Your ops manager opens Bullhorn at 08:20 every morning and works through the same list: move yesterday's candidate notes from email to the record, chase missing right-to-work docs, update placement statuses, re-enrich five records that came in from LinkedIn with no phone number. By 10am she's done none of the billable work she wanted to do.
In most small UK agencies I've scoped, recruiters lose close to 40% of their day to non-selling work. Sourcing, CV parsing, reformatting CVs for client submission, manually logging call notes. You've bought the ATS. You've bought the CRM. You may have bought a "parser" that promised to fix half of this. You've still got the problem.
This is the bracket of agency I build automation for. Five to twenty-five recruiters, UK-based, running on Bullhorn or Vincere or JobAdder, bleeding non-billable hours because the automation that came with the ATS doesn't do what the sales call promised.
The short answer
Recruitment automation for a small UK agency works best when it lives outside the ATS, not inside it. The ATS is your system of record. The real automation (candidate enrichment, CV-to-submission-ready formatting, LinkedIn-to-CRM sync, interview scheduling, email triage) runs in a separate layer that reads from and writes to your ATS over its API. Expect to save 8-15 hours per week per recruiter once built properly. Expect to spend between £600/month (SaaS stack) and £15,000-£25,000 (custom build) to get there, depending on which of the three agency archetypes you hire.
Why recruitment automation is broken in most small UK agencies
Bullhorn is the UK market leader. Vincere is the challenger. JobAdder has a solid slice. All three sell "automation" as a feature. All three describe it the same way: automated workflows, AI-powered screening, intelligent matching. The demos look clean.
Then you buy it, and six months later nothing meaningful has changed.
Three things are happening.
First, the ATS automation layer is designed for corporate in-house TA, not agency workflows. The built-in flows assume one employer, one requisition pipeline, one hiring manager. A recruitment agency has dozens of live clients, hundreds of live roles, and candidates routed across multiple client pipelines simultaneously. ATS-native automation wasn't built for this fan-out and it shows the moment you try to scale a workflow past a single client.
Second, the operations manager has a soft incentive not to touch the workflows. Most small UK agencies run on a mutual understanding between the owner and the ops manager: ops keeps the data clean, the owner doesn't ask too closely how. Replacing two hours a day of manual work with a scheduled automation means the ops manager's value to the business has to be re-justified. It rarely gets fully built because nobody inside the agency is pushing for it, and the ATS vendor isn't going to build it for them.
Third, the integrations that would actually help are paywalled. Bullhorn charges per seat plus per integration plus per API call tier. Vincere and JobAdder are structured similarly. By the time you've added Apollo for contact data, Voila Norbert for email verification, DocuSign for contract signing, Calendly for scheduling, and a Zapier Professional plan to wire them together, you're £1,000-£2,500/month into your automation stack before a single line of custom code has been written.
The agencies that get this right stop fighting the ATS's built-in automation and build a thin custom layer alongside it that calls the ATS API when needed.
What's actually worth automating in a 5-25 recruiter UK agency
Every listicle on the first page of Google will tell you to automate everything. That's vendor copy. In practice, some automations compound and some don't. Here's what I've seen actually work, in rough order of ROI per workflow.
1. Candidate record enrichment at the point of ingestion
When a new candidate lands in Bullhorn (via CV submission, LinkedIn scrape, or manual entry) a workflow fetches their current employer, most recent role, LinkedIn URL, business email, and phone if missing. The recruiter opens a fully populated record instead of chasing 20 minutes of LinkedIn stalking.
Time saved: 15-25 minutes per record. Multiply by 40 new records per recruiter per week.
Build complexity: medium. Needs API keys for an enrichment provider (Apollo, Clay, Cognism) plus a workflow layer to watch the ATS and trigger the call. Watch the cost of the enrichment vendor: Cognism at around £2,000/month for an agency licence is expensive; Apollo at roughly £99/seat is cheaper; Voila Norbert at £49/month for email-only is cheapest. Match the spend to the use case.
2. CV-to-submission-ready formatting
Recruiters spend 15-30 minutes per shortlisted candidate cleaning up a CV, reformatting to the agency's client-facing template, redacting personal details, and adding the brief match summary. This is exactly what an LLM does excellently in 2026.
Time saved: 20 minutes per submission. At five submissions per recruiter per week that's 1.6 hours.
Build complexity: low. A workflow triggered on "candidate added to client pipeline" that takes the raw CV, extracts structured data, reformats into your template, and writes the new doc back as an attachment to the ATS record. Claude or GPT does the work.
