The Harland Rowe file

This file is a case study and a teaching module at the same time. Three devices run through it:

• Terms you can tap. Words like grab, Role Card and net fee income are dotted. Tap one for a plain-language explanation. No prior knowledge of recruitment, finance or AI is assumed.
• Figures you can audit. Dotted numbers like 4,630 hours open the arithmetic behind them: the formula, the inputs, and where each input comes from. Auditability is the method.
• Method notes. Green-edged boxes step outside the story to explain what the methodology is doing at that point and why, so you can follow the discipline itself, not just this telling of it.

By the end you should be able to answer four questions: how a business finds where its time actually goes, how an AI hire is specified so a manager can hold it to account, how a term is run week by week through a seasonal trough and a surge, and how savings are counted so that nobody has to take anybody's word for anything.

History and ownership

Carol Harland founded the firm in 2003 with David Rowe, out of two desks in a serviced office on Lever Street. She had spent eleven years at a big-firm recruiter and left with one rule she has never bent: never send a CV you have not defended in person. Carol built the finance and accountancy desk; Rowe, an engineer by background, built the engineering clients and the contract book. Rowe retired in 2011 and Carol bought his shares over three years, keeping his client base and his name on the door. The firm has been profitable every year since 2004, including 2009 and 2020.

The expansion decision that shapes today's cost base is the retained search practice, launched in 2016 after Carol lost three senior mandates in a row to London firms. She hired a researcher, took a London meeting-room arrangement in 2018, and now fronts every mandate herself. The practice works: £432k of NFI on two salaries, and it wins the firm its best introductions. But it carries the £18k London arrangement and the bulk of £160k a year in travel and entertaining, and every pound of it is Carol. When she is on a mandate, which is most weeks, nobody else is running the firm.

What makes the business genuinely good is not written down anywhere. Carol and her top biller, Steve Marsden, hold twenty years of knowledge of who is good, who is movable and what North West finance and engineering employers actually pay. Clients come back because the CVs are defended, not forwarded. Carol is 58 and intends to sell or hand over within about three years. She knows what a buyer will see: £505k of profit, and £980k of NFI, 28.6% of the total, that walks out of the building if she and Steve do, with no documented process behind either of them.

People

Five people matter most to what follows. Carol Harland is still the best biller in the building when she chooses to be: she wins every retained mandate personally, interviews every shortlist candidate herself, and signs every credit note. Her diary is the firm's real operating system. Steve Marsden, 47, billed £548k in FY25, a third of the firm's perm fees, from a client list he keeps in a hardback notebook and a laptop spreadsheet called SM master list, 1,100 names, last backed up when IT forced it. He enters the minimum into the CRM to get paid, and he is openly sceptical of process, and says so. Pauline Grady, 61, nineteen years in, runs contractor compliance, right-to-work checks, AWR clocks, opt-outs and insurance certificates from a 14-tab spreadsheet only she fully understands, alongside the terms-of-business filing and half of billing. She wants to retire in the summer of 2027. Nobody is trained on the spreadsheet; when she took two weeks' leave in June 2025, four re-check dates were missed, and one became the August incident described below. Jade Okafor, 26, the shared resourcer, formats roughly 30 of the firm's 75 weekly outbound CVs, runs first-pass sifts for five consultants, and books interviews when consultants are out; work queues behind her for a day and a half when both desks are busy. She has asked twice about a consultant role. Asif Mahmood, 45, finance manager, runs Sage, the weekly contractor payroll and the Monday invoice-discounting reconciliation, and quietly maintains his own shadow debtor spreadsheet because he does not trust the Sage aged-debt report after a mispost in 2023.

The trading engine

Systems: the honest mess

Bullhorn since 2017: 118,000 candidate records, of which a March 2025 sample found 23% duplicates, with 41% of email addresses more than three years old. Consultants record what the commission scheme requires and little else; Steve's notebook, a shared contractor-availability spreadsheet and at least six personal candidate lists live outside the CRM entirely. Nine LinkedIn Recruiter seats; three job boards posted manually. Compliance lives in Pauline's 14-tab spreadsheet, where the re-check column is only as good as the last person who filled it in. Terms of business are a Word template; signed copies sit in a filing cabinet and in sent-items folders, with no central register, and Pauline's February 2025 audit found signed terms on file for 68% of the 212 billed clients. Timesheets run through the ETZ portal, which works but does not chase: that is Tyler, by phone, every Monday. Finance is Sage 50 plus the lender's portal, which do not talk to each other; Asif rekeys the invoice schedule every Wednesday.

