Candidate Rediscovery: Why Your ATS Can't Find People You Already Have

Your database is probably already your best source of placements. Your ATS just can't find the people in it. Here's why, a 12-test audit to measure it, the signals that show who is ready to move now, and the routine, messages and rules that turn it into placements.

Last updated: November 2026. Written by Artem Pravda, CPO & CDO at Execue, an AI sales employee for recruitment agencies; how Execue fits is covered near the end. Figures from vendors are labelled as such, and statistics we couldn't trace to a source are left out. This is operational guidance, not legal advice.

Quick answer

Your ATS can't find people you already have mainly because keyword search often skips notes and older CVs, matches exact words rather than meaning, and silently drops records with blank fields, duplicates or failed parsing. Candidate rediscovery (also called talent rediscovery) is the habit of fixing that and re-engaging those people before you source externally.

It is usually your best channel already. In Recruiterflow's 2026 benchmark of more than 2,100 agencies on its platform, 71% of placements went to candidates who were in the firm's CRM before the job opened. Yet in Top Echelon's 2025 survey of US and Canadian agency recruiters, only 13.8% named their own database as their top sourcing channel.

The seven causes:

1. Search skips notes and older CVs.

2. Words don't match ("team leader" vs "led a team of six").

3. Filters drop records with blank fields or archived status.

4. Parsing broke the record.

5. Duplicates split one person's history.

6. Records go stale: roughly 40-70% of a five-year-old database has changed employer.

7. Habit and ownership send recruiters to LinkedIn first.

Timing matters as much as finding: what people tell you, contract end dates and events at their employer predict a move far better than tenure or "likely to move" scores. The fix is a routine, not a purchase.

The numbers that matter

• 71%: share of placements going to candidates already in the CRM before the job opened, across 2,100+ agencies on Recruiterflow (2026, vendor data)

• 13.8%: share of agency recruiters who name their own database as their top sourcing channel (Top Echelon survey of US and Canadian agency recruiters, 2025, vendor)

• 40%: share of revenue a UK agency found came from its existing database once it started tracking candidate source (Tracker case study, 2026)

• 4.1 years: US median job tenure in January 2026, and 20.6% of workers had been with their employer 12 months or less (US Bureau of Labor Statistics)

• 10-15%: share of records that are duplicates in a typical database, per Bullhorn partner Kyloe

• About 20%: share of CVs using multi-column layouts, which parsers often scramble (2025 preprint by Alibaba-affiliated researchers)

• 35% vs 5%: reply rates to re-engagement from candidates who had replied before versus those who never had (Noon survey of about 1,000 recruiters, 2026, vendor)

• 31%: rise in voluntary quits the year after a 1% workforce reduction (Trevor and Nyberg, Academy of Management Journal, 2008)

• £40,000: the ICO's fine on Reed Online for emailing its jobseeker database to promote a CV-review service, including millions who had opted out (2022)

How to read this guide

• Want to know how bad it is in your ATS → the seven causes and the 12-test audit

• Want quick wins before buying anything → get more from the search you have

• Want to know who to call first → movability signals

• Building the habit → the weekly routine and what to say

• UK or US compliance → the rules

• Choosing a tool → what fixes which cause

• Where Execue fits → an honest note

If you want the step after this one, matching a specific job description against your database, see our guide to sourcing from your own ATS. This guide is about why the people aren't coming back when you search, and which of them to approach now.

Your database is probably already your best source

A familiar Monday (a composite, but every agency owner will recognise it): a client sends a hard brief for an IFRS 17 specialist. The consultant runs a quick keyword search in the ATS, gets nothing useful, and spends two days on LinkedIn. On Thursday a colleague mentions that she spoke to exactly that person in 2024. The candidate is in the database, under a different job title, with the IFRS 17 detail in a call note that keyword search never reads. By then a rival agency has already sent the client a CV.

The numbers behind that story are lopsided. The 71% figure above comes from Recruiterflow's agency benchmark (vendor data, and "in the CRM" includes people added days before the job opened). Another vendor benchmark, StaffingHub's 2026 report on US staffing firms, puts "ATS rediscovery" at 16% of placements, because definitions differ widely, and published agency case studies cluster at roughly a third to 40% of placements:

• Cityscape Recruitment (JobAdder): about a third of placements from its own database

• Soni Resources Group, US (Bullhorn): 38% of placements

• Kester Search Group (Daxtra): from about 10-15% to 30-40% after it fixed how job-board CVs entered the database, with time to client submission roughly halved

• Elevation Recruitment Group, UK (Tracker): 40% of revenue

These agencies range from mid-sized to large, so treat the figures as patterns rather than benchmarks for a 10-person firm.

