What the research actually says about getting better at this job
Recruitment has never had a training problem. It has had a practice problem.
Think about how most of us learned. You sat next to someone good and listened. You did a roleplay in front of the team, which was less about learning and more about surviving. Then you picked up the phone and found out the hard way. If you were lucky, someone had ten minutes afterwards to tell you what went wrong. If you were not, you repeated the mistake for another six months.
I did that for fifteen years, first as a recruiter and then running agencies. I hired trainers at a thousand pounds a day. I bought e-learning at two hundred and fifty a month. I trained my top performers myself. None of it moved the numbers the way I expected, and it took me an embarrassingly long time to work out why.
The problem was never the content. It was that almost nobody was practising.
Here is what the research says about that, and why it matters to you personally rather than to your manager.
1. Doing beats reading, and it is not close
The single most replicated finding in learning science is the testing effect, sometimes called retrieval practice. Put simply, being made to retrieve something under pressure builds far more durable capability than reviewing it.
Rowland’s meta-analysis pooled 159 separate comparisons and found a medium effect size of g = 0.50 in favour of practice over restudy. More striking, 81% of all comparisons favoured practice.
Read that again. Four out of five times, doing the thing beat reading about the thing.
Every recruiter has watched a training video on objection handling. Almost none have said the objection out loud, in real time, with something pushing back at them, before a client did it for real. That gap is where most of the value sits.
2. Feedback helps, unless it feels like a verdict on you
This is the finding that changed how I think about coaching entirely.
Kluger and DeNisi analysed 607 effect sizes across 23,663 observations. On average, feedback improved performance, with an effect size of d = 0.41. But more than a third of feedback interventions made performance worse.
Their explanation is the useful bit. Feedback works when attention stays on the task. It backfires as attention moves up towards the self. Tell someone their questioning was too closed in the third minute and they fix it. Make them feel judged as a recruiter, in front of people whose opinion they care about, and a good chunk of them get worse.
Now think about how recruitment coaching is traditionally delivered. Live call listened to by your manager. Feedback in an open-plan office. A roleplay in front of the desk you sit with every day.
We built a system where a third of the intervention was always going to misfire, then wondered why people avoided it.
Practising privately, against something that scores the call rather than judging you, is not a soft option. It is structurally the version of feedback the evidence supports.
3. The people who gain most are the ones who feel furthest behind
The largest workplace study of generative AI so far followed 5,179 customer support agents through a staggered rollout (Brynjolfsson, Li and Raymond, Quarterly Journal of Economics, 2025).
Productivity rose 14% on average. Among the newest and least experienced, it rose 34%.
The mechanism is what should interest you. The researchers found the AI was capturing and spreading the tacit habits of the highest performers, the things good people do without being able to explain them, and handing those habits to everyone else. It helped newer workers move down the experience curve faster. Customer sentiment improved. So did staff retention.
If you have ever sat next to a top biller and thought “I do not know what they are doing differently”, that study is about you. The gap is usually not talent. It is a set of small behaviours nobody has ever articulated to you.
And the honest caveat, because it matters. In that same study, the most experienced agents gained the least. Which is exactly what we see. Our strongest users do not use practice as general training. They use it before one specific call they are worried about. Different job, still valuable, but let us not pretend the 34% applies to everyone equally.
4. Confidence is not fluff, it is a performance variable
Stajkovic and Luthans pooled 114 studies covering 21,616 people and found a weighted average correlation of .38 between self-efficacy and work performance.
Self-efficacy is not self-esteem. It is the specific belief that you can execute this particular task. It is built by doing the thing successfully, repeatedly, in conditions close enough to the real thing that your brain accepts the evidence.
Which is the whole argument for reps. You do not talk yourself into confidence before a difficult client call. You earn it by having already had that conversation twenty times.
What this looks like on a Tuesday
None of the above requires you to believe anything about AI. It requires you to believe that practice beats review, that private task-level feedback beats public judgement, and that confidence comes from having done the thing before.
Rekbot exists because I could not find a way to give my own teams those three things at scale.
What it does is unglamorous. You practise a real call. It pushes back the way a real prospect would. It scores the call, not you. It tells you the specific thing to fix, and it builds your next session out of the pattern it sees in your last one. You can do it at ten at night with nobody watching, which for a lot of people is the entire point.
When a graduate cohort ran this properly, they were handling live calls after three days rather than shadowing for weeks, and reached their first self-generated client win at around week eight, against a norm of four to six months. One reached over two hundred thousand pounds in fees with more than fifty roles by week twelve.
I am not going to tell you that is typical, because it is not. It was a good cohort who used it properly. Which is rather the point of this article.
The ask
If you have an account and you have not opened it in a fortnight, this is the case for twenty minutes.
Not twenty minutes of content. Twenty minutes of reps on the one call you have been quietly avoiding. The pitch you fumble. The PSL objection you talk around. The fee conversation you concede too early.
Pick the thing you are worst at, do it badly in private six times, and watch what happens on the seventh when it is real.
The evidence has been settled for decades. The only variable left is whether you use it.
Will Bourne is the founder of Rekbot, an AI training platform that recruiters practise real client calls against and get scored on. He spent fifteen years running UK recruitment agencies before building it.
Sources
Stajkovic, A. D. and Luthans, F. (1998). Self-efficacy and work-related performance: a meta-analysis. Psychological Bulletin.
Rowland, C. A. (2014). The effect of testing versus restudy on retention: a meta-analytic review of the testing effect. Psychological Bulletin.
Kluger, A. N. and DeNisi, A. (1996). The effects of feedback interventions on performance. Psychological Bulletin, 119(2), 254-284.
Brynjolfsson, E., Li, D. and Raymond, L. (2025). Generative AI at Work. Quarterly Journal of Economics, 140(2), 889-942.
Will Bourne
https://rekbot.aiFounder and CEO, Rekbot.