The five processes that run the business

Role intake to shortlist

The perm desks take 59 roles a month, 712 in FY25. A brief arriving through jobs@ has already waited a median 1.4 working days. The role reaches Bullhorn the same day 62% of the time; the advert, 45 minutes of work, waits a median 1.2 days in the consultant's batch queue. Applications then arrive at a median 58 per role, roughly 41,000 a year, and 35% are never opened by anyone (measured from the viewed flag on a 60-role sample). Reviewing the rest takes about 3.6 hours per role; CRM and LinkedIn searching another 2.5, of which roughly 30% is wasted on duplicates and dead records. The shortlist of six CVs is reformatted into the house template at 20 minutes per CV, queueing behind Jade for a median 1.5 working days when both desks are busy. First CVs reach the client a median 6.1 working days after the brief (cohort: all 178 perm roles opened April to June 2025, followed to day 60, Bullhorn timestamps).

Shortlist to interview to offer

Client feedback on CVs takes a median 3.9 working days, chased twice by phone; 22% of clients take more than seven. Scheduling an interview takes about five calls and emails, 35 minutes of consultant time. Median time to first interview was 18 days in Q3 2025; it was 12 two years ago. Offers run about 22 a month with 82% acceptance. And 607 CV-sent candidates, 17.5%, withdrew mid-process in FY25; 31% of those withdrawals, 188 people, followed a silence of seven or more working days while they waited on feedback or confirmation.

Placement to invoice, including terms of business

The consultant confirms a start by phone, then completes a Word start form a median 2.1 working days later; 18% need chasing. Karen checks terms: a filing-cabinet lookup for the 68% of clients with signed terms on file, and for the rest the invoice goes out on the strength of terms sent with the first CV and never returned. Perm invoices are raised in a Friday batch; 31% of clients require a PO, and a missing one adds a median six working days. Rebate terms on the standard template: 100% if the candidate leaves in weeks one to four, 50% in weeks five to eight, conditional on payment within terms; in practice consultants offer a free replacement instead about half the time, which is cheaper when it works and invisible when it does not.

Contractor onboarding, compliance and timesheets

About 146 contract starts a year, peaking in January and September. Pauline runs the 11-item onboarding checklist in a median 1.8 working days; a Friday-for-Monday start compresses it to hours, which is where corners get cut. Right-to-work re-check dates sit in tab 4 with a manual follow-up column. Timesheets are due Friday 17:00; on a typical Monday 15 of 97 are missing, and Tyler chases each at about ten minutes, roughly 2.5 hours every Monday morning. Payroll runs Tuesday, client invoices generate Wednesday, the schedule is rekeyed to the lender portal the same day, the 90% advance lands Thursday, contractors are paid Friday.

Invoice to cash

Perm terms are 30 days but perm debtor days run at 66; contract runs at 48; blended, 52. Denise chases from day seven overdue, about 11 hours a week across roughly 120 overdue invoices. Queries, about nine a week, take a median eight working days to resolve and stop the whole invoice, not the disputed line. FY25 disputes and credits totalled £26k, told in full below. Invoices unpaid at 90 days are disapproved by the lender and clawed back from availability; £31k was disapproved during FY25 and funded from headroom until collected.

Performance and measurement

Eight quarters, calendar basis. NFI, placements, contractors, debtor days, disputes and facility utilisation come from Sage, payroll and lender statements. Time-to-interview, fill rate and drop-out are not tracked by the firm and were reconstructed in October 2025 from Bullhorn activity timestamps; the cohort definitions are stated under the table, because without them the numbers can be made to say anything.

FY25 against FY24: NFI down 3.0% in a market where the listed recruiters fell 4.6% to 14%. Roles taken up 6%; placements down from 211 to 202; fill rate down from 31.5% to 28.4%; time-to-interview up from a 12.5-day average to 16.5; debtor days up from 45.5 to 50.5. Volume grew, the contract book grew, and the back office that serves both stood still; the service metrics paid for it.