Elevation's case says the most about the problem. The agency only saw the 40% once its system tracked where each placed candidate came from. The likely reason: candidates first found on a job board or LinkedIn get credited to that channel, even when they're placed years later from the database. That's how the 71% and the 13.8% can both be true: the database already does the work, and nobody can see it. In Firefish's 2026 survey of 132 UK and Irish agency leaders (vendor), the existing candidate database was the highest-rated channel for return on investment, rated positively by 80%. The surveys asked different questions, so treat the contrast as indicative.

A note on the statistics you'll see elsewhere. Rediscovery articles often quote "75% of ATS records are never re-engaged", "silver medalists hire at 3x the rate of new applicants", "rediscovery is 4-5x faster and $3,000 cheaper" and "30-35% of qualified candidates are invisible to keyword search". We tried to trace each to an original study and couldn't; they circulate between vendor glossaries without links. The one solid figure in this family is Gem's: in its 2026 benchmarks, 46% of sourced hires at its in-house customers were people already in the company's ATS or CRM, up from 26% in 2021. That's corporate hiring, not agencies, and it covers sourced hires only.

My view. Most agencies don't have a sourcing problem; they have a memory problem. Every placement leaves a couple of hundred assessed, contactable people in the database (Recruiterflow's benchmark counts about 213 candidates worked per placement), and every call adds context that nobody can find six months later. The agencies that win in a slow market aren't the ones with the biggest LinkedIn budget. They're the ones that can answer two questions fast: who do we already know that fits, and who among them is ready now. The rest of this guide is about making both answers reliable. (Artem Pravda)

Seven reasons your ATS can't find people you already have

The strongest evidence here comes from the vendors themselves. Their help centres document what their search covers, and those pages describe the limits plainly.

1. Search doesn't look where the knowledge is

What a recruiter learns on a call ("did three years of IFRS 17 at an insurer, wants hybrid, open to contract") usually lives in a note, not in a CV from 2019. Default keyword search often doesn't read notes:

Two consequences follow. A candidate whose best evidence of fit sits in a note won't come back from a keyword search. And when a newer, shorter CV is uploaded, a system that reads only the latest CV (Firefish's help pages say it searches the most recent one) stops seeing the experience in the older one.

2. The words don't match

Keyword search finds the string you typed. People describe the same job in many ways: a 2021 study of job titles found 30,926 distinct English vacancy titles for 2,675 occupations, about 12 titles per occupation, and the US O*NET database lists a median of 36 alternate titles per occupation (91 for software developers). Some ATSs help at the edges. Firefish adds "close variations" of keywords, and Vincere offers fuzzy matching for typos and British versus American spellings. None of the help pages we read documents true synonym expansion in core search, so recall depends on the recruiter remembering every variant. The same goes for skills tags: "C#", "C Sharp" and ".NET developer" are three different tags unless someone standardises them.

3. Filters drop people silently

Filters on fields that are often empty don't narrow your results; they exclude everyone with a blank. Bullhorn's support team notes that radius search only returns records with a postcode. Vincere's help centre says candidates without geotagged locations, "common with LinkedIn imports", don't appear in radius searches. Archived records are hidden from default search in Bullhorn and Firefish. You see a short list and conclude the talent isn't there.

4. Parsing broke the record

A CV parser turns a document into searchable fields, and some documents defeat it. Loxo's help centre warns that graphics, tables, columns, headers and footers can parse poorly, and that a scanned PDF "will be an image instead of text", so nothing in it can be searched. A 2025 preprint by Alibaba-affiliated researchers, based on Chinese résumés, estimated that about 20% of CVs use multi-column layouts. A Bullhorn reviewer mentions parsed data landing in the wrong place, and a Vincere reviewer says parsing still needs manual editing. Skills, the field search relies on most, are the hardest to extract accurately. No parser vendor publishes independent accuracy figures, so the useful question isn't "how accurate is our parser" but "how many of our records have no searchable CV text, no postcode or no parsed title".

5. One person, several records

Duplicate checks match on narrow signals. Bullhorn's compares surname plus the first two letters of the first name, or the primary email, and depending on permissions it may only check records you own or your department owns. Loxo creates a new record every time a CV is forwarded in. Candidates change work email when they change jobs and keep several personal addresses. A Bullhorn partner estimates 10-15% of records are duplicates. Duplicates split the evidence: the note is on one record, the latest CV on another, so neither scores well in search.

6. The record describes who they were

A record's searchable text is frozen at the last CV upload. US median tenure is 4.1 years, and 20.6% of workers have been in their job 12 months or less. UK job-to-job moves run at about 9% of workers a year, and data company Lusha measured about 12% a year for US sales leaders and 21% for UK sales staff. On those rates, roughly 40% (38% on the UK-wide rate) to 70% of the people in a five-year-old database no longer hold the job their CV describes, and since agency registrants were job-hunting when they joined, probably more. A search on current title or employer misses them. The job on the record usually goes stale before the contact details, because candidates tend to register with a personal email and mobile.

