Echo Recruit blog

Timesheet approval is a credit control problem

ashutosh@echorecruit.com (Echo Recruit) — Thu, 15 Oct 2026 05:40:04 GMT

Finance chases invoices. Nobody chases approvals. But the invoice was decided weeks earlier, by whether a manager clicked a button on a Monday.

In short

An unapproved hour is not late revenue. It is revenue you may never be able to evidence.
Approval lag adds to payment terms rather than running alongside them, so a three-day delay is three days of cash, every week, compounding across the book.
The single biggest cause of a stuck timesheet is that the named approver left, changed role, or went on leave and nobody has a deputy.
Run approvals with the same weekly discipline as collections — a named owner, an ageing list, and an escalation path agreed before you need it.

On a contract desk, most people treat timesheet approval as administration and invoice chasing as finance. They are the same process. Approval is simply the part of collections that happens before the invoice exists, and it is the part with the most leverage, because an hour that is approved is nearly impossible to dispute and an hour that is not is nearly impossible to defend.

The lag is additive, not parallel

The mistake is assuming approval delay is absorbed by the payment terms. It is not. The clock on your thirty days does not start until the invoice is issued, and the invoice cannot be issued until the hours are approved.

One week's hours, from work done to cash

Contractor submits after the week ends	+2 days
Sits waiting for the client's approver	+5 days
Waits for the next invoice run	+4 days
Agreed payment terms	+30 days
Work done to money in	41 days

You negotiated 30 days. You are financing 41. The eleven extra days were never discussed with the client and do not appear in any report, because each piece of the delay belongs to a different person.

Now scale it. On a book billing $60,000 a month, eleven extra days of financing is roughly $22,000 permanently tied up — money that is yours, that you have already paid a contractor for, and that you cannot use. Halving the approval lag releases about a third of it, and costs nothing but a process.

Where approvals actually get stuck

Ask why a timesheet is sitting and you will get “the client hasn't approved it yet”, which is a restatement rather than a cause. There are four real causes and they need different fixes.

The approver has left, moved, or is on leave. Overwhelmingly the most common, and the only one with a genuinely cheap fix: name a deputy at the point the assignment starts, not at the point the first timesheet sticks.
The approver never knew it was their job. A manager who agreed to take a contractor did not necessarily agree to a weekly administrative task, and often was not told. Confirm it in writing at the start, with the day of the week it will arrive.
There is a real disagreement. Overtime that was not sanctioned, a day the manager thinks was holiday. These are good — they surface early, while everyone remembers. The failure is letting them sit as silence rather than escalating them as a query.
The approval is genuinely hard to do. A PDF to print, sign, scan and email will not get done on a Friday afternoon. If approving takes more than about thirty seconds on a phone, the process is the problem.

Run it like collections

Everything that works for receivables works here, and for the same reasons.

1An ageing list, not an inboxEvery unapproved timesheet, sorted oldest first, with the client, the approver and the value on it. Value matters: a contractor on $80 an hour outranks one on $28, and an inbox cannot tell you that.
2A reminder schedule nobody has to rememberAutomatic to the approver on day two and day four, to the approver's deputy on day six, and a human call from the account owner on day eight. The cadence matters less than that it runs without anybody deciding to run it.
3An escalation path agreed at contract stageWho is called when the approver goes quiet for two weeks. Agreeing this while everyone is friendly costs one sentence in the agreement. Agreeing it during the dispute costs the relationship.
4One owner, reviewed weeklyNot “the team”. A named person who looks at the list every Monday and can say what is happening with the oldest three. Approvals go unchased for exactly the same reason invoices do — everybody assumes somebody else is on it.

The clause that does the work

Most contract agreements say hours must be approved. Fewer say what happens if they are not, which is the sentence that matters.

A deemed-approval clause — hours are treated as approved if not disputed within a stated number of working days of submission — changes the shape of the problem entirely. Silence stops being a way to defer payment indefinitely and becomes acceptance, which is how most commercial arrangements already work elsewhere.

Two things to get right. Make the window short enough to matter and long enough to be fair; five working days is usually defensible. And make sure the submission itself is evidenced and timestamped — a clause that depends on when the timesheet was sent is only as strong as your record of sending it. If the timesheet went by message from the contractor's phone, you do not have that record.

This is a commercial mechanism, not legal advice, and whether a deemed-approval clause is enforceable depends on the contract and the jurisdiction. Have the wording checked once, properly, for a clause that will appear in every agreement you sign.

Approval lag is a leading indicator

The last reason to watch this: it tells you things before anything else does.

A client whose approvals slip from two days to nine is a client with something happening — a budget under review, a manager on their way out, a project being quietly wound down. It shows up in approval behaviour weeks before it shows up in payment behaviour, and months before they tell you.

Track average approval days per client alongside days-to-pay. When one of them moves, make the call while it is still a conversation about a timesheet rather than a conversation about an overdue balance.

In Echo Recruit

Approval as a state, not an email

Timesheets carry an approval state rather than living in an inbox, so an unapproved week is visible as a record with an age and a value rather than as a missing reply. Placed contractors submit from their own portal, which gives the submission a timestamp you can rely on. Approved hours become invoice lines at the rate the signed quote set, and the receivables ledger sorts by how late things are — so approval ageing and payment ageing sit on the same data.

Common questions

Why does timesheet approval lag cost money?

Because the lag is additive, not parallel. Your thirty-day clock does not start until the invoice is issued, and the invoice cannot be issued until hours are approved. Two days to submit, five waiting for the approver, four to the next invoice run, plus thirty days terms is 41 days. You negotiated 30 and are financing 41.

Where do timesheet approvals actually get stuck?

Four causes needing four fixes. The approver left, moved or is on leave — overwhelmingly the most common, fixed by naming a deputy at assignment start. The approver never knew it was their job. There is a real disagreement sitting as silence. Or approving genuinely takes more than thirty seconds on a phone.

What is a deemed-approval clause?

A contract term treating hours as approved if not disputed within a stated number of working days of submission — five is usually defensible. It turns silence from a way to defer payment indefinitely into acceptance. It only works if submission itself is evidenced and timestamped, so a timesheet sent by message from a phone will not support it.

Two recruiters, one candidate: the duplicate problem

ashutosh@echorecruit.com (Echo Recruit) — Tue, 13 Oct 2026 05:39:32 GMT

The candidate has been called twice this week by the same agency about the same job, and both recruiters think they sourced them.

In short

Email-only matching catches the easy duplicates and misses the ones that embarrass you, because people change employers and use a different address each time.
Detect at the moment of entry, not in a monthly clean-up. A duplicate is cheap to prevent and expensive to unpick.
Merging is destructive. Decide in advance what survives — history, consent, and ownership are the three that get lost.
Some duplicates should not be merged at all. Two people genuinely share a name, and merging them is a data breach in slow motion.

Every recruitment database has duplicates, and the reason is structural rather than careless. The same person applies through a job link in 2023, is sourced by a different recruiter in 2025, gets imported in a spreadsheet from a colleague who left, and forwards their CV from a new work address after changing jobs. Four records, one human.

The visible cost is the double approach: two people from your agency contacting the same candidate about the same role, which makes the agency look chaotic and the candidate feel like a lead. The invisible costs are worse. Your pool is smaller than your count says. A candidate who opted out on one record still gets mail on another. And the placement history that would have told you this person is a strong closer sits on a record nobody opened.

Why email-only matching is not enough

Most systems deduplicate on email address. It is the obvious key: unique, cheap to compare, usually present.

It also misses the majority of real duplicates in a recruitment context, for a reason specific to this industry. Candidates change jobs — that is the entire premise of the business — and when they do, the work address on their old record stops working and the new record arrives under a different one. The single most important event in your relationship with a candidate is precisely the event that breaks your matching key.

Then there are the ordinary variations: a personal Gmail on the application and a work address on the CV, plus-addressing, a typo in a manual entry, an apostrophe in a surname handled differently by two systems.

What to match on instead

Do not look for one perfect key. Score several weak signals and use the total to decide between three outcomes: merge automatically, flag for a human, or do nothing.

A workable matching scheme

Exact email match normalised: lowercased, plus-addressing and dots stripped	Conclusive
Mobile number match normalised to international format	Conclusive
LinkedIn or portfolio URL match	Conclusive
Same name + same current employer	Strong
Same name + an overlapping past employer and date range	Strong
Same name + same city	Weak
Rule	One conclusive, or two strong

Name alone is never enough, in any combination, and name plus city is not either — see the section below on why.

