Why keyword filters miss your best candidates, and what explainable AI matching does differently

Here is a scene I have watched many times. A role opens. A recruiter types four keywords into the database. Sixty results come back, most of them wrong. Then, two weeks later, someone remembers a candidate from last year who would have been perfect. She never appeared in the search, because her CV said “customer retention” and the search said “client relationship management”.

The cost of not finding people you already have

Most employers are not short of applicants; they are short of the right ones. Ask any hiring manager what makes recruitment hard and “finding candidates with the right experience” will be near the top of the list.

Yet most agencies and in-house teams are sitting on years of CVs. Every time they pay to advertise a role their database could have filled, keyword search is part of the reason.

Why keywords fail

• People describe the same work differently. “Bancassurance”, “insurance sales through bank branches” and “wealth advisory with a bank partner” can describe the same job.
• Languages mix. In multilingual markets, a CV may combine English job titles with Chinese company names and descriptions. Exact-match search handles that poorly.
• Seniority hides in context. “Managed the team’s onboarding” and “onboarded” contain the same keyword and mean very different things.
• Keywords reward copying. As I wrote in the AI application flood, AI tools now rewrite CVs to mirror your job ad. Keyword filters rank those copies first.

What “AI matching” should actually mean

Plenty of tools now say they use AI to match candidates. The useful question is not whether a tool uses AI, but whether you can see its reasoning. When you evaluate any matching tool, including ours, ask three things:

1. Does it compare experience with requirements, or words with words? A good match should find the candidate who did the work, whatever they called it.
2. Does every suggestion come with a reason you can check? “87% match” tells you nothing. “Five years handling commercial claims, including a team lead role” tells you what to verify.
3. Who decides? The tool should suggest. Your recruiter should decide, with the candidate’s full profile one click away.

How matching works in Talenda

Talenda first turns each CV into a structured profile, so experience, skills, education and languages sit in the same fields for every candidate. When you open a role, it compares the role’s requirements with the experience across your candidate pool, using semantic search rather than exact keywords.

Each suggested candidate comes with match reasons: the relevant experience Talenda found and how it relates to the role. Recruiters read those reasons, open the profile and decide whether to reach out. If the reason doesn’t hold up, you can see that immediately, which is exactly the point.

Your team can also ask questions in plain language through Ask Anything, such as “What insurance sales experience does this candidate have?”, and get answers grounded in the saved profile. Where the profile doesn’t say, the answer is “unknown”, not a guess.

Start with the pool you already have

Before you advertise your next role, search your existing candidates properly. Write the requirements as outcomes, not keywords. Look for evidence of the work, not the vocabulary. And use tools that show their reasoning, so the final call stays with someone who understands the role.

Your best candidate may already be in your database. The job of AI is to help you find them, and to show you why.