7. Habit and ownership

"My candidates" silos are partly built into the software: Bullhorn's duplicate check may be limited to records a user owns, so consultants create new records rather than find a colleague's. Records owned by people who have left, or by an integration user, drop out of anyone's owner-filtered views. Then there's time pressure. Posting an advert takes minutes and returns fresh, parsed applicants; mining the database takes skill and returns noisy, stale results. In Atlas's 2026 survey of agency recruiters (vendor), 37% named too much manual work, such as updating CRM records, as their main barrier. Once a few searches return junk, recruiters stop trusting the database and go to LinkedIn first, and the database decays further. And there's money: nobody searches the database first if placing a colleague's candidate means an awkward fee split. Agree a split rule for placements from another consultant's record before you mandate database-first, or the silos win.

The 12-test findability audit

Run these in your own ATS. Most take a few minutes each, and together they give you your own numbers instead of industry averages. Tests marked "documented" follow from vendor documentation of how search works; the others are our proposals.

There's no published benchmark for "good" on most of these. Run them once, write down the numbers, and run them again after each fix.

Get more from the search you already have

Before you buy anything, squeeze the search you have. Most fixes are habits and standards, not software:

• Write synonyms into the search. Chain variants with OR ("bookkeeper" OR "accounts assistant" OR "purchase ledger"), use wildcards for stems (account*), and put phrases in quotes. Check whether yours supports proximity (NEAR) operators; not all do.

• Include notes and archived records deliberately. If notes are a separate criterion or switched off by default, add them (in Bullhorn, that's the note fields under Additional Criteria); rerun important searches with archived records included.

• Save searches, and use alerts where you have them. Many agency ATSs let you save a search, and some will alert you to new matches (in Bullhorn, these "Followed Searches" are part of the paid Search & Match add-on). Set one per core skill area, so the database works for you between jobs.

• Keep hotlists. A shortlist per skill area (tearsheets in Bullhorn) of people you'd call first, reviewed weekly. A 15-minute hotlist slot in the weekly meeting works well.

• Standardise the call note. A short header at the top of every candidate note (salary, notice period, location, what would make them move, and a next-move date such as "after bonus in March") turns free text into something you can find and act on.

• Control the skills tags. One agreed tag per skill, with variants mapped to it, and a mandatory postcode at registration so radius searches work.

Which past candidates are ready to move: 10 signals ranked by evidence

The 2026 market makes timing harder to read. Many people want to move but few actually do: in Gallup's late-2025 survey, 51% of US employees were actively looking or watching for a job, but only 28% said it was a good time to find a quality one, and the US monthly quits rate was 1.9% in August 2026. That leaves a big pool of people who'd move for the right trigger. The question is which signals identify the trigger.

Two large studies that combine decades of turnover research (Griffeth and colleagues, 2000; Rubenstein and colleagues, 2017) agree: what people say and do about leaving predicts it far better than tenure, pay or age. After that come events at the employer, which change the odds for a whole group of people at once. In order of evidence, the signals are: what candidates tell you, a contractor's end date, layoffs at their employer, a manager leaving, colleagues leaving, an acquisition, a return-to-office mandate, tenure milestones, your own past placements and silver medallists, and profile changes.

1. They tell you: a "yes" to a check-in, an application to one of your jobs, a refreshed CV, Open to Work, or a date ("after my bonus in March").

• Evidence: strong. Intentions and job search are the best predictors in both large studies

• How to spot it: check-in emails or texts, inbound applications matched to existing records, job-board CV refresh dates, LinkedIn Recruiter, a next-move date logged in the note

• Watch out for: "Watchers" who never move; old Open to Work flags

2. A contractor's assignment is ending.

• Evidence: near-certain availability on a known date

• How to spot it: the placement end date already in your ATS

• Watch out for: extensions, conversion to permanent

3. Layoffs at their employer, including for the people who stay.

• Evidence: strong to moderate. A 1% workforce cut led to 31% more voluntary quits the next year (2008 study); Glassdoor saw current employees' job applications rise 40% in the year after a layoff (vendor data)

• How to spot it: news, US WARN notices, cross-checking the employer field across your database

• Watch out for: assuming a specific person was affected when they weren't

4. Their manager or team leader left.

• Evidence: moderate. In a 2026 analysis of 2.7 million person-months by people-analytics practitioner Luděk Stehlík, exit risk roughly doubled in the second month and stayed about 55% higher for months three to six