The employment-history signal is the one most systems skip and the one that does the real work here, because it survives the job change that breaks the email.

Catch it at entry

The cost of a duplicate rises steeply with age. Caught at the moment of creation, it is one dialog and five seconds. Caught six months later, two recruiters have logged activity against different records, both have notes, one has a consent record and the other does not, and unpicking it is genuine work that somebody has to be paid to do.

So the check belongs at every point a record can be created — the application form, manual entry, CV import, bulk upload from a spreadsheet — and not in a monthly hygiene job. Monthly clean-ups are how you find out the problem is large; they are not how you fix it.

For the application form specifically, the behaviour matters. A candidate applying again should not be told “you already exist” — that is your internal problem, not theirs, and it reads as a rejection. Accept the application, attach it to the existing person, and let the duplicate handling happen behind the screen.

Merging destroys things. Decide which.

A merge is not a tidy-up; it is a deletion with a survivor. Three things routinely get lost, and each has a consequence:

1HistoryApplications, stage transitions, evaluations and notes on the losing record. These must move, not vanish. A candidate who reached final interview in 2024 and whose record was later merged should still show that they reached final interview — otherwise you have quietly deleted the evidence behind a past decision, which is exactly the record you would need if that decision were ever questioned.
2Consent and opt-outThe rule is one-way and worth hard-coding: an opt-out on either record survives the merge. If one record says do not contact and the other does not, the merged record says do not contact. Getting this backwards means a merge silently re-subscribes someone who asked you to stop, which is the kind of error that turns into a complaint rather than a tidy-up.
3OwnershipTwo recruiters, one commission question. Decide the rule in advance and write it into the comp plan — earliest verifiable contact, or most recent meaningful activity, are both defensible; deciding case by case is not. If the merge tool silently picks a winner, your commission policy is being set by a default nobody chose.

Keep merges reversible for a period, and keep a record that the merge happened. Somebody will merge two genuinely different people, and the only thing worse than that mistake is having no way to see it was made.

The duplicates you must not merge

Two people share a name more often than intuition suggests, and in some naming traditions considerably more often. Match on name plus city and you will eventually combine two real, unrelated people into one record.

Take that seriously, because it is not a data-quality issue. You would be showing one person's salary expectation, CV, interview feedback and recruiter notes to anyone who opens the other person's file — and if either ever asks what you hold about them, you would hand over a file containing a stranger's employment history. That is a personal data incident, not a merge error.

Which is why the rule above is one conclusive signal or two strong ones, and why name-based matches go to a human rather than to an automatic merge. The asymmetry is deliberate: a missed duplicate costs you an awkward phone call, and a wrong merge costs you considerably more.

In Echo Recruit

Duplicates surface before the second phone call

Applications deduplicate against the existing pool on arrival rather than in a monthly sweep, and potential duplicates surface on the desk before two recruiters ring the same person about the same role. Candidate evaluations are append-only and status transitions are recorded, so history moves with a record rather than being overwritten — a candidate who reached final interview still shows that they did.

Common questions

Why isn't matching on email enough to find duplicate candidates?

Because candidates change jobs — that is the premise of the business — and when they do, the work address on the old record stops working and the new record arrives under a different one. The single most important event in your relationship with a candidate is precisely the event that breaks your matching key.

What should you match candidate records on?

Score several weak signals rather than seeking one perfect key. Conclusive: normalised email, normalised mobile number, or a LinkedIn or portfolio URL. Strong: same name plus same current employer, or same name plus an overlapping past employer and date range. Weak: same name plus same city. The rule is one conclusive signal, or two strong ones.

What must a candidate record merge never destroy?

Three things. History — applications, stage transitions and evaluations must move, not vanish. Consent — an opt-out on either record survives the merge, always. And ownership — decide the commission rule in advance, because if the merge tool silently picks a winner, your comp policy is being set by a default nobody chose.

Six numbers that tell you the desk is healthy

ashutosh@echorecruit.com (Echo Recruit) — Thu, 08 Oct 2026 05:39:38 GMT

A number is worth reporting if a bad value would make you do something differently on Monday. By that test, most recruitment dashboards report almost nothing.

In short

Counts of things that exist — open roles, candidates, activities logged — describe the size of the desk, not its health.
Every metric below is paired with the decision it should trigger. If there is no decision, do not put it on the wall.
Sort by the worst case, not the average. An average time-to-fill of 31 days is compatible with a role that has been open since March.
Five of the six can be computed from data you already have; the sixth is the one most agencies are missing, and it is the one about cash.

Walk into most agencies and the screen on the wall shows activity: calls made, CVs sent, roles open, revenue against target. It is a picture of effort and a picture of the month so far, and it has one significant weakness. Every number on it can look fine on the morning of the day two placements fall through and a client goes sixty days past due.

The test for whether a number belongs on that screen is simple: if this value were bad, what would somebody do differently before lunch? If there is no answer, it is reporting, not management. Here are six that pass the test.

1. Slate coverage, per requisition

Not “candidates in the pipeline”. The number of candidates on this role who are at or past a stage the client actually sees — submitted, interviewing, offered — measured against how many that client needs to see before they hire.

A role with twenty candidates at “contacted” and none submitted is not 20% covered. It is 0% covered with a lot of activity attached, and it will still be 0% covered on Friday unless somebody notices today.

The decision it triggers: which roles get worked this afternoon. Sort ascending by coverage, not by how recently someone touched them.

The wrong version: total candidates linked to the job. That number goes up when somebody adds people and never goes down, so it rewards the wrong activity and never looks bad.

2. Time in current stage, oldest first

Time-to-fill is a post-mortem statistic. It tells you about roles that are already finished, and it arrives too late to change any of them. What you want is the live version: for every candidate in play, how many days since they last moved.

The specific thing this catches is the candidate sitting at “submitted to client” for eleven days. Nobody has dropped the ball exactly — the recruiter chased once, the hiring manager was travelling — but the candidate is now interviewing elsewhere and does not know they are still being considered here.

The decision it triggers: a specific chase, to a specific person, today.

The wrong version: average time in stage. Averages are built to hide exactly the case you are looking for.

A useful sharpening: set an expected duration per stage — two days at CV review, five at client submission, three between interview and feedback — and report the ones that have blown it, not the raw age. A candidate at day four of client submission is fine. A candidate at day four of “awaiting feedback” may not be.

3. Requisitions with no activity in seven days

A count, and the list behind it. Open roles where nothing has moved for a week: no new candidate, no stage change, no message to the client.

Every agency has these, and the reason is almost never that the role was abandoned on purpose. It is that the role belonged to somebody who left, or it was covered by two people who each thought it was the other’s, or the client went quiet and nobody decided whether it was dead.

The decision it triggers: work it, or close it. Both are fine. Leaving it open and untouched is the only bad option, because it inflates your open-role count, ruins every ratio calculated against it, and lets a client believe you are working on something you are not.

4. Pass rate against the bar, per requisition

Of the applications received for this role, what share cleared the shortlisting bar you wrote down? This one is diagnostic in both directions, which is what makes it unusual.

How to read a pass rate

Above ~60% clearing the bar You are not screening, you are counting. The bar is decoration and the recruiter is still doing the whole job by hand.	Bar too low
Roughly 10–40% The bar is doing real work and there is still a slate to choose from.	Working
Under ~5% Either the brief describes a person who does not exist at that salary, or the advert is reaching the wrong people. Neither is fixed by reading more CVs.	Wrong upstream

The bands are judgement, not law — a niche role with twelve applicants and one hire is a healthy 8%. The signal is a rate that is extreme and stays extreme across several roles.

The decision it triggers: a conversation with the hiring manager about the brief, or with whoever is placing the advert — before another three weeks of applications arrive against the same wrong bar.

5. Offer-to-start survival

The share of offers accepted that turn into someone actually starting, and the reasons for the ones that do not. Counter-offer, competing offer, background check, or simply changed their mind.

This is the number that converts directly into revenue you have already counted and will not receive. It is also the earliest warning you get about a hiring process that is too slow: candidates who drop between offer and start are usually candidates who were in another process that moved faster.