What Carol's monthly numbers show: NFI by desk against last year, placement count, cash at bank, facility headroom, top ten debtors. Nothing in the pack measures time-to-interview, fill rate, drop-outs, terms coverage, compliance currency, queue lengths or per-consultant conversion. Carol manages those by walking the floor, which worked at 18 staff and does not at 30.

What it costs them

Every figure below is a multiplication of numbers already stated.

• Consultant capacity burned on admin: £103k a year. The 4,630 hours of Exhibit 3 at the loaded consultant rate. Capacity, not cash: it is what the firm pays for its own admin at consultant rates.
• Fees lost to latency-attributable drop-outs: £107k a year. Of the 188 withdrawals that followed a seven-plus-day silence, 26 were the lead candidate on a role that subsequently died unfilled. Attributing half of those to latency, conservatively, and none of the other 581 withdrawals.
• Cash cost of debtor drift: £8.1k a year. Five days of drift against the firm's own FY24 average, financed at the 6.35% discount rate, plus £31k of lender disapprovals funded from headroom until collected.
• FY25's actual named losses: £24.6k. Padfield Group, £13.9k: a finance director placed in November 2024, fee £22k, no signed terms could be produced, settled at £11k plus £2.9k legal. Brindle Logistics, £8.2k: a financial controller who left in week five, 50% rebate. And the right-to-work near miss, £2.5k: a welder at Ashbrook whose visa-expiry follow-up column sat unfilled after Pauline's June leave; his status was in fact valid, but the evidence was not on file, he was stood down four days while a fresh check ran, and the exposure had it been invalid was a civil penalty of up to £45k.

Why now

Carol intends to be out within about three years, and she has read enough sale prospectuses to know what diligence will find here. A buyer sees £505k of profit and then applies the discounts: 28.6% of NFI concentrated in two people; a top biller whose client book lives in a notebook; compliance, billing control and the terms archive in the head and spreadsheet of a manager who retires in 2027 with no successor; signed terms for 68% of clients; a CRM that is a quarter duplicates; and service metrics nobody inside the firm could produce without this profile's reconstruction. None of it is fatal, and all of it prices the multiple.

What moved this from a worry to a decision was February 2025. The Padfield work was good: right candidate, placed in six weeks, still in post. The firm lost £11k of a £22k fee anyway, plus £2.9k of legal costs, because nobody could produce a signed terms document for a client it had billed three times before. Carol signed the settlement herself and described it afterwards as paying eleven grand to learn what a filing cabinet is not. Six months later the Ashbrook scare showed her the same shape on the compliance side, with a £45k penalty behind it instead of a halved fee. Two near misses in one year, both process failures, neither anyone's dishonesty, both survivable. The third, in diligence season with a buyer's lawyers watching, might not be. She asked for this profile to be prepared.

Carol Harland did not call because of a technology ambition. She called because of February 2025, and because she has read enough prospectuses to know that the third process failure, in diligence season with a buyer's lawyers watching, is the one that prices the business. TCEL's position from the first meeting was the one it holds everywhere: AI belongs inside your strategy, not on top of it. So the first work was not a tool selection; it was reading the firm's own numbers until they ranked themselves.

1. Latency in the perm engine. Median time to first interview has moved from 12 days to 18 in two years. Roles taken rose 6% in FY25; placements fell; cohort fill rate fell from 31.5% to 28.4%. The firm's own reconstruction attributes £107k of FY25 fees to withdrawals that followed silences of seven or more working days, and 35% of the year's roughly 41,000 applications were never opened. The largest measurable leak, and it compounds: every lost day costs candidates, and every lost candidate costs fees.
2. Compliance and terms as unpriced liability. Signed terms on file for 68% of 212 billed clients. FY25's named losses, £24.6k, were all process failures, and the control for all of it is one manager's 14-tab spreadsheet, undocumented, with a successor problem dated summer 2027.
3. Consultant capacity burned on admin. 4,630 hours a year, 2.8 FTE, £103k at the firm's own loaded rate: capacity paid for at consultant rates and spent at clerk rates.
4. Cash drag. Debtor days drifted from 45.5 to 50.5, £8.1k a year in financing, plus £31k of lender disapprovals funded from headroom.
5. Key-person concentration. Carol and Steve are 28.6% of NFI with nothing written down. This leak is real, and it is not addressable by hiring anything. It is addressed by choices only the owner can make.