• How to spot it: your client contacts are often their managers; a job change on a contact flags the team they left (see job change tracking)

• Watch out for: correlation, not proof; a good replacement manager

5. Colleagues are leaving the same employer.

• Evidence: moderate to strong. Co-workers' job search predicts individual quitting (peer-reviewed, 2009)

• How to spot it: several people in your database leave employer X in a quarter

• Watch out for: planned restructures; small numbers

6. Their employer was acquired.

• Evidence: moderate. People who joined through a startup acquisition were about twice as likely to leave as comparable hires (Wharton, US Census data)

• How to spot it: deal news; employer name changes on profiles

• Watch out for: retention bonuses delay exits

7. A return-to-office mandate.

• Evidence: moderate. Departures rose 14% after mandates at S&P 500 firms, skewed to senior and skilled staff (working paper)

• How to spot it: news, client intelligence

• Watch out for: mandates are often softened

8. Tenure milestones (around 12, 24 and 36 months; the end of an equity vesting period).

• Evidence: weak to moderate, mostly vendor data

• How to spot it: start dates on refreshed profiles

• Watch out for: rounded dates; longer tenures in 2026

9. Past placements and silver medallists.

• Evidence: indirect, but they're your relationships

• How to spot it: placement start dates and finalist status in your ATS

• Watch out for: never approach someone you placed during the client's guarantee period

10. Their profile changed: new skills, a certification, a promotion.

• Evidence: a fit signal more than a timing one

• How to spot it: re-enrichment, candidate self-updates, new CVs

• Watch out for: keyword padding

One commercial watch-out for signals 3-7: if the employer is your client, approaching its staff can breach your terms of business or cost you the account. Check off-limits rules before you call.

What doesn't work as well as it's sold. "Likely to move" scores from sourcing tools publish no accuracy figures, and data provider People Data Labs tested such a signal and shelved it in 2024 after finding "low-signal results". Tenure-anniversary timing comes mostly from vendor data, much of it on startup employees. Use both to decide who to call first, never to decide who to ignore.

What not to use. Don't infer pregnancy, parental leave, health or any other protected characteristic as a "signal", and don't guess gender or ethnicity from names; the UK regulator, the ICO, criticised AI recruitment tools for exactly this in its November 2024 audit. Birthdays are out, and never infer immigration status: work authorisation is a legitimate field to ask about and record, not a signal to guess.

A simple rule of thumb (our suggestion, not a validated model):

• Act this week: they told you they're open, their next-move date has arrived, or a contractor ends within eight weeks.

• Act this month: their employer had a layoff, acquisition or leadership change in the last six months, or several colleagues have left.

• Watch: everything else, including tenure milestones.

Reach out within weeks of an employer event, not months. Turnover after a colleague leaves peaks within a few months, and laid-off people do find work: two-thirds of long-tenured US workers displaced in 2023-25 had been re-employed by January 2026, so the window closes.

The weekly rediscovery routine

Every vendor playbook and practitioner guide we read lands on the same few habits. None needs new software to start.

What it looks like when it works:

• Signature Consultants, US, used reminders to account managers as consultants neared the end of contracts and reports about 30% of contractors redeployed (Bullhorn case study).

• The Conexus Group, UK, says automated emails to lapsed candidates average a 10% response rate "even after years of no activity" (Bullhorn case study).

• Accrue Partners, US, stopped external sourcing for 30 days and reactivated 2,000 past candidates who had rated it well, contacting 500 for interviews and adding about $500,000 of placement revenue (Bullhorn partner blog).

• PAC Solutions, a 13-person agency, tags its best candidates so stay-in-touch follow-ups run automatically, and reports twice the placement rate (Recruiterflow case study).

All four are vendor-published and self-reported, and only PAC is in the 5-49 range, so treat them as patterns rather than benchmarks.

High-volume temp desks are different. Many light-industrial and warehouse assignments have no real end date, so rediscovery is daily availability, not an eight-week sweep. Text your pool by shift and site, rank replies by last-worked date and reliability (no-shows, do-not-return flags), and keep certifications such as forklift licences current on the record. Texting platforms and staffing back-office systems handle most of this; see our guide to back office software for temp agencies. Before texting a long-dormant US pool, check old numbers against the FCC's Reassigned Numbers Database, because a number that has changed hands means texting a stranger without consent.

What to say when you get back in touch

Expect modest reply rates and segment hard. In Noon's two 2026 surveys of about 1,000 recruiters each (vendor data), messages to previously contacted candidates got 9.9% replies against 16.9% for first contact, and the split by history was stark: 35.2% from people who had replied before, 5.2% from people who never had. Time since last contact showed no clear pattern (between 7% and 14% across the buckets, though the group contacted over a year earlier was small); whether they had replied before mattered far more. What candidates complain about isn't being contacted again; it's irrelevant messages.