The decision it triggers: if counter-offers dominate, the conversation about money has to happen earlier in the process rather than at offer. If competing offers dominate, your client’s process is too slow and that is a conversation with the client, evidenced.

Worth pairing it with: guarantee-period survival — placements still in post at the end of the rebate window. Both of these are money you have recognised and might have to give back.

6. Days sales outstanding, by client, aged

The one most agencies do not have on the same screen as the rest, because it lives in the accounting system and the accounting system belongs to somebody else.

It belongs next to the pipeline for one reason: the client who is delightful to work with, gives you exclusivity and briefs you properly may also be the client who pays at 74 days. Those two facts are known by two different people in your business, and the person deciding where to spend Thursday is the one who does not know the second.

What a single slow client costs to carry

Invoiced to this client in a year	$180,000
Agreed payment terms	30 days
Actual average days to pay	74 days
Extra days of financing you are providing	44
Average balance outstanding because of the delay $180,000 ÷ 365 × 44	≈$21,700
Cost at 10% cost of capital	≈$2,170 a year

Plus the part that is not in the arithmetic: that $21,700 is the money you were going to use to hire a resourcer.

The decision it triggers: where the next exclusive goes, whether this client’s next agreement gets different payment terms, and whether the reminders are going out on a schedule or when somebody remembers.

What to leave off

Three popular numbers that fail the Monday test, and why:

Calls made and emails sent. Effort, not outcome. They rise when people are being measured on them, which is the definition of a metric that stops meaning anything the moment you put it on a wall.
Total candidates in the database. A number that only goes up, attached to a liability that also only goes up. If it is doing anything, it is encouraging you to keep data you should have deleted.
Revenue to date, on its own. It is the score, not the game. By the time it is wrong, everything that made it wrong happened weeks ago and is visible in the five numbers above.

The reporting rule underneath all six

Every one of these reports a change or an exception rather than a total. That is not a stylistic preference. A total is the sum of everything that has ever happened, which means it moves slowly, hides the individual case, and cannot tell you whether this week was good. An exception list is short, is different every morning, and each line has a name attached to it.

If you only take one thing: replace the largest number on your wall with the shortest list.

In Echo Recruit

Reporting what changed

The dashboard is built on this rule. Slate boards carry the real column totals rather than the cards on screen, so a column showing five of forty says so. Tasks group overdue first. Requisition health, pipeline conversion and sourcing performance sit in analytics; receivables ageing and days-to-pay sit in billing on the same data, so the slow-paying client is visible to the person deciding where the week goes.

Common questions

Which metrics should a recruitment agency report on?

Six that trigger a decision: slate coverage per requisition; time in current stage, oldest first; requisitions with no activity in seven days; pass rate against the written bar; offer-to-start survival; and days sales outstanding by client, aged. The test for each is whether a bad value would make somebody do something differently before lunch.

Why is average time-to-fill a poor metric?

Because it is a post-mortem statistic about roles that are already finished, and averages are built to hide the case you are looking for. An average of 31 days is compatible with a role open since March. The live version — days since each candidate last moved, sorted oldest first — triggers a specific chase to a specific person today.

How do you read a shortlisting pass rate?

Above roughly 60% clearing the bar, the bar is decoration and the recruiter is still doing the whole job by hand. Roughly 10–40% means it is doing real work with a slate still to choose from. Under about 5%, either the brief describes someone who does not exist at that salary or the advert is reaching the wrong people.

The placement is not the finish line

ashutosh@echorecruit.com (Echo Recruit) — Tue, 06 Oct 2026 05:39:48 GMT

Your ATS marks them placed and stops caring. Your obligations to that person are about to run for another nine months.

In short

“Placed” is a status change in the ATS and the start of a long relationship everywhere else.
Document expiry, extension conversations and the four-week check-in are all dated events you already know about — they only become emergencies when nobody is holding the date.
A leave balance is a ledger, not a number somebody edits.
A contractor who finished well is the cheapest placement you will ever make, and only if you know they finished and when they are free.

Watch what happens in most agencies on a contractor’s first day. The requisition closes. The candidate’s record moves to “placed” and stops appearing in anybody’s view. The recruiter gets a bell, a number on a board, and their next role.

And then a folder appears. It has the signed contract in it, a scan of a passport, a spreadsheet of hours, and whatever else accumulates. It is owned by whoever is least able to refuse, and it is where the next nine months of this relationship are going to be managed.

Every expensive thing that happens on a contract desk happens in that folder.

What is still owed after the start date

Not in a compliance-manual sense. In a “this will cost you money or a client” sense.

1Documents with an expiry dateRight-to-work evidence, visas, professional certifications, background checks, insurance where the contractor carries their own. Each has a date on it, and the date was known on day one. A visa that lapses on the 14th is not a surprise — it is a calendar entry nobody made. When it does lapse, the contractor cannot work, the client loses cover, and you are explaining both.
2An onboarding checklist somebody ownsEquipment, systems access, site induction, the client’s own paperwork. The failure mode here is not that items get missed; it is that nobody can say which ones are outstanding, so the answer to “are they set up?” is always “I think so”.
3Weekly timesheets, with approvalThe engine of the whole desk. Hours in, approved by a named person at the client, billed. Late or unapproved hours are the largest and quietest leak on a contract book.
4Leave, as a ledgerWhere leave applies to the engagement, the balance is not a number in a cell. It is the sum of accruals and bookings, and it needs to be reconstructable — because the one time it will be questioned is at the end, when somebody is owed a payment for untaken days and disagrees with your figure.
5The four-week check-inA conversation with the client and a conversation with the contractor, separately, about a month in. This is the earliest warning you get about a placement that is not working, and it lands while the guarantee or rebate period is still live and while the situation is still fixable. It is the highest-value half hour on the whole desk and it is the first thing to be skipped.
6The extension conversation, before week elevenOn a twelve-week assignment the extension is decided around week nine or ten. Start it at week eleven and you are negotiating against a contractor who has already taken a call from somebody else. The date to act on is the contract end date minus three weeks, and it exists the day the contract is signed.
7The exit, and the record it leavesFinal timesheet, equipment returned, access revoked, client sign-off — and then the thing almost nobody captures: how it went, and when this person is free again.

Why the folder always loses

Every item on that list is a date or a state. A folder holds neither. It holds documents, which means the only way to find out whether anything is due is for a person to open it and look — which happens when there is time, which is never, or after something has gone wrong, which is too late.

The second problem is ownership. The recruiter moved on to their next role, which is correct: it is what they are for. Nobody replaced them as the owner of this relationship, so it is owned by whoever notices, which in practice means it is owned by the client when they complain.

The third is the one nobody budgets for. When the same person is in an ATS as a candidate, in a spreadsheet as a timesheet, in accounting as a payee and in a folder as a set of documents, the four of them disagree. Their end date is different in two places. Their rate was updated in one. And the question “who is on site at Northwind this week?” takes somebody forty minutes to answer with a caveat.

The re-book argument

Everything above is defensive. Here is the offensive version, which is usually the one that gets a change funded.

A contractor who has already worked for you, whose documents you have already checked, whose quality you have already seen, and who finished an assignment well is the cheapest placement available to you. No sourcing. No screening. A conversation and a start date.

Capturing that requires exactly two facts that the folder does not hold: how the assignment ended, and when they become available. Record those two things at exit and your best source of next quarter’s placements is a list you can sort. Fail to record them and every assignment ends with somebody walking off your book and onto somebody else’s.

The same applies to the client side. A contractor finishing in three weeks is a conversation with that client about what comes next — but only if the end date is somewhere that generates a prompt rather than somewhere that generates a search.

The test

Ask your desk four questions and time the answers.

Which documents expire in the next sixty days, and for whom?
Which assignments end in the next thirty days, and has the extension been raised?
Whose timesheet for last week has not been approved?
Who finished well in the last six months and is available now?

If all four take under a minute, the post-placement half of your desk is under control. If any of them requires opening a folder, you have found the thing to fix — and it is almost certainly costing you more than the sourcing problem you were planning to spend money on.

In Echo Recruit

The people you placed stay in the system that placed them

The placement stays inside the same workspace as the candidate who became it. Employee records carry effective-dated assignments and pay. Onboarding checklists track document expiry, so a lapsing right-to-work document is a task rather than an incident. Leave is an append-only accrual ledger, timesheets carry approval, performance runs in cycles with calibration, and exits record clearance and an alumni record.

