What a real AI hiring audit looks like, and what it finds.

"AI audit" has become one of the most diluted phrases in recruitment. Half the vendors offering them run a structured product demo with the conclusion already written. The other half hand back a 40-page deck of generic recommendations the operations team has no chance of executing.

An honest AI hiring audit does something different. It looks at how your operations actually work today, where time and money are leaking, and which of those leaks AI can credibly close. Just as importantly, which ones it can't.

Here's what one looks like when you do it properly.

What we actually look at

A real audit examines four things in this order. The order matters, and most audits get it backwards.

1. Process before tooling

Map the end-to-end hiring process for one representative role: timestamps, owners, handoffs, decision points, exceptions. We do this before looking at any tool, because tools are downstream of process. Most of what gets blamed on "the ATS is bad" is actually a process that doesn't fit the tool, or a tool being asked to do work it was never designed for.

2. Data hygiene

Pull a sample from the ATS and CRM and look for the basics. Are candidates tagged consistently? Are pipeline stages defined the same way across recruiters? Are time-to-fill and cost-per-hire calculated identically? Are notes structured or free-text? AI runs on data. If your data is messy, no AI tool will save you. It'll just produce confident garbage faster.

3. Real cost mapping

For each role, where is the time going? Coordinator hours, recruiter hours, manager review hours, client-side hours. Most SME recruitment leaders are surprised at how concentrated the cost is. Usually two or three workflow steps eat 60 to 70% of the labour. AI investment that doesn't target those steps is wasted spend.

4. The human part nobody automates

Where in the process does the firm actually add judgement? Match quality, candidate sell, client positioning, negotiation. These are the parts AI shouldn't touch directly, but should support. The audit needs to surface them clearly so they don't accidentally end up in the "automate this" bucket.

The three problems we find almost every time

Across the SME recruitment operations I've audited, three patterns appear so consistently that I now expect them on day one. They're unglamorous. They're not what vendors want to talk about. They're where almost all the value lives.

Problem 1: The pipeline is a fiction

"Stage 2: Phone Screen" means something different to every recruiter on the team. One person uses it for "scheduled," another for "completed," a third treats it as a wastebasket for candidates they're not sure what to do with. The pipeline reports management looks at every Monday are, in practice, made up of categories nobody agrees on.

You can't deploy AI ranking, scoring or routing on top of this. You'll get sophisticated guesses pointed at the wrong stage. The fix is unglamorous: define stages in writing, train the team, audit weekly until everyone's pipeline reads the same way.

Problem 2: The same candidate exists three times

Duplicate candidate records. Wildly common, almost universally underestimated. Most ATS deployments have anywhere from 8% to 25% duplication once you actually look. This destroys your ability to track candidate history, trains AI matching on noise, and silently frustrates candidates who get re-contacted by different recruiters at the same firm.

Deduplication is unglamorous, manual, and the prerequisite to any AI you'd actually want to deploy.

Problem 3: Nobody owns the metrics

Time-to-fill, cost-per-hire, source quality, client conversion. These are the numbers that should be steering the firm. In most operations I audit, they exist in three different spreadsheets, calculated three different ways, with nobody clearly responsible for keeping them current. By the time leadership sees them, they're stale, contradictory, or both.

AI that depends on these metrics, anything that ranks roles, flags risks, predicts time-to-fill, will inherit all that ambiguity. Step one is always the same. Pick the canonical version, assign an owner, automate the calculation. Then, and only then, layer AI on top.

What the audit gives you

A proper audit ends with three things, not 40 slides:

1. A map of where time and money are leaking, by workflow, with rough numbers.
2. A short list of fixes ordered by impact-per-week-of-effort. Usually 3 to 5 items, not 30.
3. A clear separation between what should be redesigned, what should be automated, and what should genuinely use AI. They are not the same.

The most useful thing an audit does is stop you from buying the wrong AI tool. The second most useful thing is showing you the redesign you can do this quarter that will pay for itself before any AI gets deployed.

Bottom line: if an AI hiring audit doesn't start with your process and end with a fixed-scope, prioritised list of changes, it's a sales deck. Ask for the audit that gives you the list. Then run it.