A three-touch structure works for most segments:

1. A personal note that references your last conversation and gives a reason to talk now

2. A relevant role or market insight, only if it genuinely fits

3. A no-pressure close that makes "not now" easy

Example openers by segment (ours, adapt freely):

• Silver medallist: "Hi Priya, you were down to the last two for the finance lead role at a logistics client last spring, and they rated you highly. A similar role has just opened with a shorter commute for you. Worth ten minutes this week?"

• Contractor ending soon: "Hi Tom, your current contract ends on 12 December. Do you want me to line up your next one, or are you planning a break? A quick yes or no is enough."

• Dormant registrant: "Hi Sam, you registered with us in 2023 when you were looking at senior bookkeeping roles, and we've kept your details since. Is that still what you're after, or has it changed? If you'd rather not hear from us, reply 'stop' and we'll remove you. How we use your data: [privacy link]."

• After an employer event: keep it neutral. "Hi Dan, it's been a while. I'm working on two finance briefs this month that look close to your background. If the timing's right, happy to share details; if not, no problem." Don't say "I saw your company is cutting jobs"; it feels intrusive.

• Temp availability by text (US): "Hi Maria, it's Jess at [Agency]. Warehouse days next week in Joliet, $19/hr. Available? Reply YES or NO. Reply STOP to opt out."

Before you put anyone forward, get their fresh agreement to that specific introduction, confirm their CV is current and check your submission history with that client. In the UK, the Conduct Regulations bar passing on a work-seeker's information without their consent (regulation 28) and require you to give them details of a role when you offer it (regulation 21); in the US, VMS programmes typically need a right to represent for each submission. A two-year-old record should never go to a client cold.

In the UK, each message should say how you have their details, link your privacy notice and give an easy way to opt out or ask for deletion. The rules section explains why.

The rules: UK, EU and US

This is the part most rediscovery guides skip, and in the UK it changed in 2026.

UK.

• Lawful basis. Keeping candidates on file and contacting them about new roles normally rests on legitimate interests, with a written legitimate interests assessment (LIA). The "recognised legitimate interests" added by the Data (Use and Access) Act 2025 don't cover recruitment. The ICO's November 2024 audit found many recruitment AI and sourcing providers relying on legitimate interests without an LIA.

• Sourced candidates. If you took someone's details from LinkedIn or a CV database, UK GDPR requires privacy information within a reasonable period and at most one month, or at your first message if that comes sooner; the ICO's draft recruitment guidance says the same. Re-enriching an old profile from a data vendor is a fresh collection.

• Treat unsolicited role emails and DMs as marketing. The ICO's draft guidance says direct messages to candidates you've identified are direct marketing. Updates on an application, or alerts a candidate signed up for, are arguably service messages, but a "here's a role you might like" email is safest treated as marketing. For marketing emails and texts to individuals (personal addresses, and sole traders even on work addresses), PECR requires consent or a valid "soft opt-in". The ICO has fined recruitment businesses for this: Join the Triboo £130,000 in 2023 for 107 million spam emails to jobseekers, and Reed Online £40,000 in 2022 for emailing its jobseeker database about a CV-review service, including millions who had opted out. Reed had called it a service email; the ICO said intent doesn't matter. Whether the soft opt-in covers candidates who registered with you is unsettled, and it doesn't cover sourced candidates. The safest route is an explicit "keep me updated about suitable roles" opt-in at registration, an opt-out in every message, and instant suppression of anyone who objects. Since 5 February 2026, PECR fines can reach £17.5 million.

• Retention. The Conduct Regulations require records to be kept at least one year after you last provided services to a work-seeker; UK GDPR caps it at "no longer than necessary". The ICO says indefinite retention is unlikely to comply, and criticised a provider whose one-year clock reset every time a profile was refreshed. Reset the clock only when the candidate genuinely engages.

• AI ranking. Since 5 February 2026, fully automated significant decisions are allowed under legitimate interests, but only with safeguards: tell candidates, let them make representations and challenge the outcome, and offer human review. The ICO's March 2026 report on automated decisions in recruitment found many tools employers thought only "supported" decisions were in practice making them, and a recruiter who rubber-stamps a ranking doesn't count as human involvement. Let a consultant genuinely review matches and override them, say in your privacy notice that you use automated matching, and keep special category data out of it.

EU. AI used to filter or rank candidates is "high-risk" under the EU AI Act. The 2026 Digital Omnibus moved those obligations from August 2026 to 2 December 2027. UK and US agencies placing into EU roles can be in scope. Don't count on the "preparatory task" exception: systems that rank or match candidates profile them, and profiling systems in this category are always high-risk. Keep a person genuinely deciding, and ask each vendor how it classifies its tool.