The people you placed get their own portal — their hours, their balances, their documents, their review — so the timesheet arrives as a submission rather than as a message on a Sunday night.

Common questions

What does an agency still owe after a contractor starts?

Seven things, all dated or stateful: documents with expiry dates, an owned onboarding checklist, weekly approved timesheets, leave as a ledger rather than a number, the four-week check-in, the extension conversation at end date minus three weeks, and an exit that records how it went and when the person is free again.

Why does managing a placement in a folder fail?

Because every item on that list is a date or a state, and a folder holds neither. It holds documents, so the only way to find out whether anything is due is for a person to open it and look — which happens when there is time, which is never, or after something has gone wrong, which is too late.

When should you start the extension conversation?

At the contract end date minus three weeks, a date that exists the day the contract is signed. On a twelve-week assignment the extension is decided around week nine or ten. Start it at week eleven and you are negotiating against a contractor who has already taken a call from somebody else.

A subject access request arrives on a Friday

ashutosh@echorecruit.com (Echo Recruit) — Thu, 01 Oct 2026 05:39:58 GMT

“Please send me all the personal data you hold about me.” One sentence, no explanation required, and a clock starts.

In short

A request does not have to use the words “subject access request”, be in writing, or go to a particular address. It counts from when anyone at your agency receives it.
The database is the easy part. The time goes on mailboxes, exports, notes and whatever a recruiter saved locally.
Recruiter notes are disclosable. That fact should change what goes in them, starting before you ever receive a request.
You may need to redact other people — a note naming a client contact contains their data too.

Most agencies handle their first access request badly, and not out of bad faith. They handle it badly because nobody has ever done one, the deadline is shorter than it sounds, and the data turns out to be in nine places rather than one.

This is a practical runbook. It is not legal advice, the specifics differ by jurisdiction, and if you place people across borders the details vary — but the shape of the work is much the same everywhere, and having done it once on paper makes the real one manageable.

Recognise it

The commonest failure is not recognising the request at all. Under the UK and EU GDPR a request can be made verbally or in writing, to anyone in the organisation, and need not mention any particular phrase. “What have you got on me?” said on a phone call to a recruiter is capable of starting the clock.

The deadline is generally one month from receipt, extendable in limited circumstances for complex requests. The practical consequence: the clock may already have been running for a week before it reaches anyone who knows what to do.

Two cheap protections. Tell everyone who speaks to candidates what one looks like and where to forward it — a single line in induction is enough. And publish an address for it in your privacy notice, so most arrive somewhere sensible.

Where the data actually is

The instinct is to export the candidate record and send it. That is the first ten percent. Personal data is anything from which the person can be identified, and in a recruitment agency it is scattered:

The candidate record — contact details, parsed CV, applications, stage history, scores
Documents in storage — the CV itself, right-to-work scans, certificates
Recruiter notes and call logs — free text, and the part people forget
Email and messages — including internal threads about them, not just correspondence with them
Interview scorecards and feedback, including the client's if you hold it
Payroll and timesheet records, if they were ever placed
Exports — the spreadsheet someone downloaded to work on at the weekend, which is outside every control you have

Internal email is where the days go. A search across mailboxes for a name and an address is the single most time-consuming step, and the one most likely to surface something you would rather had not been written.

The runbook

1Log it and acknowledge the same dayRecord the date received — that is the date the clock started, not the date it reached the right person. Acknowledge, and say when they will hear back.
2Verify who you are talking toProportionately. If the request comes from the address already on the record, that is usually enough; if it comes from an unfamiliar address claiming to be a candidate, ask for something reasonable. Do not use verification as a delaying tactic, and do not demand a passport scan for a request about a CV — you would be collecting more sensitive data than you hold.
3Search every system on your listUse the list you built when you mapped where data lives. Search on name, every email address you hold, and phone number — people apply from more than one address, which is also why duplicate records matter here.
4Review for third partiesA note saying “spoke to Priya at Northwind, she thought he was light on controls experience” contains Priya's personal data as well. You generally cannot disclose another identifiable person's data without consent or another basis, so redact or summarise. This step takes judgement and is the reason a request cannot simply be an automated export.
5Send it with the explanationNot a raw database dump. A readable pack, plus the things the law expects you to tell them: why you hold it, who you have shared it with, how long you keep it, where it came from if not from them, and their other rights. Send it securely.
6Record what you sentKeep a copy of the pack and the date. If they follow up — and an access request is quite often the step before an erasure request or a complaint — you need to know exactly what was disclosed.

Notes are disclosable. Write them accordingly.

The single most useful thing to take from this: recruiter notes are personal data, and the person can ask to see them.

That is not an argument for keeping no notes. Notes are how a desk works, and a candidate record with no history is useless. It is an argument for writing notes you would be comfortable showing the person — which, conveniently, are also better notes.

Observations, not verdicts on character. “Asked for £75k, would consider £70k for remote” is useful and fine. “Difficult, bit of an ego” is neither.
Reasons tied to the requirement. “No production experience with the stack” beats “not a fit” — and it is also the note that defends the decision if the rejection is ever questioned.
Nothing about protected characteristics, or anything adjacent to them. Notes about family plans, health or age are a problem in a disclosure pack and a much larger problem in a discrimination claim.

Run one as a drill. Pick a candidate placed two years ago, and have someone assemble everything you hold on them as if the request were real. Time it. The drill costs an afternoon and tells you which system is the bottleneck while the answer still has no deadline attached.

What usually goes wrong

The clock started earlier than you think. It runs from receipt by the organisation, not by the right person.

Somebody deletes something. Tidying a record after a request arrives is the worst possible instinct — it converts an administrative task into something far more serious. Freeze the record instead.

The duplicate record is missed. You search the address they wrote from and find one record; a second exists under an old work address. Deduplication is a data protection control as much as a hygiene one.

Nobody can say what is held or for how long. Which is the same underlying gap as not having decided a retention period — the request just makes it visible.

In Echo Recruit

One workspace to search, not nine

Candidate records, applications, evaluations, documents and message history sit in one workspace scoped by row-level security, so the search step is a lookup rather than an expedition. CVs live in private storage served by signed links that expire in five minutes, and retention is enforced by a scheduled sweep that deletes the file before clearing the pointer — so there is no orphaned document sitting outside any record. Applications deduplicate against the pool on arrival, which is what stops a second record being missed.

Common questions

How long does an agency have to answer a subject access request?

Generally one month from receipt, extendable in limited circumstances for complex requests. The clock runs from receipt by the organisation, not by the person who knows what to do — so it may already have been running a week before it reaches them. This is a practical runbook, not legal advice, and specifics differ by jurisdiction.

Does a subject access request have to be in writing?

No. Under the UK and EU GDPR it can be made verbally or in writing, to anyone in the organisation, and need not use any particular phrase. “What have you got on me?” said on a call to a recruiter is capable of starting the clock. Tell everyone who speaks to candidates what one looks like and where to forward it.

Are recruiter notes disclosable?

Yes — recruiter notes are personal data and the person can ask to see them. That is not an argument for keeping no notes; it is an argument for writing notes you would be comfortable showing the person. Observations rather than verdicts on character, reasons tied to the requirement, and nothing about protected characteristics.

The intake call that decides the whole search

ashutosh@echorecruit.com (Echo Recruit) — Tue, 29 Sep 2026 16:13:12 GMT

A search that goes wrong usually went wrong in the first hour, and the first hour is the one nobody protects.

In short

A job description is a list of everything anyone could want. A brief is a ranked set of trade-offs. Only one of them is useful.
Comparative questions get real answers where direct ones do not. Nobody gives up a must-have when asked; they give it up to keep something better.
Test the salary against the requirement in the room, not three weeks later with a rejected shortlist as evidence.
Leave with the decision process written down — who interviews, in what order, how fast. It predicts whether you will lose the candidate at offer.

Ask a recruiter why a search stalled and you will hear about the market. Sometimes that is true. More often the search was mis-specified on day one, everyone worked hard against the wrong target for three weeks, and the market took the blame.

The intake call is the cheapest point of intervention in the entire process. An hour there saves a fortnight later, and it is the only hour where the hiring manager is obliged to think rather than react.

