You Don’t Have an AI Problem. You Have a Data Problem.

87% of companies have now incorporated AI into their recruitment processes (DemandSage) — and yet most agency owners will tell you they’re not seeing transformative results. The tools are there. The subscriptions are active. And the outcomes are… fine, at best. Confusing, at worst. In a straight-talking episode of RecTalk, Nitin Sharma sat down with Dylan Humphreys from First Frontier AI and Poonam Mawani from Azuki Accounts to answer the question that isn’t being asked often enough: why isn’t the AI working? Their answer: for most agencies, the problem isn’t the AI. It’s everything that comes before it.

The Wrong Starting Point

Most recruitment agencies approach AI the same way they approach any shiny new technology — they find a tool, buy a subscription, and try to work out what to do with it afterwards. It’s the same pattern that played out with CRMs a decade ago. And it produces the same result: underutilised software, frustrated teams, and the nagging feeling that everyone else has figured this out except you.

Dylan’s framework flips the sequence entirely. The right starting point for AI adoption isn’t “what tools are available?” It’s “what business problem am I actually trying to solve?” That sounds obvious. It’s consistently ignored. CV overload is a business problem. Candidate re-engagement is a business problem. Losing track of warm leads in a messy CRM is a business problem. Once you’ve named the problem clearly, you can work backwards to whether AI, automation, or simply a better process is the right solution — and often it’s the latter, not the former.

The agencies that get genuine value from AI are the ones that have done this diagnostic work first. They’re not looking for AI to fix everything. They’re using it to fix specific, named things — and measuring whether it’s working.

Poor Prompting = Poor Outcomes

One of the most practical points in the episode is also one of the least discussed in recruitment circles: AI is only as good as the instructions you give it. This applies to everything from writing job adverts to screening CVs to generating outreach copy. Vague prompts produce vague outputs. Garbage in, garbage out — the principle hasn’t changed, it’s just applied to a new layer of technology.

Agencies that complain AI-generated content sounds generic, or that their AI screening tool keeps missing obvious candidates, or that the automation they set up keeps doing the wrong thing — almost always have a prompting problem rather than a technology problem. Learning to write clear, specific, contextual instructions for AI tools is a genuine skill, and it’s one most recruitment teams haven’t been trained in. It’s also not complicated to develop. It just requires treating the prompt as part of the process, not an afterthought.

The Real Bottleneck: Your Data

Here’s the uncomfortable truth Dylan and Poonam surface: most recruitment agencies have a data problem that AI cannot fix and will actively make worse. Databases full of duplicate contacts, outdated job titles, missing phone numbers, and candidates last touched in 2019. CVs stored inconsistently across different systems. Client records that no one fully trusts. When AI tools are layered on top of this — for screening, matching, outreach, or reporting — they don’t clean the mess up. They automate it.

The agencies seeing the best returns from AI investment are the ones that treated data hygiene as a precondition, not an afterthought. That means auditing what’s in the CRM before trying to make it intelligent. It means building consistent data capture habits before automating the capture process. It means deciding what “clean” looks like for your specific business before asking AI to work with it.

None of this is glamorous. It also doesn’t require a big budget or a consultant. It requires treating your data as the asset it actually is — because every other tool you build on top of it is only as good as what’s underneath.

Where AI Actually Delivers in Recruitment Right Now

To be clear: AI does add genuine value in recruitment today. Dylan is bullish on it — but specific about where. The areas that show the most reliable returns are:

• CV screening and candidate triage. AI can process large volumes of applications faster than any human, flag the most relevant profiles, and surface candidates who might otherwise be missed. The caveat: it needs clear criteria and human oversight for the judgement calls. It should narrow the field, not make the final decision.
• Market intelligence and BD research. Pulling together data on target companies, industry hiring trends, competitor activity, and prospect signals is time-consuming manual work that AI can compress dramatically. Consultants who use AI for research before client conversations are consistently better prepared and more credible.
• Back-office and operational automation. Poonam’s perspective from the finance and operations side is that the highest-value automation wins aren’t sexy — they’re repetitive tasks like invoice processing, timesheet chasing, reporting, and compliance documentation that consume hours every week without adding any strategic value.
• Outreach personalisation at scale. AI can help draft personalised candidate and client outreach faster — but again, the quality of the output depends entirely on the quality of the data and the instructions it’s given.

Real Talk

AI isn’t going to rescue a messy CRM, a vague process, or a team that doesn’t understand what it’s trying to achieve. Start with the problem. Fix the data. Learn to prompt properly. Then bring in the tools. In that order.

This post is inspired by the RecTalk episode with Dylan Humphreys and Poonam Mawani: AI in Recruitment: Hype, Reality & The Future of Agency Growth. Watch the full conversation on YouTube.

Looking for the AI and recruitment tech tools actually worth investing in? Browse 1,000+ vetted recruitment suppliers and 49,000 real reviews — free — at rectools.io.