Lookalike Client Modelling: Turn Your ATS Placement History into Your Next 100 Target Clients (2026)

TL;DR

Most agency ICPs are written from opinion: the owner's instinct, a sector the desk likes, a slide from last January's planning day. Meanwhile the most honest ICP document you own sits unused in your ATS: every placement, fee, start date and repeat order your agency has ever made. Lookalike client modelling borrows a method from B2B sales: take the accounts that already bought, filter them down to the genuinely good ones, find what they share, and search for more companies that look like them [1] [2]. In recruitment the "seed" is not a closed-won deal. It is a client that paid a fair fee, briefed you again and let you fill roles without a fight. This guide gives you the method in seven steps: export the data from Bullhorn, Recruiterflow or any ATS, score clients on value and repeat rate, pull out the shared traits, write a lookalike ICP, build a list of 100 target companies, rank them, and feed them into signal-led BD so you reach them when they are actually hiring. A spreadsheet is enough to start. boilr then takes the ICP and runs the signal watching, scoring and drafting, with the Company Brain keeping the evidence in one place.

Why Your Placement History Beats a Workshop ICP

If you have not written an ICP yet, start with our guide to defining an ICP for recruitment agencies. This article is the data-led version. It answers a narrower question: what do the clients who already made you money have in common, and where are the 100 companies most like them?

• It is behaviour, not opinion. A placement record shows who actually briefed you, paid and came back. A workshop shows what the room believes.
• It exposes the gap between "big" and "good". The biggest logo on the list is often not the best client once fee level, time-to-fill and discounts are counted.
• It surfaces patterns nobody wrote down. Maybe your best clients are all founder-led, or all hire in clusters of three, or all use the same tech stack. Consultants feel this; the ATS can prove it.
• It survives consultant churn. When the person who "just knew" the right clients leaves, the data is still there. See our piece on why consultant churn drains the agency's knowledge.

A caveat before the method: client concentration is a real risk, and most agency revenue rests on a small group of accounts. Practitioner sources describe the 80/20 pattern in staffing firms, but it is a rule of thumb, not a measured benchmark [9], and the thresholds people quote for "too much from one client" vary by source [7]. Lookalike modelling is partly a way to widen that base with clients of the same quality, not to clone one account until it becomes a dependency.

What "Lookalike" Means in Recruitment BD

In B2B sales, lookalike modelling means using your best existing customers as the seed, then scoring other companies by how closely they match, usually on firmographics first and other layers (technology, behaviour, timing) on top [2]. The vendor guides on this are written for software sellers, and their numeric rules of thumb are not validated studies [1]. The logic still transfers, but you have to translate three things:

B2B software sales	Recruitment BD equivalent
Closed-won deal	A client with a placed role, a paid invoice and a fee you are happy with
Deal size / ACV	Fee per placement, total billings per client, margin after rebates and discounts
Churn	Client goes quiet after one placement, or asks you to drop to a PSL rate
Expansion	Repeat briefs, a second desk or office using you, exclusive or retained work
Product usage depth	Time-to-fill, shortlist-to-interview ratio, fall-off rate
Buying committee	The hiring manager, the TA lead, the founder or CHRO who signs the terms

The big recruitment-specific difference is timing. A software buyer might fit your profile and buy in any quarter. A recruitment client only buys when a role opens. That is why the lookalike list is the who and signals are the when. You need both, and we come back to that in step seven.

Step 1: Export the Data You Already Have

You do not need a data team. You need one export per ATS and an afternoon. The fields below are what matter; most systems hold them, though the exact report names differ.

The fields to pull

• Client company - the account name, plus domain if your ATS stores it (it makes later matching far easier).
• Placement date and type - permanent, contract or temp, and the start date.
• Fee or revenue - flat fee or margin per placement, ideally after any discount or rebate.
• Job created date - so you can calculate time-to-fill.
• Role family and seniority - the job title or category.
• Source of the client - inbound, referral, outbound, signal, existing relationship.
• Fall-off or replacement flag - where you track guarantee claims.
• Firmographics - industry, headcount, location, funding stage (usually not in the ATS, so add these in step 3).