The job description is not the brief

The document you are sent lists every attribute anybody in the hiring team mentioned, with no weighting and no trade-offs. It is a committee artefact. It will tell you the candidate needs seven years, three technologies, a degree, and to be a self-starter.

What you need is different: if I can only bring you one person, what makes them the one? That question cannot be answered by a list, because the answer is a ranking, and rankings only come out under forced choice.

The five questions that produce a brief

1“Who did this before, and what happened to them?”A vacancy exists for a reason. Promoted, left, was let go, or the role is new — each implies a different search and a different pitch. If the last person was pushed out, ask what they were missing; that is the real requirement, and it is rarely in the job description. If the role is new, ask who is doing the work today, because that person is your benchmark and possibly your candidate.
2“What will they have delivered ninety days in?”This converts a list of attributes into work. A manager who cannot answer it has not thought the role through, and that is worth knowing before you source. The answer also gives you the single best line for approaching passive candidates — people respond to a problem to solve far more than to a list of requirements they already meet.
3The forced choice, asked four or five times“Two candidates, identical except one has eight years in the industry and the other has four but has shipped exactly the thing you are building. Who do you see first?” Then vary the pair — depth against breadth, domain against craft, someone cheaper and rawer against someone at the top of band. Each answer is a weight. Four or five of them is a rubric.
4“Which of these would genuinely end it?”Read the must-haves back and ask which ones are actually disqualifying. Right to work, a licence the job legally requires, a hard location constraint — those are real. “Seven years” almost never is, and treating it as one throws away the person with six who would have got the job. A manager who insists everything is a knockout is telling you the brief is not agreed internally, which is a different problem and worth surfacing now.
5“Who else says yes, and how fast can they?”Names, order, and availability. Not the process in theory — the actual diaries. A four-stage process with a panel member who travels every other week is a six-week process, and you should say so in the room rather than discover it when a candidate accepts elsewhere.

Test the salary before you leave

The most common failure is not an unreasonable requirement or an unreasonable budget. It is the pair of them, held by people who have never put them next to each other.

Do it in the room. Take the requirement as agreed, name two or three real people you know at that level — no names, just shape and current package — and say what they are paid. If the band does not reach, you have three options and the manager picks one:

Move the money. Sometimes available, especially if you have made the gap concrete rather than theoretical.
Move the requirement. Drop the domain, take the person with four years and the exact stack, plan to train.
Move something else. Remote days, title, scope, a faster process. These are worth real money to candidates and cost the client less than salary.

Having that conversation on day one is an awkward twenty minutes. Having it on day twenty-one, with a rejected shortlist as evidence, costs you the search and some of the relationship.

A tell worth listening for: “we'll know them when we see them.” It sounds like discernment and usually means the criteria are not agreed between the people who have to sign off. Push once — “what did the last person you said that about have?” — and if the answer is still vague, expect a long search with shifting feedback. Price it accordingly, or decline it.

Leave with four things written down

Not notes. A short document you send back the same day, because the act of confirming it in writing is what makes it binding enough to refer to in week three.

The ranked requirement — what is disqualifying, what is weighted and how heavily, what cannot be read off a CV and will be asked at screening instead.
The ninety-day outcome, in the manager's own words. This is your approach message.
The package, tested — range, plus what else is movable.
The decision process — names, order, target turnaround at each stage.

That document does two jobs. It is the specification you screen against, so two recruiters on the same role apply the same bar. And it is the thing you hold up in week three when the feedback starts drifting — not to win an argument, but to make the drift explicit and get the brief formally changed rather than informally ignored.

The brief will change. That is fine.

Around the third week, the market answers. Candidates who match the brief turn out not to exist at that salary, or they exist and are not interested, or the manager meets someone who reframes what they want.

This is normal and not a failure of intake. The failure is when it happens silently — when the bar moves and nobody records that it moved, so the candidates rejected in week one were judged against rules that no longer apply and nobody can reconstruct why.

So treat a brief like a version, not a document. When it changes, say so, write the new one, date it, and note what changed. The cost is five minutes. The benefit is that six months later, when somebody asks why a particular person did not get a call, there is an answer that is not a recollection.

In Echo Recruit

The brief as a structure, per requisition

Every job carries its own sourcing brief and its own shortlisting rubric — target titles, must-haves, locations, and rules marked knockout, scored or manual. Rules a CV cannot prove become screening questions instead of silent penalties. Rubrics are versioned and each evaluation records the version that produced it, so a bar that moves in week three does not rewrite what was decided in week one. There is also a criteria builder that drafts the first version from what you tell it about the role, with a rationale per rule, which you edit before it applies.

Common questions

What questions should you ask on a recruitment intake call?

Five. Who did this job before and what happened to them. What will the hire have delivered ninety days in. A forced choice between two candidate shapes, asked four or five times with the pair varied. Which stated must-haves would genuinely disqualify. And who else says yes, in what order, with what diary availability.

Why isn't the job description enough to run a search from?

A job description lists every attribute anyone in the hiring team mentioned, with no weighting and no trade-offs — it is a committee artefact. A brief is a ranked set of trade-offs. Rankings only come out under forced choice, which is why comparative questions produce real answers where direct ones do not.

How do you test the salary against the requirement?

In the room, on day one. Take the requirement as agreed, describe two or three real people you know at that level and what they are paid, and see whether the band reaches. If it does not, the manager picks one of three: move the money, move the requirement, or move something else — remote days, title, scope, a faster process.

Eleven questions to ask before you let software rank candidates

ashutosh@echorecruit.com (Echo Recruit) — Tue, 29 Sep 2026 05:39:41 GMT

You are not evaluating whether the model is good. You are evaluating whether, eighteen months from now, you can reconstruct why a particular person did not get a call.

In short

Ask what is deterministic and what is a model. If everything is a model, nothing is explainable.
The single most revealing question is what happens to a requirement the CV cannot prove. A silent zero is a broken process.
An evaluation should be a new row, not an edit. Overwriting means last week’s decision no longer exists.
Automated employment decision tools are regulated in a growing number of places, and the obligations mostly fall on you, not the vendor.

Screening software has got good enough to be worth buying and confident enough to be worth checking. The demo will be impressive, because ranking a pile of CVs is genuinely something this technology does well. The demo is not the thing to evaluate.

The thing to evaluate is what happens afterwards, when a rejected candidate’s solicitor, a client’s procurement team, or a regulator asks a question about a specific decision made on a specific day. These eleven questions are the ones whose answers you can verify rather than take on trust, roughly in the order they become expensive.

On how it works

1. Which part of this is arithmetic and which part is a model?

There should be a clear line. Arithmetic — does the candidate have these skills, this many years, a title that matches — is deterministic, repeatable and explainable to a non-technical person. A language model reading a CV and forming a view is none of those things, and it is also better at the parts arithmetic cannot do.

A good architecture runs the deterministic pass first and uses the model for judgement on top, with both visible. A vendor who cannot draw that line for you either does not have one or does not want you looking at it.

2. Can you show me the weights?

Not the model weights — the rubric weights. Skills 50%, experience 25%, role match 25%, or whatever they are. You should be able to read them, change them, and see the score move.

“Our matching algorithm is proprietary” is a complete answer to this question, just not the one being asked for. If you cannot see the weights, you cannot tell a hiring manager why one candidate ranked above another, which means you have outsourced the part of your service you are actually paid for.

3. What happens to a requirement the CV cannot prove?

The most revealing question on the list, and the one most likely to get a vague answer.

Salary expectation, notice period, visa status, willingness to be on site: none of these are on a typical CV. If the system scores them as zero when absent, it is systematically marking down everyone who writes a normal CV, and the people who float to the top are the ones whose documents happen to mention everything. That is not a screen for capability; it is a screen for CV formatting.

The right answer is that unprovable requirements are excluded from the calculation entirely and surfaced as questions for the screening call.

4. Does the system ever reject anybody?

It should not. Rank, flag, sort, band — yes. Automatically move a human being out of a process with no person involved — no. Partly because it is the kind of automated decision-making that attracts specific legal obligations in several jurisdictions, and partly because the candidate worth placing is disproportionately the one your rules did not anticipate.

On what it remembers

5. Is an evaluation a new row, or an update?

When a candidate is re-scored — because the rubric changed, or because their profile was updated — does the system write a new evaluation, or overwrite the old one?