US.

• California: the Civil Rights Council's rules, in force since 1 October 2025, treat automated decision systems that disadvantage protected groups as discrimination, cover employment agencies, and require their records to be kept for four years. Separately, applicant data has been covered by the CCPA since January 2023 for businesses over its thresholds (around $26.6 million revenue in 2025, among others), and new rules for automated tools that replace or substantially replace human hiring decisions apply from 1 January 2027.

• New York City: Local Law 144 requires a bias audit and notice when an automated tool substantially assists or replaces decisions about applicants, and a December 2025 state audit pushed the city towards tougher enforcement.

• Illinois: amendments in force since 1 January 2026 bar AI with a discriminatory effect in recruitment, ban zip codes as a proxy, and require notice of AI use.

• Colorado: its replacement AI law, SB 26-189, applies to decisions from 1 January 2027, though enforcement is tangled in litigation; check its status before relying on it.

• Texting and email: a few federal district court decisions (for example Gerrard v. Acara Solutions, 2020) have held job texts aren't advertisements, and the federal autodialer rules have been narrower since 2021. But several states, such as Florida and Oklahoma, and the mobile carriers effectively require consent for bulk texts. Capture consent at application, register your business texting number (10DLC or toll-free verification), honour STOP and other reasonable opt-outs within ten business days, respect state quiet hours, and check old numbers against the FCC Reassigned Numbers Database. Bulk emails should follow CAN-SPAM formalities.

Before your first campaign, check:

1. Where each record came from and when the person last engaged

2. A current LIA covering storage, search, enrichment, matching and re-contact (UK)

3. A retention clean-up first, with a suppression list for people who opted out

4. A privacy notice that mentions future roles, automated matching and retention periods

5. Privacy information in the first message to anyone sourced

6. Consent or opt-out position by channel: personal email, work email, SMS, LinkedIn

7. A consultant genuinely reviewing every ranked list

8. A data-protection complaints procedure, required under the Data (Use and Access) Act since 19 June 2026 (UK)

Candidate rediscovery tools: what fixes which cause

Semantic search over your own database is now standard in many agency CRMs. Job-change and "open to work" alerts, and anything that reads notes and calls, are still patchier and often cost more. Match the tool to the causes your audit found.

On price:

Standalone rediscovery platforms such as hireEZ, SeekOut and Eightfold mainly connect to corporate systems like Workday, Greenhouse and Lever, often with seat minimums. Gem sells staffing plans, but its rediscovery needs a separately priced AI seat. For most small agencies, the AI tier of their own ATS or an add-on is the practical route.

Test before you buy. Run audit tests 1-3 on a trial with your own data: a known placement, a fact that lives only in a note, and a skill phrased three ways. If the tool finds them, it's real; if it only returns exact keyword matches, you're paying AI prices for Boolean search.

A scorecard

No independent benchmarks exist for most of these, so measure against your own baseline.

An honest note on Execue

This guide is written by Execue's CPO, so here's the straight version.

What Execue is: an AI sales employee for recruitment agencies. Its agent, Q, watches two things: your own history (placements, candidates and client contacts) and the market (leadership changes, acquisitions, funding rounds, new hires and roles posted). Every morning it gives you a short list of who to contact and why, with a drafted message for each. Once you approve, it sends from your own email and LinkedIn, handles the reply (which you also approve), books the meeting and chases no-shows.

Where it meets rediscovery: one event, two leads. The same move in your history is usually both a business-development signal and a rediscovery signal. An illustrative example:

• In 2023 you placed Sarah as finance manager at a logistics firm. Q sees she has just started as finance director somewhere new.

• Lead one, a client: a new leader often rebuilds her team. Q puts her on the morning list with a drafted note that mentions the placement.

• Lead two, candidates: the team she left is in your database, and signal 4 says they're now more likely to move. That's your cue for a neutral check-in with them, run through your own routine and approved by you.

What a morning list might look like (illustrative; names invented):

• Sarah P., placed by you in 2023, has just started as finance director at a new company. Why now: a new leader in her first months. Draft ready: congratulations, and an offer to help build her team.

• A logistics firm in your niche has been acquired. Why now: acquired staff are more likely to leave, and the acquirer is likely to restructure. Your next step: check who in your database works there, and run a neutral check-in through your own routine.

• A healthcare group you've placed with has a new head of operations. Why now: new leaders often bring in their own people. Draft ready: an introduction that mentions your recent operations placements.