Where to find it in common ATS tools

ATS	Where to start	Watch out for
Bullhorn	The Placement Activity Report shows permanent placements with salary and flat fee, and contract placements with pay and bill rates [3]. The dashboard defines Days to Fill as the days between a job's added date and its first placement [4].	The report only includes placements that have the fee or rate fields filled in, so incomplete records are missing from totals [3]. Wide exports can also time out, so export in slices.
Recruiterflow	The Revenue Report breaks down revenue from filled placements and can be grouped by client company [5]. The Placements tab includes a CSV export for external analysis [6].	Placements without a fee value do not appear in the revenue report [5]. Check which columns your CSV actually contains before you plan around it.
Any other ATS or CRM	A placements or "starts" list with client, fee and dates, exported to CSV or Excel.	Duplicate client names ("Acme Ltd", "ACME Limited") will split one client into several rows. Clean them first.

Report names and navigation change between releases, so treat the above as a starting point and confirm in your own account. The data-quality point matters more than the tool: placements with no fee recorded vanish from every revenue view. If your consultants skip that field, fix it as part of this project, because every later step depends on it.

Step 2: Score Every Client on Value and Repeat Rate

Revenue alone is a poor definition of "best". A client that gave you one huge placement and then ghosted you is not the same as a company that gives you three mid-sized fees a year without negotiation. Build a simple client table with one row per company and these columns:

• Total billings (last 24-36 months) - the sum of fees or margin.
• Average fee as a percentage of salary - the clearest read on price sensitivity.
• Number of placements - repeat business is the main signal of a real relationship.
• Months between first and last placement - shows whether they came back over time.
• Median time-to-fill - slow clients cost desk hours.
• Fall-off rate - how often placements collapsed within the guarantee period.
• Share of your total billings - to see concentration. Recruiter trade writing lists revenue concentration and repeat business rate as core client metrics [8].

A simple weighted score

Keep it boring. Rank each column from 1 to 5 across your client base, then weight them. One workable starting split, which you should adjust to your own economics:

Factor	Suggested weight	Why
Fee level (percentage held, not discounted)	25%	Protects margin
Repeat rate (placements per year active)	30%	Proof of a relationship, not a lucky brief
Total billings	20%	Size still counts, just not alone
Time-to-fill (faster is better)	15%	Desk hours are your real cost
Low fall-off	10%	Rebates quietly erase fees

The weights are a judgement call, not a research result. What matters is that you write them down once, apply them to every client the same way, and review them later. Our guide to lead scoring for recruitment agencies covers the same logic applied to prospects.

Step 3: Choose the Seed Group and Clean It

The seed group is the set of clients you want more of. Pick the top slice of your scored table, but filter hard. Lookalike guides from the B2B software world warn that a polluted seed list means the model faithfully finds you more of the wrong thing [1].

Include

• Clients with at least two placements, or one placement plus a clear repeat brief in progress.
• Fee held at or near your standard rate.
• Fall-off below your agency average.
• Active in the last 12-18 months, so the profile reflects today's market.

Exclude or tag separately

• One-off emergency fills - a company that hired you once under pressure and never again.
• Heavily discounted PSL accounts - they may be profitable, but they are a different segment with different economics.
• Clients won through a personal relationship that cannot be repeated (a former colleague who moved companies).
• Companies that no longer exist or have been acquired - their traits describe a past situation.

How big should the seed group be? One vendor guide argues that below roughly 30-50 qualifying accounts in the trailing 6-12 months, lookalike modelling is closer to guessing [1]. That threshold is a rule of thumb from a sales-tool vendor, and a boutique agency may only have 15-25 good clients in total. That does not make the exercise useless. It means you should treat the output as a set of hypotheses to test, not a model to trust, and widen the seed over time as your data grows.