If it overwrites, then the score that was in front of the recruiter on 4 March no longer exists anywhere, and the reason a candidate was rejected has been silently replaced by a reason nobody acted on. Append-only evaluation history is a small engineering decision that is nearly impossible to retrofit once there is a year of data, which is why it is worth asking before you buy rather than after.

6. Are the criteria versioned, and does each evaluation point at a version?

Related but not the same. The bar moves during a search. If your rubric is a single mutable object, then changing it in week three changes the apparent basis of every decision made in weeks one and two — retrospectively, and with no record that anything happened.

7. What does it do when the model is unavailable?

API outages happen, keys expire, and rate limits arrive at the worst moment. There are three possible behaviours: fall back to the deterministic score and say so on screen, refuse and say so, or quietly return something worse without mentioning it. Only the third is disqualifying, and it is more common than you would like.

8. Does a status change survive a later status change?

If a candidate reaches final interview and is then rejected, does the record still show that they reached final interview? A system that stores only a current status has thrown away the history you would need to answer almost any question about fairness.

On who can see what

9. Is access enforced by the database or by the page?

Scores, salary expectations and notes are sensitive. Ask specifically whether the restriction is in the data layer or in the user interface — that is, whether someone calling the API directly gets the same refusal as someone clicking the button. Interface-only permissions are a lock on the door of a room with no walls.

10. What happens to the CVs of people you did not hire?

A retention period, an enforcement mechanism, and a way to answer a deletion request without a week of searching. If the answer is “they stay in the database”, you are buying a growing liability with a search box on it.

11. Is our data used to train models?

Ask about the vendor’s own training and about their subprocessors’. Get the answer in the contract rather than on a call, because the answer on the call is always no and the answer in the terms is occasionally different.

The regulatory backdrop

Automated tools used to screen candidates are increasingly regulated, and the obligations generally land on the employer or the agency using the tool rather than on the company that built it. A vendor saying “we are compliant” is answering a question about themselves.

New York City Local Law 144. Automated employment decision tools used to screen candidates for jobs in NYC require an annual independent bias audit, publication of a summary of the results, and notice to candidates before the tool is used. Practically, this means asking a vendor what data they will hand over so an auditor can do the work.
The EU AI Act. AI used in employment and worker management is classified as high-risk under Annex III, which brings obligations around risk management, data governance, logging, human oversight and transparency. The compliance timetable has been adjusted more than once since the Act entered into force, so check the dates that apply to you rather than relying on a summary — including this one.
Illinois. The Artificial Intelligence Video Interview Act sets notice, explanation and consent requirements for AI analysis of video interviews, along with deletion obligations on request. Separate amendments to the Illinois Human Rights Act address AI use in employment decisions more broadly.
Data protection law generally. Under the GDPR and UK GDPR, individuals have rights concerning decisions based solely on automated processing that produce legal or similarly significant effects, plus the usual rights of access and erasure over everything you hold about them.

This is a description of the landscape, not legal advice, and it is a snapshot written by a software company rather than a lawyer. Which jurisdictions apply depends on where the role is, where the candidate is and where you are — often all three. If you place people in more than one country, this is worth an hour of somebody qualified.

What the answers add up to

Notice that nine of the eleven questions are about records rather than about intelligence. That is the point. The quality of the ranking is what you evaluate in a trial, with your own roles, in a fortnight. The quality of the record-keeping is what you cannot evaluate later, because by the time you need it, the year of data has already been written in whatever shape the vendor chose.

If you ask only one, ask the third: what happens to a requirement the CV cannot prove. The answer tells you whether anyone at the vendor has thought about the candidate’s side of this at all.

In Echo Recruit

Our answers to our own eleven

Scoring is deterministic first: skills weighted 50%, experience 25%, role match 25%, against the requisition’s own brief, with the weights on screen. Claude is used for three things — reading a CV, drafting a rubric, and ranking an already-scored slate with its reasoning shown — and is told the numeric score is evidence rather than verdict. Turn the key off and the deterministic path still runs and labels itself.

Rules a profile cannot prove are excluded from the arithmetic and become screening questions. Evaluations are append-only and rubrics are versioned, so each decision points at the rules that produced it. Status transitions are recorded, so a rejection never erases that somebody reached final interview. Nothing is ever rejected automatically. Commercial fields are restricted by row-level security in the database rather than by the page. Customer data is not used to train models.

Five places a handover breaks

ashutosh@echorecruit.com (Echo Recruit) — Mon, 28 Sep 2026 19:34:43 GMT

Recruitment work is a relay. The baton drop happens at the exchange, not mid-stride.

In short

All five failures below have the same shape: the outcome travels to the next person and the reasoning does not.
None of them is dramatic on its own, which is exactly why none of them ever gets fixed.
The cost is mostly re-work — time spent re-establishing something that was already known — and it never appears on any report.
The fix is the same each time: attach the why to the record rather than to the person.

Nothing breaks in the middle of a tool. It breaks at the exact moment one person hands work to another — the CV moves, the desk changes hands, the rate gets confirmed, the invoice gets cut, the decision gets questioned later.

Here are the five exchanges where it happens, what each one actually costs, and the smallest thing that fixes it.

The CV moves. The context doesn't.

A recruiter spends twenty minutes on a call learning why a candidate is looking, what they will and won't accept, what's already been discussed. When the CV gets forwarded — to a client, or to another desk — what travels is the PDF: name, history, maybe a two-line note.

What doesn't travel is the reasoning behind it. The client asks something already answered on the call. The candidate gets asked to repeat themselves to whoever inherits them. It's small friction, but it's usually the first place trust erodes — and candidates read being asked twice as nobody having read it once.

The fix: a short structured summary attached to the submission rather than to the recruiter — what they want, what they've refused, what's already been covered. Three lines, written once, travelling with the person.

What re-covering old ground costs a single desk

CVs sent to clients each week	15
Share where someone asks a question already answered	1 in 3
Time to re-establish it email, a call back to the candidate, waiting for the reply	15 min
Hours a year 5 × 0.25 × 48	60
At a loaded cost of $35/hr	$2,100 a year, per desk

Substitute your own three numbers. The figure is not the point — the point is that this is a salary line nobody has ever costed, on the cheapest of the five gaps.

Cover a desk, inherit nothing.

Someone's on leave, or covering another consultant's desk. The pipeline is there — statuses, candidate lists — but the judgment calls that produced those statuses usually aren't.

Why was this candidate parked instead of progressed? Was that client call about a rate concern or a timeline concern? Nothing was hidden on purpose. It just lived in one person's head, a personal notebook, or a direct message nobody else was on.

This is the gap that turns a fortnight's holiday into a fortnight of stalled pipeline, and it is the one that becomes serious when somebody leaves rather than goes away for a week.

The fix: make the reason for a stage change a required part of the stage change, not a note somebody might add. A status without a reason is half a record.

Said on a call. Written down differently.

A rate gets agreed verbally — bill rate, pay rate, standard terms. Somewhere between the call and the paperwork, a number shifts, a term gets dropped, or “standard” gets interpreted differently by whoever types up the contract.

Nobody's lying. Two people just remembered a fifteen-minute conversation slightly differently, and only one version made it onto paper. The other version is still in the client's head, and it will surface at the worst possible moment — usually when the first invoice lands.

The fix: confirm commercial terms in writing the same day, in a form the other side has to respond to. A follow-up email saying “to confirm: 22% of base salary, 90-day sliding rebate, 30 days from invoice” costs two minutes and converts a memory into a record. What a fee agreement has to nail down to survive this trip is a longer subject.

Signed terms. An invoice built from memory.

The contract exists, signed and filed. Weeks later, when it's time to bill, someone opens billing software and rebuilds the invoice from what they think the contract says — sometimes checking the PDF, sometimes not, especially against a deadline.

The two documents can quietly diverge, and the first sign is usually a client querying a line item. By then the person who agreed the terms may have moved on, and the person defending the invoice is reconstructing a conversation they were not part of.

This is the most expensive of the five because it has a currency symbol attached to every occurrence. We took it apart separately in the four times a staffing agency types the same fee.

The fix: the invoice should be assembled from the agreed terms rather than typed against them. Where that is not possible, the person raising it should be reading the executed agreement, not the CRM field somebody transcribed from it.

Rejected. Nobody remembers why.

A candidate gets screened out — a reasonable call, made fast, in the middle of a busy day. Months later, that candidate reapplies, or a client asks why they weren't put forward, or an audit asks for the screening record.