Candidate search (in beta). You describe the role in plain language, and Q searches your own database and the wider market, which it monitors continuously, for that profile. It reads the notes in your ATS as well as CVs, so what a consultant learned on a call counts, and it surfaces people you've already placed, interviewed or mapped alongside new names. It's at its strongest on hard, specific roles where keyword search fails. That answers causes 1 and 2: the notes are searched, and you describe the person you want instead of guessing the keywords on their CV. Anyone Q finds in the wider market is a sourced candidate, so the privacy-information rules above apply, and a consultant reviews every list before anyone is contacted.

Who it fits: agencies of 5-49 people on Bullhorn, Vincere, JobAdder, Firefish, Recruit CRM, Loxo, Recruiterflow or Crelate; Execue's AI configures the connection during setup.

What it costs: Core at $254 per seat per month and Scale at $509, billed annually in US dollars, with custom pricing for enterprise.

Where it isn't the answer: it isn't your ATS. It doesn't merge duplicates, fix parsing, set your retention policy or run your contractor-end-date sweeps and temp availability texts; those routines still live in your ATS and your team's week. It isn't a bulk sender: every message is approved by a person and goes out from your own accounts. On a high-volume temp desk, Execue helps you win the clients, while daily availability and redeployment stay in your ATS and texting tool. And if your audit shows a database full of duplicates and blank postcodes, fix that first; better search on bad data still returns bad results.

Where to start

This week (one hour): run audit tests 1, 2, 5 and 9. If fewer than half of your known placements come back on page one, or notes aren't searched, you've found the cause, and the fix starts with the search habits above, not a new tool.

This month: make database-first a rule on every new job, start the weekly contractor-end-date sweep, and set up one always-on check-in for candidates dormant for six months, with your privacy wording in it. Turn on source tracking for placements.

This quarter: run the hygiene pass, refresh your LIA and retention policy (UK), start a monthly review of employer events where you hold candidates, and decide whether your ATS's search, an add-on or a different tool fixes the causes your audit found.

If the bottleneck is search that can't read your notes or meaning, or knowing which of the people in your history have just moved, that's where Execue works: Q finds them, drafts the outreach and books the meetings, with you approving each step. See how at execue.io.

FAQ

Q: What is candidate rediscovery (talent rediscovery)?

A: Candidate rediscovery, also called talent rediscovery, is finding and re-engaging people already in your ATS or CRM, such as past applicants, silver medallists, placed candidates and contractors coming off assignment, before you source externally. It's usually an agency's cheapest and fastest channel, because you already hold their CV, history and contact details.

Q: What is a silver medalist in recruitment?

A: A silver medalist (silver medallist in UK spelling) is a candidate who reached the final stage for a role but wasn't hired. They've already been assessed and liked by a client, so they're among the first people to approach when a similar role opens, as long as you check they're still interested and get their agreement before putting them forward.

Q: Why can't my ATS search my own candidate database properly?

A: Usually one of seven causes: keyword search doesn't cover notes or older CVs, the words on the CV don't match your search terms, filters exclude records with blank fields, the CV parsed badly, the person has duplicate records, the record is out of date, or habit and ownership keep candidates in other consultants' silos. A 12-test audit shows which apply to you.

Q: Does Bullhorn search notes?

A: Not in the standard keyword box: Bullhorn's knowledge base says the keywords field doesn't search notes, which need a separate criterion. Its AI search and match features sit in the Pro tier or premium add-ons rather than the published Starter and Core tiers. Check your own configuration with a notes-only test.

Q: How much of an agency's business comes from its own database?

A: More than most agencies think. Recruiterflow's 2026 benchmark of 2,100+ agencies found 71% of placements went to people already in the CRM before the job opened, and published agency case studies cluster around a third to 40%. Only 13.8% of US and Canadian agency recruiters name the database as their top channel, probably because placements get credited to wherever the candidate first came from.

Q: How quickly does a candidate database go out of date?

A: The job usually goes stale before the contact details. US median tenure is 4.1 years, and annual job-change rates run from about 9% (UK workers overall) to 21% (UK sales staff), so roughly 40-70% of a five-year-old database has probably changed employer. Personal emails and mobiles tend to outlast work contacts.

Q: What are the best signals that a past candidate is open to a move?

A: What they tell you comes first: a yes to a check-in, an application or a refreshed CV. Next comes a contractor's end date, then events at their employer: layoffs, a manager leaving, colleagues leaving, an acquisition or a return-to-office mandate. Tenure anniversaries and "likely to move" scores are weaker; use them only to decide who to call first.

Q: What reply rate should I expect when re-contacting old candidates?

A: Roughly 5-15%. A 2026 vendor survey found about 10% replies to re-engagement overall, 35% from people who had replied before and 5% from people who never had. A UK agency reports about 10% from lapsed candidates. Relevance matters more than the template.