Three observations from the operator's chair. The exit multiple is set by what diligence finds, not by what the P&L says: a buyer does not price the Padfield loss at £13.9k; a buyer prices the system that allowed it, across every client, forward. The cheapest multiple points this firm can buy back are process points, and they are buyable in months. The firm tells itself a candidate-shortage story; the numbers tell a latency story: shortage is real in North West finance, but shortage does not explain 14,400 unread applications, nor 188 withdrawals that followed the firm's own silence, nor a six-day internal queue before the first CV goes out. A shortage story locates the problem in the market, where nothing can be done; the measurements locate most of it inside the building, where something can. Compliance is a sale-killer, not a cost line: in a share sale, the right-to-work file, the AWR clocks and the terms archive are the pages a buyer's lawyers actually read, and a firm that can show every tracked date and every signed terms document is worth more, because warranty negotiation starts from the file rather than from the owner's assurances.

The decision logic follows from the split, and one interest is declared with it: GRABS is TCEL's sibling, and a diagnostic that always lands on the family product is worth nothing. The disconfirmation test is on the page above. The largest leak by value, key-person concentration, was ruled a strategy problem that no hire can touch, and if the other leaks had been the same shape, the recommendation would have been board work, not Workers. They are not the same shape: the latency leak, the compliance leak and the admin burn are defined work, well understood, short of hours, exactly what a fixed-term hire fixes. TCEL recommended intervening first where the KPI the whole perm business turns on is measured, role intake to interview, with the compliance and billings work put under management in the same term, because it is the cheapest insurance the sale process will ever buy. GRABS was commissioned to run its Process Diagnostic before anything was bought.

One process, one week: role intake to interview, Monday 20 to Friday 24 October 2025. Measured from the firm's own records: Bullhorn timestamps, the jobs@ inbox, consultants' sent items and calendars, and the job-board logs. Nothing recalled in a workshop. In the week the firm took 15 new perm briefs, received 804 applications, sent 73 CVs and held 29 first interviews. Eleven candidates withdrew from live processes; four of those followed a silence of seven or more working days, including one who wrote, "I assumed the role had gone."

Tap each row of the grab map for where the work waits, who touches it, and what the wait costs at the firm's own numbers.

The surprise finding. The firm believes it operates in a candidate-shortage market. The week's measurements say it is short of attention, not candidates. Of the 158 applications that arrived on the Monday, 55 were still unopened at Friday 17:00. On the Thursday, a management accountant was interviewed whose CV had sat unopened for 14 days until the client chased. Two measured facts carry the week: a shortlisted role holds about 8.1 hours of actual work spread across 6.1 working days of elapsed time, so the work is being worked on 18% of the time; and the applicants the market supposedly does not have are arriving at roughly 800 a week, a third of them read by nobody.

The diagnostic closed with the Role Card for the hire that would fix it, a second card added by TCEL's analysis for the compliance and billings work, and the readiness list with named owners. Everything it produced was the firm's to keep, hire or not.

Signed Friday 31 October 2025 by Carol Harland and Asif Mahmood, before either Worker started. GRABS counts events; the client prices them, in advance, in writing. Two of Asif's choices set the tone. He priced every hour over a 1,950-hour paid year rather than the 1,650 contracted hours the profile used, so no saving is flattered by the denominator. And he excluded commission from the rates: commission follows billings, not hours, and adding it would double count the selling the hours are released for.

Each hire is proposed on one page, the Role Card: six parts, in the language of a job spec. Both terms ran 90 days from Monday 3 November 2025.