The decision was fine. The problem is it was never written down, so “fine” is now unprovable. And unprovable is the operative word: this is the one gap of the five that can become a legal question rather than a commercial one, because automated and semi-automated screening is now regulated in a growing number of places and the obligation to explain a decision sits with the agency.

The fix: write the bar down before the CVs arrive, so a rejection reason is a reference to a rule rather than a recollection. Writing the bar down costs an hour at intake and is the only version of your judgement that still exists six months later.

The same failure, five times

Read the five back and the pattern is hard to miss. In every case the outcome transfers cleanly — the CV arrives, the status is visible, the contract is signed, the invoice goes out, the candidate is marked rejected. In every case the reasoning stays with the person who had it.

That asymmetry is not an accident of any particular tool. Outcomes are what systems are built to store, because outcomes are structured — a stage, a date, an amount. Reasoning is unstructured, takes effort to write, and benefits somebody other than the person writing it. So it does not get recorded, and it walks out of the building every time a desk changes hands.

Which gives you the one-line test for any process you are about to change: if the person who did this left tomorrow, what would the next person be missing? If the answer is “the why”, you have found a handover worth fixing — and you can fix it without buying anything, by making the reason a required field rather than a courtesy.

Worth ranking them before you act. Gaps three and four have money attached and are worth fixing first. Gap five is the one that can become someone else's question rather than yours. Gaps one and two cost the most hours and hurt the most on the day somebody resigns — which is not the day to start.

In Echo Recruit

Context that travels with the record

Echo Recruit exists because of this exact pattern. Every handover — CV to client, desk to desk, call to contract, contract to invoice, decision to record — keeps its context attached, not just its outcome. Signed quote terms apply themselves to the account and are inherited by the placement. Approved hours become invoice lines at the rate the quote set. Shortlisting rubrics are versioned and evaluations are append-only, so a rejection points at the rule that produced it.

That is a product mention, not the diagnosis. The five gaps above show up whether or not you ever use our software, and four of the five fixes cost nothing but a decision.

A fee agreement that can't reach the invoice isn't an agreement

ashutosh@echorecruit.com (Echo Recruit) — Mon, 28 Sep 2026 06:19:22 GMT

Most fee disputes are not about the percentage. They are about what the percentage applies to, and that sentence is usually the vaguest one in the document.

In short

The basis clause — what the fee is a percentage of — is worth more money than the percentage and gets a tenth of the attention.
A fee floor enforced by policy is a suggestion. A fee floor enforced by an approval before the document can be sent is a floor.
Never edit a sent quote. The client still has the version you sent them.
The agreement should be structured data with a signature on it, not prose somebody has to re-read at invoicing time.

Ask an agency owner what their standard terms are and they will tell you a percentage. Ask what the percentage applies to and there is a pause, because the honest answer is that it depends which template was used and who negotiated it.

That pause is where the money is. A fee agreement is a document whose job is to still be unambiguous eleven months later, in front of a client’s procurement team, when the person who negotiated it has left. Most of them are not written to that standard, and the gap does not show up until it is expensive.

The clauses that decide what you get paid

The fee model, named

Contingency percentage, retained across stages, contract markup or margin, or a flat fee per hire. Say which, once, in a way that does not require interpretation. An agreement that describes a retainer in one paragraph and a success fee in another is an agreement you will be arguing about.

The basis — the clause worth the most

What is the percentage of? Base salary only? Base plus guaranteed bonus? Total first-year cash including a sign-on? Does it include a car allowance, a relocation payment, an expected commission at target?

On a $95,000 base with a $20,000 guaranteed bonus, a 20% fee is either $19,000 or $23,000. That is a 21% difference in what you are paid, decided by a sentence most agreements handle with the phrase “the candidate’s remuneration”.

Write it as a list of what is in and what is out. If a client wants base only, that is a perfectly reasonable deal — it is just one you should be pricing knowingly rather than discovering at invoice time.

The guarantee, in full

Four separate questions, and agreements routinely answer two of them:

How long is the period, and does it run from the start date or the offer date?
Is it a replacement or a rebate? Replacement protects your revenue; a rebate hands it back. Never let the client choose after the fact.
If it is a rebate, is it full or sliding? A sliding scale by week is fairer to both sides and much harder to argue about, because the arithmetic is in the document.
What voids it? Redundancy, a change of role, the client failing to pay the invoice on time, a restructure. A guarantee with no exclusions is a guarantee against the client’s own decisions.

Payment terms, with a start point

“30 days” is half a term. Thirty days from what — the invoice date, the start date, the end of the month in which the invoice was issued? The last of those can add another four weeks and is the one clients’ standard terms quietly assume.

Currency, and who carries the movement

If the salary is in one currency and you invoice in another, name the rate source and the date it is taken. Otherwise the difference is a discussion, and you will lose it, because the client’s finance team does this more often than you do.

The awkward scenarios, decided in advance

Counter-offer accepted after resignation. Candidate withdraws before starting. Candidate you introduced is hired eight months later for a different role. Candidate hired into a second role at a sister company. Each of these will happen; each is unpleasant to negotiate in the moment and straightforward to agree in advance, because at signing time neither side knows which way it will cut.

Who can say yes

Name the roles at the client who can approve an offer and accept a candidate. This costs nothing to include and resolves the specific dispute where a hiring manager engaged you directly and procurement later says they never authorised it.

This is a list of the commercial decisions a fee agreement has to make, not a template and not legal advice. The wording that makes them enforceable depends on your jurisdiction and is worth a lawyer’s afternoon — once, properly, for a document you will use hundreds of times.

Getting it signed, in a way you can evidence

Two questions get asked about a signature when there is a dispute: did this person agree, and can you show that it was them?

“Yes, see the email thread” is weaker than it feels. A name typed at the bottom of a reply, forwarded through three mailboxes, is evidence of a conversation rather than evidence of an agreement. What you want is a signature bound to the document, timestamped, with some verification that the signer controlled the address it was sent to — a one-time code to that address is cheap and makes a meaningful difference to what you can demonstrate.

And both sides should sign. A document only your client signed is an acceptance of your terms; a document you both signed is an agreement, and it looks like one to everybody who reads it afterwards.

Make the floor an actual floor

Every desk has a number below which a deal stops being worth doing. Almost every desk enforces it as a policy, communicated in a meeting, remembered unevenly.

A policy is a suggestion under pressure, and fee negotiation is pressure by definition — it happens at the end of a long conversation, with a recruiter who wants the deal, against a procurement person who does this professionally. The floor loses that fight regularly, and the agency finds out at the quarterly review.

The version that works has three parts:

The floor is a number in the system, not in a handbook, and it can differ by fee model and by market.
Going below it requires a written reason. Not to create bureaucracy — to create a record. Six months of those reasons is the most useful pricing document your business will ever produce.
It requires an approval before the document can be sent, not after. An approval sought after the client has the PDF is a rubber stamp with extra steps.

Never edit a sent quote

When terms change — and they do, usually twice — the tempting thing is to correct the document and resend. Do not.

The client has the first version. If your copy has changed and theirs has not, you now have two documents that both claim to be the agreement, and the only way to tell which is which is a conversation about who has the latest. A revision should be a new, linked version with its own number and its own signature, leaving the original intact and visibly superseded.

The same principle covers the invoice: once a document stops being a draft, both parties’ details, the terms and the numbering should be frozen onto it. An invoice that changes shape because somebody edited the client record afterwards is not a record of anything.

Then the invoice should assemble itself

Everything above only pays off if the agreed terms actually reach the invoice. If they are prose in a filed PDF, they reach it by way of a person re-reading it under time pressure — which is how a fee gets typed four times and goes out wrong on the fourth.

The agreement has to be structured: a fee model, a percentage, a basis, a guarantee, payment terms, a currency — as fields, with the signature attached to them. Then a placement inherits them, approved hours bill against them, and the invoice is reviewed rather than calculated.

Two habits finish the job. Number documents atomically, so two invoices can never take the same number — a duplicate number is a week of reconciliation for both finance teams. And put reminders on a schedule rather than on somebody’s memory: a sequence before and after the due date, with a receivables ledger sorted by how late things are, turns chasing from a monthly panic into a Tuesday.

