James ran a seven-person tech recruitment agency in Manchester. His consultants were good. His clients were happy. But every week, 120 CVs arrived in the shared inbox, every consultant spent 90 minutes a day chasing candidates for status updates, and every Monday morning evaporated into three hours of client reporting. The work wasn't hard. It was relentless. Here's how three AI agents changed the numbers — and what it actually took to get there.
The Starting Point: Volume That Good People Can't Absorb
Ashford Talent is a Manchester-based specialist recruitment agency placing mid-level developers, data engineers, and product managers into UK tech firms and scale-ups. Seven staff: four consultants, two researchers, one administrator. Around 40 active client accounts at any given time, placing roughly 22 professionals per month.
Revenue was solid. The team was capable and experienced. But James, the director, described something we hear from almost every agency owner: growth had stalled, not because there weren't enough jobs or candidates, but because the team was maxed out on process management.
Here's what the week actually looked like:
- CV screening — 120 CVs per week, each reviewed manually against the relevant job spec. At three minutes per CV, that's six hours of team time every week spent on triage that produces no candidate relationships and no revenue
- Candidate status updates — consultants spending 90 minutes a day sending "where are you in the process?" replies, fielding FAQ messages, and arranging interviews by email back-and-forth
- Interview scheduling — every interview required 3–5 emails to align three calendars. Average time per interview arranged: 45 minutes
- Monday morning client reports — 40 active accounts, each needing a weekly pipeline summary pulled manually from the CRM. Three hours of administrative work before the week had properly started
Across the team, roughly 30% of every working week was going to process. Not client development. Not candidate relationships. Process.
What the Audit Uncovered
James booked a £500 AI Audit. We spent a day mapping every recurring process in the business — time cost, decision complexity, data structure, and automation suitability. The output is a prioritised matrix: what to automate first, what to leave alone, and what to revisit later.
For Ashford Talent, four processes scored above 4.0 on our five-point suitability scale:
- CV screening and shortlisting — structured input (CV + job spec), structured output (score + shortlist flag). Automation score: 4.7/5
- Candidate status communications — templated, trigger-based, largely one-directional messages. Automation score: 4.5/5
- Interview scheduling — pure calendar logic, no judgment required. Automation score: 4.8/5
- Client pipeline reporting — data in from the CRM, formatted summary out. Automation score: 4.4/5
One process we specifically recommended leaving alone in phase one: the initial candidate screening call. Reading ambition, assessing cultural fit, sensing whether someone is genuinely looking to move — that still requires a consultant. Automating everything around that call means consultants can do more of the work only they can do.
"I kept telling myself we needed to hire another consultant. After the audit, I realised we needed to automate. Those are very different problems — and very different costs."
— James Ashford, Director, Ashford Talent
Projected weekly hours recoverable from the four priority processes: 28–35 hours. At a blended staff cost of £22 per hour, that's £616–£770 of recovered capacity per week.
Building the AI Operating System
We built on the same infrastructure we use for every deployment: OpenClaw running on Google Cloud Run, connected to the agency's shared inbox, their CRM (Vincere), and a calendar integration for the four consultants. Total monthly infrastructure cost: £38.
Three agents. Each scoped tightly. Each with a clear handoff back to a human before anything consequential happens.
Agent 1: CV Screening and Shortlisting
This agent monitors the inbound CV inbox — direct applications, job board forwards, and LinkedIn messages forwarded by consultants. When a CV arrives, it identifies the relevant live vacancy from the CRM and scores the candidate on five criteria: technical skills match, seniority level, sector background, location or remote compatibility, and salary alignment.
Candidates above the scoring threshold are flagged in the CRM with a structured summary note. Consultants review the flagged candidates only. Everything below threshold goes into a holding folder, reviewed in a weekly batch if volume allows.
In the first week of live deployment, the agent processed 114 CVs. Consultants reviewed 28 flagged candidates in under 20 minutes total. The same 114 CVs would previously have taken around six hours to triage manually.
Agent 2: Candidate Outreach and Interview Scheduling
This agent handles all status communication between application received and offer made. When a candidate is shortlisted, it sends a personalised acknowledgement. When a consultant progresses a candidate to interview, the agent triggers the scheduling sequence: it checks both the consultant's and the client's available slots via the calendar API, presents three options to the candidate, confirms the chosen time, and sends calendar invites to all parties.