Onboarding: the vault, built in days

Day 1: the Bullhorn API user created under Asif's ownership; the house CV template captured with Jade in under an hour; criteria sessions begin. Day 2: Pauline's first half-day capture, tabs 1 to 7 of the spreadsheet, Becky present; Karen and Denise give the rebate scale, the PO list and the chase sequence in ninety minutes. Day 3: Pauline's second half-day, tabs 8 to 14, including the follow-up column rules that failed in June; Tyler gives the timesheet ritual in 45 minutes; Dec signs criteria for seven live engineering roles. Day 4: Steve Marsden gives 40 minutes, supplies criteria for his two live roles verbally and declines to hand over the notebook; the finance consultants sign theirs; the DPIA is signed; the disclosure wording goes into every candidate-facing template; the client-disclosure decision is minuted; and Becky and Sharon begin scanning the filing cabinet's 144 signed terms into the register, finished 21 November.

The day-5 finding. Friday 7 November, the first full scan of the 118,000-record candidate base: 26,400 probable duplicates (22.4%, confirming the March sample), 48,300 primary addresses older than three years, and 37 candidates with do-not-contact or data-removal requests sitting in free-text notes with no suppression flag set. The flags were set the same day and the breach assessment minuted. A cleanup sprint ran the following week across the 11,600 records attached to live roles and the last 24 months of applicants, and all candidate outreach was gated until the suppression work was verified, so keep-warm contact started a week later than planned. The full-base merge ran in the background through the term: 24,900 duplicate pairs merged by 16 January, 1,500 ambiguous pairs listed for human decision.

The connectors. Both Workers reach Bullhorn through its REST API under one dedicated licence (£90 a month, a simulation assumption), owned by Asif. Diaries through Microsoft 365 free/busy. The timesheet portal is read through its reporting API. Sage 50 has no usable API, so a nightly export runs from the office server into the vault and invoice drafts travel back through Sage's import routine for Karen to post; the lender portal has no API at all and remains Asif's Wednesday rekey, unautomated and said so. One connector failed in the term: a Windows update on 4 December silently stopped the nightly Sage export. The Monday Report's data-freshness line showed it on the 8th, the Worker's chase run held itself automatically because the balances were stale, and Asif re-enabled the job by noon. The export now writes a heartbeat that is checked daily.

30-day review, Wednesday 3 December 2025. Baseline integrity confirmed; the criteria co-factor at 100% since 24 November; the keep-warm descope to active candidates ratified. Continue.

The complaint. Thursday 4 December, a regret email, human-released by Claire Dunwoody, drew a reply objecting to the line "your application was reviewed with the assistance of automated screening against the role criteria": the candidate said it read like being marked by a machine. Sophie Dean rewrote the disclosure across every template to: "Software helped us sort applications for this role; every decision, including this one, was made by the consultant named above." Pauline confirmed the new wording still meets the transparency requirement. The disclosure stayed; the temperature changed.

60-day review, Monday 5 January 2026. December throughput redirected honestly, no December claims made; the Worker's proposal to extend direct booking to six more clients approved for three, by Carol. Continue.

The incident. Tuesday 13 January 2026. Interview confirmations for two candidates at Lathom & Birch went out in an approved batch showing the client's registered office on King Street West when the interviews were at the Altrincham site. The approval queue had displayed the message template, not the resolved address, so the error passed a human approval. One candidate travelled to the wrong building; the interview was missed and the client's finance director was kept waiting twenty-five minutes. Aisha Rahman phoned the candidate and the client within the hour, apologised under her own name, put it in writing and rebooked for the 15th. Two rules changed the same day: the approval queue now renders fully resolved text, never templates; and any confirmation where the role's default location differs from the calendar invite routes to a human. The catch log records it as a miss, because the log records misses too.

90-day review, Friday 30 January 2026. Numbers as the week-12 report; handover attested; four priced catches and one logged miss on the term's log; renewal decision deferred to 13 February. Term complete.

• Client-side interview feedback is still the longest wait: median 3.8 working days against 3.9 at baseline, with 22% of clients over seven. It is a commercial conversation between a consultant and a client, not a Worker task, and the term did not pretend otherwise.
• The retained search practice was left deliberately alone. Carol's network is the product; nothing in it was automated and nothing should be.
• Steve Marsden's desk adopted partially. His roles ran a median time-to-interview of 14 days against his desk's 10. The notebook is intact, and into February he was still keeping a shadow longlist for each live role; the morning collision list is what keeps that workable.
• December perm hiring stayed dead, as it always does: 27 briefs against the prior December's 31. No claim survives contact with December, and none was made.
• The jobs@ inbox and the website form remain unowned: 1.4 working days median to reach a consultant at the last sample, outside both cards' scope.
• Disputed invoices still stop whole invoices, not disputed lines, and still take about eight working days: a Sage limitation the term did not touch.