In Echo Recruit

The quote is the contract

Terms are agreed on a quote as structured fields — contingency, retainer, contract markup or flat fee, in the client’s currency — and both sides e-sign, each confirmed by a six-digit code sent to their own address, with a certificate of completion. A fee below the floor needs a written reason and an admin’s approval before the quote can be sent. Revising a sent quote creates a linked new draft rather than editing history.

On signature the terms apply themselves to the client account, and every later placement inherits them. Commission and approved hours become invoice lines at the rate the quote set; reminders run on a 15/7/5/3/2/1/0-day schedule; numbering is atomic; and both parties’ identity is frozen onto the document the moment it stops being a draft.

The four times a staffing agency types the same fee

ashutosh@echorecruit.com (Echo Recruit) — Thu, 24 Sep 2026 05:47:44 GMT

One number, agreed once, typed four times by four people on four different days. The interesting question is not whether the copies drift — it is which copy reaches the client.

In short

A fee agreed in a document has to be re-entered at least three more times before it reaches an invoice, and each entry is done by a different person on a different day.
The copy that goes out is the last one, typed by whoever does billing — usually from the placement record, usually at month end, usually without the agreement open.
The loss is rarely one big error. It is a percentage point here and a missing bonus inclusion there, on a fraction of placements, which is why nobody has ever put a number on it.
You can measure yours this afternoon: five placements, four places each, count the mismatches.

Here is a sequence every agency owner recognises. A client agrees 22% on base salary, with a 90-day rebate on a sliding scale. It is written into a fee agreement, signed, and filed. Six weeks later a candidate starts on $95,000. The invoice goes out at 20% of base, for $19,000. Nobody notices for five months, and when somebody does, the conversation is not about the $1,900 — it is about why the client should trust the next invoice.

Nothing in that story required anyone to be careless. It required a number to exist in four places at once, which is the normal condition of a recruitment desk.

The four copies

Follow one fee from agreement to cash and you will pass it through four systems, each of which needs it in a different shape.

1. The signed agreement

A PDF. It is the authoritative copy and the only one with a signature under it, and it is also the copy nothing can read. You cannot ask a folder of PDFs which of your clients are below 18%, which have a rebate rather than a replacement, or which ones agreed that the percentage applies to total first-year compensation instead of base. Those are questions with answers; they are just answers that take someone an afternoon and a highlighter.

2. The CRM or tracker field

Somebody types 22 into a field on the client record. Usually the person who opened the account, usually before the final round of negotiation, sometimes from the draft rather than the executed version. This is the copy the desk works from day to day, which means it is the copy that shapes what recruiters believe the terms are. It is also the copy with no audit trail: a field that says 20 today looks exactly like a field that has always said 20.

3. The placement record

When the candidate accepts, the recruiter who filled the role records the placement and the fee. They type it from memory, or from the tracker field, or from a message where the account manager said “think it’s 22 but check”. This copy is where the commission calculation comes from, so it is the number the recruiter has an interest in, and the number nobody else checks.

4. The invoice

Finally, at month end, someone in finance opens the placement record and types an invoice. They are not the person who agreed the fee, they do not have the agreement open, and they are working through thirty of these. If the placement record is wrong, the invoice is wrong, and the invoice is the only one of the four copies the client ever sees.

Four entries, three of them transcriptions. The error rate of a human copying a number between systems is not zero, and it does not need to be high to matter.

What the drift is worth

Nobody publishes a credible figure for this, because the only people who could measure it are the people it happens to. So do not take a number from an industry survey — do the arithmetic with yours.

A forty-placement year, with a small error rate

Placements billed in the year	40
Average salary placed	$95,000
Fee agreed $20,900 per placement	22%
Placements invoiced two points light one in ten, at 20% instead of 22%	4
Shortfall on each	$1,900
Left on the table	$7,600

Substitute your own placement count, average salary and honesty about the error rate. The figure that matters is not ours — it is what falls out when you put your own four numbers in.

$7,600 is not a catastrophe. It is roughly a third of a placement, given away for free, in exchange for nothing, by a process everyone believes is working. And it is the cheap version of the failure. The expensive versions are:

The inclusion question. The agreement says the fee applies to total first-year cash including guaranteed bonus. The invoice is calculated on base. On a role with a $20,000 guaranteed bonus that is not two points, it is a fifth of the fee.
The rebate you cannot evidence. A candidate leaves on day 84. The client claims a full refund; you believe it was a sliding scale. The agreement says which — if anyone can find the executed version rather than the third draft.
The client who starts checking. Once a client has caught one invoice, every subsequent invoice is checked, queried and paid late. The cost of that is not in any of your reports, and it is larger than the error that caused it.

Why “be more careful” does not fix it

The instinctive response is a process note: check the agreement before invoicing. It will work for about six weeks. It fails for the same reason every manual reconciliation fails — it asks the person with the least context, under the most time pressure, at the end of the chain, to catch a mistake made by somebody more senior at the start of it.

The second instinctive response is a better ATS. That does not fix it either, because the ATS is not where the fee lives and never was. The break is not inside any of your tools. It is in the four handovers between them, and a tool cannot fix a handover it is not present for.

The only structural fix is that the fee stops being a number and starts being a record: it exists once, in a form software can read, with a signature attached to it, and everything downstream reads it rather than being told it.

Audit your own desk this afternoon

Before you buy anything, find out whether you have this problem and how big it is. It takes about ninety minutes and you need nothing but access.

1Pick five placements from last quarterNot the five you remember. Take the five most recent that have been invoiced, so the whole chain has run. Include at least one contract placement if you run a contract desk — those have rates as well as fees, and rates drift faster.
2Find the fee in four places for eachThe executed agreement, the client record in your CRM, the placement record, and the issued invoice. Write all four down. Note how long each one took you to find — that number is as informative as the fee itself.
3Check the basis, not just the percentageDoes the agreement say base salary, or total first-year compensation? Does the invoice agree? A matching percentage applied to a different basis is the failure that is worth the most and looks the most correct.
4Count the mismatches and price themMultiply by your annual placement volume. If five out of five match, you have a disciplined desk and a reconciliation cost instead of a leakage cost — which is still a cost, just one paid in somebody’s Thursday.
5Ask who could have caught itFor each mismatch, name the person and the moment. If the answer is “finance, at month end, if they had opened the PDF”, then your control is one tired person with thirty invoices to raise.

What “fixed” actually looks like

Fixed does not mean everyone is more careful. It means the retype is not possible, because there is only one copy:

The fee is agreed as data — a fee model, a percentage, a basis, a guarantee, and payment terms as structured fields — not as prose inside a document.
The signature attaches to those fields, so the thing that was signed and the thing the system holds are the same object rather than two objects that ought to agree.
On signature, the terms apply themselves to the client account. Nobody transcribes them into a CRM field afterwards, because the CRM field is the agreed term.
A placement inherits the account’s terms rather than asking the recruiter what they were.
The invoice is assembled from the placement and the terms it inherited. Finance reviews and sends; they do not calculate.
Changing terms produces a new version rather than editing the old one, so the question “what were the terms in March?” has an answer.

Read that list back and notice what it removes: not effort, but opportunity. There is no moment at which a person is asked to remember a number, so there is no moment at which they can remember it wrong.

In Echo Recruit

Where the fee lives, and what reads it

The fee is agreed on a quote — contingency percentage, retainer, contract markup or flat fee, in the client’s currency — and both sides e-sign it, each confirmed by a six-digit code sent to their own address. On signature the terms apply themselves to the client account. Every later placement inherits them, and invoice lines are generated from the placement and the approved hours behind it rather than typed.

Two details matter more than they sound. A fee below your floor needs a written reason and an admin’s approval before the quote can be sent, so the floor is a floor rather than a policy. And revising a sent quote creates a linked new draft instead of editing the original, because the client still has the PDF you sent them.

The general case

The fee is the clearest example because it is a single number with a dollar sign on it, but it is not the only one. The same shape shows up wherever your stack hands work across a boundary: a bill rate agreed with the client and a pay rate agreed with the contractor, typed into two systems. Hours approved in an email and keyed into a spreadsheet. A start date in the ATS, a different one in onboarding, and a third on the first invoice.

Each of them is the same fault: one fact, several copies, no single owner. Fix the fee first — it is the one with money attached and the one you can measure in an afternoon. Then look for the others, because now you know the shape.