Software for recruitment agencies, built by people who run one.

High volume recruitment loses most of its applicants in the first day. Someone applies at eleven at night, gets a call two afternoons later, and has already accepted something else. Every hour of delay is advertising spend that produced nothing.

Hiring more recruiters does not fix it, because the constraint is not effort, it is the hours in a day. Applications arrive in the evening and at the weekend. People do not.

We built and run a recruitment platform used daily across a national field sales network. That means we have already met the problems that only appear at volume: duplicate candidates arriving through three channels, screening that has to explain itself to a hiring manager, and reporting that has to reconcile across offices.

It also means we know which parts are not worth automating. Not every conversation should be a bot, and a supplier who tells you otherwise has not run the operation.

Screening and interview scheduling as a first build sits in the lower half of the £8,000 to £30,000 range. Candidate communication at volume, where speed of first contact decides the placement, is often custom AI work at £8,000 to £25,000 because the judgement being automated is a reading task.

An audit at £2,500 to £6,000 is a reasonable entry point where an applicant tracking system is already in place, since the useful question is what it is failing to do rather than what should replace it.

Time from application to first meaningful contact, measured in minutes and including evenings, because that is when candidates apply. Then interview attendance rate, which is where most agency effort quietly leaks away.

Also record how many candidates confirm an appointment and how many of those actually attend. A candidate who never engages with a confirmation is a different problem from one who confirms and does not arrive, and the two need different fixes.

The agency surveys published in 2026 say most agencies have adopted AI in some form and only a small minority have embedded it in how they operate, with disconnected data and a lack of training named as the blockers. Writing job adverts is where most start, because it is the easiest thing to do with a chat window, and it is the lowest-value thing, because the advert was never the bottleneck. The bottleneck is the hours after someone applies. Our recruitment guide sets out the figures and the four jobs where AI actually pays back, from the experience of running a recruitment platform.

Recruitment is the use case regulators watch most closely, because decisions about individuals are involved. The rules on significant automated decisions, Articles 22A to 22D of UK GDPR as amended by the Data (Use and Access) Act 2025, require safeguards where a decision about a person is made by automated means: the candidate is told, can contest it and can obtain human review. So screening is advisory, a recruiter decides, and the candidate is told AI is used. Systematic profiling needs a DPIA. Discrimination law makes the agency, not the vendor, answerable for a model that reproduces past bias, so outcomes are monitored by protected characteristic. The Conduct Regulations and the Employment Agencies Act set the rules on fees, terms and the information given to work-seekers, which automation must follow to the letter.

Our published prices apply to every sector: an AI adoption audit at £2,500 to £6,000 fixed, credited in full against any build; a first AI implementation at £4,000 to £15,000; a custom AI system at £8,000 to £25,000; applications at £8,000 to £30,000 per phase; websites at £4,000 to £16,000; training from £1,200. For a agency that usually means screening and message chasing as the first implementation, voice as a separate phase once the fallback is designed, and the CRM you already run kept as the record.

The running cost is stated before go-live, per feature, and it scales with volume rather than headcount. From operating our own AI products, the model is rarely the expensive part; the channels around it, telephony and messaging, usually are, and a well-designed workflow sends most of its work to the small, cheap models. Every system we build reports cost per feature from day one and has spending ceilings that were tested by watching them fire.

Onrolo, the recruitment operations platform we build and operate, is in daily use across a national recruitment network on web, iOS and Android. It is where the timelines and the running-cost figures on this site were measured, and it is why the advice here is specific: the small model does most of the work, the channels cost more than the model, roughly one outbound call in eight reaches a human, and the fallback matters more than the voice.