Step 4: Find the Shared Traits

Now add the firmographics the ATS lacks. For each seed client, fill in a handful of columns from a public source: LinkedIn, Companies House or the local register, and the company website. Do the same for a comparison group: clients who were low value, discounted or one-off. The comparison group is what turns a description into a discriminator. If 80% of your best clients are Series B and 70% of your worst are too, "Series B" is not a trait.

Trait categories to check

Layer	Questions to ask of your best clients	Recruitment example
Firmographic	Industry, headcount band, geography, ownership, funding stage?	Software firms with 80-300 staff in the UK, PE-backed or Series B
Hiring pattern	How many roles per year? Single hires or clusters? Which seniority?	Hire in clusters of three or more engineers after a funding round
Buyer	Who briefed you? Founder, hiring manager, TA lead, HR director?	Head of Engineering briefs directly, no formal PSL
Internal capacity	Do they have an in-house recruiter? A TA team of how many?	One overstretched in-house recruiter, no TA team
Commercial	Terms accepted, exclusivity, payment speed, rebate disputes?	Accept 20% on contingent, pay within 30 days
Origin	How did the relationship start? What was the trigger?	Started after a new CTO joined, first brief within 60 days
Tech or domain	Stack, regulatory regime, product type?	Fintech regulated by the FCA, cloud-native stack

The "Origin" row is the one most agencies skip, and it is the most valuable for signal-led BD. If five of your best clients first briefed you within two months of a new senior hire, that is not just a trait, it is a trigger you can watch for. See our guide to executive moves as a recruitment BD signal.

Do it by hand first

For 20 to 40 clients you do not need a model. Put the seed and comparison rows side by side and look for the three or four columns where the groups clearly differ. Ask the consultants who own those accounts to sanity check: they will quickly say "that is true" or "that is coincidence". Their judgement plus the table beats either alone.

Step 5: Write the Lookalike ICP

Translate the traits into a short, testable definition. Keep it to what discriminates, in plain language that a consultant (or an AI agent) can apply without interpretation. Use the same hypothesis style we recommend in our quarterly ICP refresh guide.

A worked example (illustrative, not real data)

Imagine a six-person technology desk exports three years of placements and finds that its best clients share four traits. The lookalike ICP might read:

• Who - UK software and fintech companies with 60-250 employees, venture or PE backed.
• Situation - hiring three or more engineers in a rolling quarter, with no more than one in-house recruiter.
• Buyer - a Head of Engineering or CTO who briefs directly.
• Trigger - funding round, new senior engineering hire or an open-roles spike in the last 60 days.
• Disqualifiers - formal MSP or VMS programme, standing fee cap below your floor, or a PSL that is closed to new agencies.

Every line has a way to be wrong, which is the point. Add a date and the evidence next to each ("8 of 11 seed clients; 2 of 9 comparison clients"). Disqualifiers deserve as much care as the positives; our piece on negative signals that disqualify clients shows how to keep them honest.

Step 6: Build the List of 100 and Rank It

A lookalike ICP is only useful if it becomes a list. Aim for roughly 100 companies per desk to start. That is small enough for a consultant to know, and large enough to give you a steady flow of triggers across a quarter. (Some vendor guides recommend far bigger outbound lists, in the thousands [1], but a recruitment desk works accounts one consultant at a time and a shorter list gets better research.)

Where the 100 come from

• Company databases and LinkedIn filters - apply the firmographic traits (industry, headcount, geography, funding).
• Lookalikes of named seeds - competitors, suppliers and investor portfolio companies of your best clients.
• Your own dormant accounts - clients that matched the profile but went quiet. They are warmer than any stranger.
• Candidate-side intelligence - companies your placed candidates came from or moved to.
• Referrals - ask the top five seed clients who else in their network struggles to hire.

Rank before you reach out

Score each of the 100 against the ICP the same way you scored clients. Put dormant accounts and strong-trait matches in tier A, partial matches in tier B, and speculative fits in tier C. Check every company against your do-not-contact and existing-client lists first. If you contact people in the UK or EU, make sure your outreach has a documented lawful basis; our guide to GDPR and legitimate interest in signal-led prospecting covers the practical side.