Q: Is it legal to email candidates in my database in the UK?

A: Usually, if you do it properly: legitimate interests with a written assessment, privacy information for anyone you sourced, an opt-out in every message, a retention policy you actually apply and, for marketing emails or texts to personal addresses, consent or a defensible soft opt-in under PECR. Whether the soft opt-in covers registered candidates is unsettled, so an explicit opt-in at registration is safest. Take legal advice.

Q: How long can a UK agency keep candidate data?

A: The Conduct Regulations require at least one year after you last provided services; UK GDPR says no longer than necessary, and the ICO says indefinite retention is unlikely to comply. A common approach, similar to what APSCo suggests, is a "do you still want to hear from us?" message after 12-24 months of silence, then deletion or anonymisation if there's no reply.

Q: Do AI matching tools create legal risk?

A: They can if the score effectively decides. In the UK, a recruiter rubber-stamping a ranking isn't meaningful human involvement. In the EU, obligations for high-risk recruitment AI, including ranking, apply from December 2027. In the US, New York City, Illinois and California regulate automated hiring tools, and Colorado will from 2027. Keep a person genuinely deciding and tell candidates you use automated matching.

Q: What's the difference between candidate rediscovery and talent pooling?

A: A talent pool is a group of candidates you've tagged to stay in touch with, usually by skill or role. Rediscovery is the wider habit of searching your whole database first on every job and re-engaging whoever fits and is ready, including people you never put in a pool.

Q: What are the best candidate rediscovery tools for a small agency?

A: Usually the AI tier of the ATS you already use, or an add-on for it, rather than an enterprise platform. Loxo includes natural-language search from $149 a user a month; Recruiterflow, Recruit CRM, Zoho Recruit, JobAdder, Firefish and Mercury offer AI search or matching; Bullhorn users need Pro or an add-on such as Skima, Daxtra or Textkernel. Test any of them on your own data first.

Related Reading

• Sourcing From Your Own ATS: Matching Candidates to a Job Description

• Job Change Tracking for Recruiters

• Sourcing Automation: The Complete 2026 Playbook

• Best Recruitment Software for Small Agencies

• Back Office Recruitment Software for Temp and Contract Agencies

• How to Get Clients for a Recruitment Agency

Written by Artem Pravda (CPO & CDO, Execue), drawing on vendor help-centre documentation from Bullhorn, Vincere, Recruit CRM, Recruiterflow, Firefish, Crelate and Loxo; vendor pricing and product pages from Bullhorn, Loxo, Recruiterflow, Firefish and Gem, and third-party pricing reports; Kyloe's duplicate estimate (via Bullhorn); agency case studies published by Tracker, JobAdder, Bullhorn, Daxtra and Recruiterflow; Recruiterflow's 2026 recruiting benchmarks, Top Echelon's 2025 State of the Industry report, Gem's 2025 and 2026 recruiting benchmarks, Atlas's 2026 agency report and Noon's 2026 re-engagement surveys; the US Bureau of Labor Statistics (employee tenure, January 2026; JOLTS, August 2026; displaced workers); Gallup (Q4 2025); Firefish's 2026 agency report; StaffingHub's 2026 benchmark; Lusha's measured job-change rates; meta-analyses by Griffeth, Hom and Gaertner (2000) and Rubenstein, Eberly, Lee and Mitchell (2017); Trevor and Nyberg (2008), Felps and colleagues (2009), Kim (Wharton, 2019), Luděk Stehlík (2026) and the University of Pittsburgh return-to-office study; Glassdoor Economic Research (2025); People Data Labs; academic work on job titles (JobBERT, 2021) and CV parsing (2025); O*NET via Noon; the ICO (AI in recruitment audit, November 2024; draft recruitment guidance; automated decision-making in recruitment report, March 2026; Reed Online penalty notice, 2022); legislation.gov.uk (Conduct Regulations 2003, regulations 21, 28 and 29; UK GDPR; SI 2026/82); the ICO's Join the Triboo penalty (2023); California Civil Rights Council regulations; FCC consent-revocation rules and TCPA case law; law-firm briefings from A&O Shearman, RPC, Covington, DLA Piper, Browne Jacobson, K&L Gates, FPF, Mayer Brown, Hunton, Littler, Akin Gump, Seyfarth, Hinshaw, Squire Patton Boggs and Womble Bond Dickinson; and the New York State Comptroller. Most case-study and benchmark figures are published by vendors and self-reported; the 40-70% database-staleness range, the signal ranking, the rule of thumb, the example messages and the audit tests not marked as documented are our own estimates and suggestions. Laws on data, AI and marketing are changing quickly: check current guidance and take legal advice before running a campaign.