And the one failure on the record, restated once: on 13 January a Worker confirmation sent two candidates the wrong interview address past a human approval, one missed interview resulted, and the correction and both rule changes ran the same day under Aisha Rahman's name.

Attested Friday 30 January 2026 by Pauline Grady, line by line against the vault index. The firm keeps, outright:

• the cleaned candidate base: 24,900 duplicate pairs merged, 48,300 stale addresses flagged, 37 suppressions enforced, 1,500 ambiguous pairs listed for human decision;
• the criteria library: 61 signed criteria sheets, one per role sifted, each with its tier reasoning attached;
• the compliance calendar: 403 tracked dates (41 right-to-work re-checks, 118 insurance certificates, 99 AWR clocks, 52 CSCS cards, 93 opt-out confirmations), every action evidenced;
• the terms-of-business register: keyed by company number with an alias table; signed coverage 181 of 213 active clients, 85%, against 68% in February 2025, with every new client signed before first CV since 9 January;
• the playbooks: seven, covering sift-and-tier, formatting, scheduling, acknowledgement and keep-warm, the terms gate, the compliance calendar and the chase sequence;
• the records: the Work Log's 11,284 task entries, the catch log with its four priced catches and one logged miss, the approval-queue history, and the baseline-to-day-90 measurements.

One change-management item is open at attestation: the approach-before-call rule holds because a Worker publishes the morning collision list, and if neither Worker is renewed, nobody owns that list. It is recorded as unassigned.

"It chases the dates so I don't have to. I still open tab 4 on a Friday, old habit, but it's ten minutes now, and nothing in it is a surprise." Pauline Grady, at the attestation, recorded with her permission

The renewal decision, due 13 February, is a real trilemma. Renew both Workers at the assumed £7,000 a month. Renew only the Compliance and Billings Worker, on the argument that the desk playbooks are the more runnable in-house while the compliance exposure is the sale-killer. Or renew neither and run everything from the vault and the playbooks, accepting that the collision list dies. In every branch, the commission scheme is restated before any year two. Renewal is a choice, never a default.

The term moved the firm's diligence position more than its P&L. A buyer's lawyers now find a terms register at 85% and climbing with a hard gate behind it, a compliance calendar with evidence attached where a 14-tab spreadsheet stood, service metrics the firm can produce monthly instead of reconstructing once, and a documented process behind the perm engine. Asif's one-page pack has a second page: time-to-interview, terms coverage, compliance currency and queue lengths, cohort-defined. None of that adds a pound of profit this quarter; all of it defends the multiple on £505k, which is where the real money in a three-year exit sits. The lender conversation has moved the same way: disapprovals ran £31k in FY25 and the term's three months produced one, collected inside the window; Asif briefed the lender on the AI-prepared drafts in November, as the facility's process covenant requires, and the April facility review starts from a book at 46 days with a register behind it, which is the firm's side of that conversation.

The key-person risks now read differently, and honestly. Pauline's retirement in 2027 is no longer a cliff: the spreadsheet's rules live in the vault, and her successor inherits a documented system rather than a person's memory. Steve remains Steve: his book is in his notebook, his longlists shadow the system, and that is a negotiation about equity, earn-out and succession, not a process problem; it should be had this year, not in diligence season. Carol's network is the retained practice, and it should be priced as hers in any sale structure rather than disguised as firm goodwill; the £160k of travel and entertaining that rides with it belongs in that pricing conversation too.

The questions that remain are the owner's, and none of them is an AI question. Whether to grow the finance desk or the contract book, now the perm engine has reclaimed capacity. Whether the two sole-supplier PSL positions and the 15% discounts are assets or anchors. Whether Ashbrook at 9.8% of NFI, close to the facility's concentration limit, is a relationship or an exposure. How the retained practice is priced, staffed and eventually handed to Fiona Marsh. And Jade Okafor asked again in January about the consultant role. Nobody has answered her yet.