Tier	Definition	Approach
A	Matches 4+ traits, or a dormant former client	Personal research, senior consultant, act on first signal
B	Matches 2-3 traits	Monitor signals, draft-assisted outreach on trigger
C	Matches 1-2 traits, high uncertainty	Light-touch watch list, small test volume, review after 90 days

Step 7: Feed the List into Signal-Led BD

A perfect list contacted at the wrong time still gets ignored. The 100 tell you who deserves your attention. Hiring signals tell you which of them to contact this week: a funding round, a new senior hire, a jump in job postings, an office expansion. Layer the two and a consultant stops cold-calling a list and starts acting on moments. Our guide to signal stacking shows how to combine triggers into a stronger case for reaching out.

Approach	Who you contact	When you contact them	Typical outcome
Cold list	Any company in the sector	Whenever the consultant has time	Low relevance, low reply
Lookalike list only	Companies that resemble your best clients	Whenever the consultant has time	Better targets, still no timing
Signals only	Any company showing a trigger	When the trigger fires	Good timing, uneven fit
Lookalike plus signals	Companies that fit and are hiring now	Within days of the trigger	Best use of consultant time

The last row is the model to aim for. The rows above it describe outcomes qualitatively; we are not claiming measured conversion figures for them, so run your own comparison over a quarter (see the KPIs below).

How boilr Fits Into This Workflow

An honest framing first. boilr does not currently read your placement history out of the ATS and build the lookalike model for you. The analysis in steps 1 to 5 is yours, done in a spreadsheet. What boilr does is take the result and run the daily work that normally breaks down: watching the companies, spotting the signals, scoring against your ICP and drafting the outreach.

• Agent - you tell your boilr agent your ICP, target industries and open roles, and you can update the ICP at any time. Paste in the lookalike ICP from step 5.
• Companies - the account list your ICP produces, so the 100 are researched and enriched rather than sitting in a spreadsheet.
• Signals - monitoring for funding rounds, leadership changes, hiring velocity, headcount growth and office expansion across your target companies and sectors.
• ICP scoring - signals and companies are scored against your ICP, so tier A accounts with a fresh trigger rise to the top.
• Tasks - finished work arrives in your Tasks inbox with context, a suggested angle and a ready-to-send message. You edit, verify and send.
• Company Brain - learns from every outreach you verify and send, capturing which messages and ICPs get replies and bookings, so the agency's evidence sits in one shared place. It learns from outreach results, not from placement records, so keep your placement analysis as the starting hypothesis and let reply data refine it.
• Integrations - Recruiterflow is live (boilr pushes ICP-matched companies, contacts and signals into your ATS). HubSpot is live too and avoids duplicates by matching companies on domain. Bullhorn is on the roadmap, not live, so Bullhorn teams currently bring the export in manually.

What stays human

• Choosing the seed group - only your consultants know which clients were genuinely good.
• Reading the traits - deciding which patterns are causal and which are coincidence.
• Verifying and sending every message - nothing goes out unreviewed.
• The relationship - the first call, the brief, the fee conversation.

Got a lookalike ICP and want it working against live hiring signals? Book a demo or start with boilr and give your agent the profile you built from your own placements.

KPIs: How to Know the Lookalike List Is Working

Compare the lookalike list against your previous prospecting over the same period. The aim is to see whether lookalike targets convert better, not to hit a benchmark someone else measured.