For inbound candidate FAQ messages — "any update on the role?", "how did the interview go?" — the agent drafts a response using the current CRM status and places it in a review queue. The consultant approves or edits with one click.
Interview scheduling time dropped from 45 minutes per interview to 8 minutes. Candidate update emails dropped from 90 consultant-minutes per day to around 15 minutes of queue review. Consultants get the time back. Candidates get faster responses.
Agent 3: Client Pipeline Reporting
Every Monday, 40 clients were supposed to receive a pipeline update. In practice, some got them by 10am, some got them by Thursday, and some were chased. The administrator who owned the task had 40 separate reports to compile, each requiring her to open the CRM, locate the account, pull last week's activity, and write a formatted email. Three hours, on a good Monday.
The reporting agent runs at 7am every Monday. It pulls the previous week's activity from Vincere for each active account — CVs submitted, interviews completed, offers made, placements confirmed — and generates a formatted email update for each client. By 9am the queue is in the administrator's dashboard. She reviews and sends in under 20 minutes.
The Numbers at 90 Days
At the 90-day mark, we pulled performance data across all three agents and compared it to the pre-build baseline. The numbers exceeded the audit projections on every metric.
| Metric | Before | After 90 Days |
|---|---|---|
| Weekly CV screening time | 6 hrs | 20 mins |
| Daily candidate comms (team total) | 6 hrs | 1 hr |
| Interview scheduling (per interview) | 45 mins | 8 mins |
| Monday client reporting | 3 hrs | 20 mins |
| Monthly placements | 22 | 38 |
| Monthly infrastructure cost | £0 | £38 |
| Weekly team hours recovered | 0 | 32+ |
Monthly placements went from 22 to 38 — a 73% increase — without adding a single person to the team. The 32 weekly hours recovered went back into business development calls and candidate relationship building: exactly the activities that move placement numbers.
The full build cost — audit, build, and 90 days of operation — came to £3,400. At approximately £700 per week of recovered capacity, the system paid for itself in under five weeks.
"The Monday morning reporting session used to be the thing we all dreaded most. Now I barely think about it — the agents have already done it. That shift, from dreading a task to not noticing it, is when you know something has actually changed."
— James Ashford
What Didn't Work (And What We Fixed)
Every deployment has rough edges in the first 30 days. Two issues needed attention here.
The CV screener was too conservative on career changers. In the first three weeks, candidates moving from adjacent sectors were consistently scored below the shortlist threshold, even where a consultant would have flagged them. The criteria were measuring direct sector experience too heavily. We added a "transferable skills" scoring dimension and adjusted the weighting. Over-exclusion dropped significantly in week four.
Three clients had non-standard reporting preferences. Two wanted only confirmed interviews reported, not applications submitted. One needed the report sent on Friday instead of Monday. The base template didn't account for this, which caused a minor issue in the first two weeks. We added a client preferences field to Vincere and the reporting agent now reads it before generating. It took one day to build.
Both were caught through the oversight layer — the human review step we maintain for the first 30 consecutive days of any deployment. As we describe in the 90-Day AI Transformation Playbook, the oversight period isn't a sign that the system isn't working. It's the mechanism that makes it reliable.
The Pattern Behind the Numbers
Ashford Talent's story repeats a pattern we've now seen across professional services in the UK — from the Yorkshire accountancy firm we wrote about last week to consulting and legal businesses we're working with now. The tools differ. The problem is the same: capable people doing 25–35% of their week on process that software can handle reliably.
For recruitment agencies specifically, the implications go further than time savings. Every hour recovered from CV triage, scheduling logistics, and reporting is an hour that can go into the only things that actually move placement numbers: building client relationships and having real conversations with candidates. The agencies that compete on those things will pull away from the ones still doing it all by hand.
Industry data backs this up. A 2026 GRID report found that firms using AI for candidate management are twice as likely to have grown revenue year-on-year. The gap between early movers and everyone else is already opening.
If you run a recruitment agency — or any professional services firm where operational process is absorbing capacity that should be going into client and candidate work — the starting point is the same one James used. Get in touch and we'll walk you through what an AI operating system would look like for your business.