3. Interview scheduling with clients
The classic three-way email chain. Candidate available Tuesday or Thursday, client available Monday or Wednesday, recruiter loses an hour coordinating. Calendly's paid tier handles one-to-one. Three-way scheduling needs custom.
Time saved: 30-45 minutes per interview booked. Five to ten interviews per week per recruiter.
Build complexity: medium. Needs calendar API access for all three sides plus a booking workflow with fallback logic.
4. Call note capture and ATS update
A recruiter finishes a 20-minute candidate call. They now spend 5-10 minutes typing up notes and pasting them into Bullhorn. A workflow that takes an audio recording (Otter, tl;dv, Fathom, or a raw voice memo), transcribes it, summarises the relevant points, and writes structured notes to the candidate record replaces that admin with a background job.
Time saved: 5-10 minutes per call. A recruiter does 10-15 calls a day.
Build complexity: low to medium. Most of the transcription tools have APIs. The wiring is the work.
5. Lead and BD signal monitoring
Your ops manager manually checks LinkedIn for funding news, leadership changes, and growth signals in your target accounts. A workflow that monitors a list of companies for news and writes summaries to a BD dashboard replaces the manual scan with scheduled runs.
Time saved: 5-8 hours per week for whoever does BD research.
Build complexity: medium. Needs news APIs or LinkedIn scraping, plus a scoring and summarisation layer.
What I wouldn't automate first
Resume screening as the primary placement decision. Using an LLM to rank candidates for a client brief is a minefield. False negatives are expensive (you miss a good candidate). False positives waste client submission slots and damage your reputation. Use LLMs to draft match summaries, not to gate candidates.
Bulk candidate outreach emails. The reason templated outreach works so badly for recruitment is that candidates can tell. A half-automated personalised email is worse than no email. If you're going to do this, make the personalisation deep and narrow.
Reference checks. The legal exposure outweighs the time saved for most UK agency use cases. Build a template, keep a human in the loop.
What the typical automation stack looks like in 2026
A working recruitment automation stack in a small UK agency has four layers.
Layer 1: the ATS. Bullhorn, Vincere, JobAdder, or Adapt. System of record. Don't replace it. Work with its API.
Layer 2: the workflow orchestration layer. Zapier, n8n, Make, or custom code (Trigger.dev, Temporal, or a dedicated Python service). This is where the logic lives. Watches the ATS for triggers, calls external APIs, writes results back.
Layer 3: the data sources. Apollo, Clay, Cognism, LinkedIn (via scraping), Voila Norbert, Companies House API. You'll probably use two or three, not all. Pick based on your ICP.
Layer 4: the LLM decision layer. Claude or GPT for reformatting, summarising, extracting, and drafting. This is the "AI" in AI recruitment automation. One or two API calls per workflow is typical.
The question is what you use at layer 2. This is where agency archetype matters.
Who builds this for you (and who pretends to)
I wrote a longer version of this framing in my AI automation agency buyer's guide. The short version, applied to recruitment specifically:
- The Zapier or n8n reseller will quote you £800-£5,000 per workflow, built in their Zapier account or an n8n workflow instance. Works fine for simple one-to-one triggers. Breaks when you need fan-out logic (one candidate, seven client pipelines) or when the Zapier task count hits your plan limit.
- The AI-labelled consultancy will sell you a £12,000 recruitment automation roadmap. The roadmap is usually competent. The implementation after the roadmap is either out of scope or subcontracted.
- The engineer-led custom shop will quote you £12,500-£30,000 to build a recruitment automation layer that lives in your AWS or Render account and writes to Bullhorn over its API. You own the code. Seat costs don't scale as you add recruiters.
Laires Labs sits in the third category. The build floor is £12,500 because I cannot do smaller builds economically, and because smaller builds are better served by the first kind.
Real costs: SaaS stack vs custom build over 2 years
Same 10-recruiter UK agency, same desired automation (the first four items on the list above). Here's what each path actually costs across two years.
SaaS stack path
- Bullhorn (10 seats): £1,200/month × 24 = £28,800
- Apollo (5 seats): £500/month × 24 = £12,000
- Zapier Professional plus task tier top-ups: £240/month × 24 = £5,760
- Calendly team: £120/month × 24 = £2,880
- LLM API costs (moderate usage): £200/month × 24 = £4,800
- Setup and build by a Zapier reseller agency: £8,000 one-off
- Annual tweaks and maintenance: £3,000/year × 2 = £6,000
2-year total: £68,240. Of which £48,240 is recurring seat, task, and API fees you'd pay whether the automation exists or not. Net automation layer cost: around £20,000.