Everything you just read is one method applied once. It has three steps, and you have now seen each of them working in a second industry; the companion file applies them to an industrial distributor and arrives, by the same route, at different hires.

And underneath all three, the counting discipline from Part seven: events counted by the supplier, prices signed by the client in advance and argued down rather than up, and three columns of different evidence that are never blended into a number you would have to take on faith.

Run it on your own numbers

Three questions, straight from this file to your firm:

1. What share of the applications you received last month was never opened by anyone? If you cannot answer from your CRM's timestamps, that is the first finding.
2. How many of your billed clients have signed terms on file, as a percentage you could show a buyer's lawyer tomorrow? Harland Rowe thought it knew; the audit said 68.
3. What is your median time from brief to first interview, measured on a stated cohort, and what was it two years ago? If the answer is a feeling rather than a figure, so is your service.

The board pack is a ready-to-edit presentation for taking this case to your own board, in your voice, with every figure left as a slot for your numbers.

Or write to [email protected].

Harland Rowe is fictional; the industry it trades in is not. The firm's shape, its pains and the engagement's results were calibrated against the published sources below, graded by how much weight each will bear. Where the simulation needed a number the public record does not supply, it stayed inside the nearest published range, and the ranges are shown.

Industry and sector research

• REC, UK Recruitment Industry Status Report 2024/25: industry structure, the temp and contract share of industry value, placement volumes, and a market in which recruitment activity fell in 2025.
• TC Group / Recruitment Accountants, Recruitment Sector Benchmarking Report 2025 (78 UK agencies): the 60:20:20 staff-overhead-profit benchmark and actual 2024 to 2025 cost ratios; Harland Rowe's 60.5% staff cost and 14.7% margin sit inside the reported spread.
• Recruiter HOT 100 2025: top-decile NFI per head of £136k; the simulated firm's blended figure sits deliberately below the elite band.
• Recruiterflow benchmarks across 2,100 search and recruiting firms, with candidate.fyi funnel data: CV-to-interview, interviews per placement, offer-acceptance and fill-rate ranges behind Exhibit 4.
• SmartRecruiters Benchmarks 2025 and NatWest Mentor: UK time-to-hire averages; standard contingent fee percentages (15 to 20%) and retained thirds from multiple converging fee guides.
• Published UK invoice-discounting pricing (service fee plus discount charge of base plus 1 to 3%): the simulated 6.35% facility sits inside the band.
• GOV.UK: employer National Insurance from April 2025 and auto-enrolment minimums, the basis of every loaded rate; and the civil penalty regime for illegal working, tripled in 2024, behind the £45k exposure in the Ashbrook near miss.

What AI delivers in hiring, and what it must not do

• Screening-assistance evidence, vendor claims discounted: a 40 to 60% reduction in human review time is the defensible band; the headline "70 to 75%" figures are vendor showcases and were not used.
• Scheduling automation (Paradox, GoodTime and class): cycle times from days to same-day in mature deployments; the simulation automates 85% of bookings only where clients agreed standing slots.
• ICO guidance on AI in recruitment (2024, 296 recommendations), EHRC guidance, and UK GDPR article 22: the legal frame behind no solely automated rejection, disclosure, data minimisation and the DPIA.
• Harvard Business School, Hidden Workers (2021): 88% of employers say automated screening rejects qualified candidates; the reason the simulation runs a monthly blind re-sift for false negatives, not just a bias report.
• University of Washington (2024): large language models preferred white-associated names in 85% of CV comparisons; the reason the Worker ranks only against human-signed role criteria and never against placement history.
• Reuters (2018): Amazon's scrapped CV-screening engine, trained on its own hiring history, is the standing parable for that same rule.
• Pew Research and Talent Board candidate-experience research: most candidates distrust AI-decided hiring, which is why judgement contact (feedback, rejection, offers, advice) stays entirely human and only transactional contact is automated.
• RAND Corporation (2024): over 80% of AI projects fail, most often on misstated problems and data quality; the day-5 duplicate-and-suppression finding is that result, lived. S&P Global (2025): 42% of firms abandoned most AI initiatives, usually on change management; the shadow-longlist collision is that result, lived.