KPI	What it tells you	How to read it
Reply rate: lookalike vs non-lookalike	Whether fit improves response	Compare like for like, same message style and period
Meeting-to-brief rate	Whether the companies actually have mandates	Low rate means fit on paper, no urgency
Average fee percentage on new clients	Whether the ICP selects for price-sensitive buyers	Should match or beat your seed average
Time-to-first-brief	Speed from first contact to a real role	Falls when signals and fit line up
Repeat rate at 6-12 months	Whether lookalikes become real clients	The slowest metric and the most important one
Client concentration	Whether the base is widening	Watch your top client's share of billings

Common Mistakes

• Seeding with revenue alone - the biggest accounts may be the most discounted and slowest to fill.
• Missing fee data - placements with no fee recorded disappear from revenue reports, which biases the whole analysis [5].
• No comparison group - a trait shared by good and bad clients tells you nothing.
• Cloning one account - a lookalike of your single biggest client is a concentration risk in disguise [7].
• Treating it as finished - markets move. Re-run the analysis every quarter or two.
• Contacting the whole list at once - without a trigger you are back to cold outreach with a nicer spreadsheet.
• Ignoring the wins outside the model - a great client that does not fit the profile deserves investigation, not exclusion.

A 30-Day Plan

Week	Action	Output
1	Export placements, clean client names, fix missing fees	One clean client table
2	Score clients, pick the seed group and comparison group, add firmographics	Seed and comparison sheets
3	Review traits with the consultants who own the accounts, write the lookalike ICP	One-page ICP with evidence and disqualifiers
4	Build and tier the list of 100, load it into your tooling, switch on signals	Ranked target list with live triggers

• Day 30 - first outreach on tier A accounts with a fresh trigger.
• Day 60 - first read of reply rate and meeting-to-brief rate against your old approach.
• Day 90 - adjust two or three ICP criteria based on what replied.

Frequently Asked Questions

What is lookalike client modelling for a recruitment agency?

It is the practice of using your best existing clients as a template to find new ones. You export placement data, score clients on value and repeat business, identify the traits they share, then search for other companies with the same traits. It borrows the lookalike method used in B2B sales and adapts the seed to recruitment economics [2].

Do I need a data scientist or a machine learning model?

No. For most agencies a spreadsheet and a few hours with the owning consultants is enough. A statistical model needs far more accounts than a typical desk has, and vendor guides themselves caution that small samples make model output unreliable [1]. Start by hand and treat the result as a hypothesis.

Which ATS data matters most?

Fee level, number of placements per client, time between first and last placement, time-to-fill and fall-off. Bullhorn and Recruiterflow both expose placement and revenue reporting, though missing fee values will drop records from the totals [3] [5]. Fix the data entry habit first.

How many good clients do I need before this works?

More is better, but there is no proven minimum. One vendor guide suggests 30-50 qualifying accounts [1], which is a rule of thumb. With fewer, the method still helps you structure judgement, but treat every conclusion as a test and widen the seed over time.

Does boilr connect to Bullhorn to read my placement history?

Not today. Recruiterflow and HubSpot are live integrations, and Bullhorn is on the roadmap. Even where integrations exist, they push companies, contacts and signals from boilr into your system rather than analysing your placement history. You do the analysis, then give the resulting ICP to your boilr agent.

How often should I redo the lookalike analysis?

Quarterly is a sensible default, with extra reviews after a major shift such as a new market or pricing change. Our ICP decay guide covers the refresh process in detail.

What if my best client does not fit the pattern?

Keep them as a named exception and find out why they work. Sometimes the outlier points to a segment you had not considered. Do not delete it from the seed just to make the pattern cleaner.

Is lookalike targeting compliant with GDPR?

The list of companies is not the issue; the contact and outreach are. If you message individuals in the UK or EU, document your lawful basis and respect opt-outs. Read our guide on legitimate interest and check current regulator guidance, as the rules differ between corporate and individual recipients.

Sources

Information sourced from public industry reports and vendor documentation as of November 2026.

1. firstsales.io - Lookalike account sourcing: mine closed-won for pipeline
2. Landbase - How lookalike modeling works
3. Bullhorn Knowledge Base - Placement Activity Report
4. Bullhorn Knowledge Base - Recruiting Manager Dashboard (Days to Fill)
5. Recruiterflow Help Centre - Revenue Report
6. Recruiterflow Help Centre - The Placements tab
7. Top Echelon - Why client concentration is dangerous in 2026
8. Recruiter - Measuring what matters: client metrics
9. Madison Resources - The 80/20 rule for staffing firms