Engineer-led custom build path
- Bullhorn (10 seats): £1,200/month × 24 = £28,800 (same, nobody escapes the ATS seat fee)
- Apollo or equivalent data source (one API key, not per seat): £100/month × 24 = £2,400
- Render or AWS hosting: £40/month × 24 = £960
- LLM API costs: £200/month × 24 = £4,800
- Custom build by engineer-led shop: £18,000 one-off
- Maintenance retainer or ad-hoc work: £3,000-£6,000 across 2 years
2-year total: £58,000-£61,000. Savings vs SaaS path: £7,000-£10,000 across 2 years on a 10-recruiter agency.
The real story. The custom build path doesn't save much money in year one. It starts saving meaningful money by year two and accelerates from there, because the SaaS path adds seats as you grow, and the custom path doesn't. A 25-recruiter agency running the SaaS path will pay roughly 2.5x the recurring fees of the 10-recruiter version. The custom path recurring cost barely moves. That's the core economic argument for going custom once you're past about 12 recruiters and stable.
How to phase the build (if you're going custom)
The mistake I see most often in recruitment agencies going custom is trying to ship everything in one project. Automation scopes balloon. Priorities shift halfway through. The ops manager turns sceptical. The build ends up either half-finished or delivered late.
A better way to phase it, based on what I actually ship to agency clients.
Phase 1 (weeks 1-3): one high-ROI workflow, end to end. Pick candidate enrichment or CV formatting. Ship one workflow into your production ATS account. Catch the edge cases. Let the team use it for a fortnight. Fix what breaks.
The point of phase 1 is not the workflow itself. It's proving that your ATS API, your hosting, your logging, and your team's trust all work under real load. Once that foundation exists, every subsequent workflow is 60% cheaper to build.
Phase 2 (weeks 4-7): two more workflows that share infrastructure. Now that the auth tokens, the database, the error handling, and the deployment pipeline are in place, a second workflow ships faster. Pick interview scheduling or call note capture. Ship. Same drill: production use for two weeks, fix what breaks.
Phase 3 (weeks 8-12): the harder ones. Fan-out logic (one candidate across multiple client pipelines), BD signal monitoring, anything involving scraping or rate-limited APIs. These need the most engineering judgement and the most testing. Ship them only after phases 1 and 2 have settled.
Phase 4 (month 3 onwards): maintenance and iteration. APIs change. Enrichment vendors deprecate endpoints. A client asks for a custom match template. This is retainer territory, usually £500-£1,500/month or hourly ad-hoc depending on how much is changing.
The agencies that succeed with custom automation treat it like software, not like an IT project with a fixed end date. The ones that fail try to ship everything in a single "transformation" engagement and never circle back.
What not to automate
Three cases where the honest advice is "don't build anything."
You're under six recruiters. The maths doesn't work. Use Zapier free tier, use Voila Norbert free tier, lean on your ATS's built-in flows, accept the friction, save the money, hire another recruiter instead. Automation stops being a good return on investment under about five people because the hours-saved total is too small to cover the build cost.
You're restructuring the business. If you're mid-pivot from permanent to contract, or exiting a vertical, or consolidating ATSes, don't automate yet. Build the new ops manually first, let the workflows settle, then automate. Automating a process that's about to change wastes half the build.
Your ops manager isn't on board. This is the political one. An automation that the person running it doesn't trust gets bypassed. Build consensus before building software.
Book a call
Twenty minutes, no pitch. Send me your ATS, your rough recruiter count, and the two workflows you hate most. I'll come back with a scoped build plan: what I'd ship in phase one, what I'd phase after, what each phase would cost, and which of the five workflows above I wouldn't touch for your specific setup. You keep the plan whether you hire me or not.
Don't book if you're under six recruiters, if you're mid-pivot or restructuring, or if your ops manager isn't on board yet. Sort the structural stuff first or the build won't stick.
Book a 20-minute call or read the full AI automation agency buyer's guide if you're still deciding between archetypes.
Thinking about a system like this?
20-minute call, no slides. We'll map it against your stack and I'll tell you if it's worth building.
Book a callQuestions buyers actually ask
Any workflow where a step that used to be manual (enriching a candidate record, scheduling an interview, reformatting a CV, logging call notes) is handled by software triggered by an event in your ATS or adjacent system. In 2026 most recruitment automation has at least one LLM call somewhere in the chain for drafting, summarising, or extracting, which is why vendors often label it 'AI recruitment automation' even though the underlying workflow is deterministic.
Robin Laires
Ten years software engineering, former tech lead at Jellyfish — one of the UK's largest independent digital agencies. Now I build custom AI systems that replace manual business processes: ads ops, sales intelligence, intake routing, research pipelines. One engineer, installed into your stack.