Does Coaching Work? What the Research Actually Says
In brief
What four meta-analyses actually show about coaching, why the headline number needs a footnote, what predicts a good outcome, the side effects, and why the ROI multiples cannot be trusted.
A senior leader deciding whether to spend money on a coach deserves a straight answer before signing anything. The two answers usually on offer, that coaching is proven and that it can't be, are both wrong. The published research supports a narrower claim.
The Headline Number
Meta-analyses combine many separate studies into a single estimate. Four independent ones exist for coaching, run by different research teams using different inclusion rules, and they converge on the same range. Theeboom, Beersma, and van Vianen combined eighteen studies in 2014 and found an overall effect size of 0.66 (a standardized measure researchers use to compare how large a difference coaching made across different studies), with individual outcomes ranging from 0.43 for coping up to 0.74 for goal-directed self-regulation, the strongest single result in their data. Jones, Woods, and Guillaume restricted their 2016 review to workplace coaching only, a stricter filter, and found a lower overall effect of 0.36, though the results-level outcomes in their data reached 1.24, a number drawn from far fewer studies, which the authors themselves flag as unstable.
Two further meta-analyses went stricter still, including only randomized controlled trials: studies where people are assigned to coaching or not by chance rather than by choice, generally the strongest design a field can produce. De Haan and Nilsson found an effect of 0.59 across thirty-seven RCTs published between 1994 and 2021. Nicolau, Candel, Constantin, and Kleingeld, reviewing twenty RCTs the same year, found 0.43, then ran a statistical test for publication bias, the tendency for studies with positive results to get published more often than studies that found nothing. The test came back significant. Correcting for it dropped their estimate to 0.27, roughly forty percent smaller than the number they started with.
The effect is real and moderate, comparable to or a little better than standard managerial-skills training, and smaller than coaching marketing tends to imply. The two RCT-only reviews land at 0.59 and 0.43, and the one that corrected for publication bias landed at 0.27.
Why Even That Number Needs a Footnote
De Haan and Nilsson's own RCT-only review reports signs of publication bias despite its 0.59. Nearly everything measured in this literature is self-report: the coachee rating their own improvement, on their own instrument, at the end of an engagement they or their employer paid for. Several individual studies have found coachees reporting real improvement with no corresponding change on independent ratings from a manager or a peer, a pattern researchers treat as a serious open problem, since self-report in this specific literature runs measurably higher than outside observation. Nicolau and colleagues also flag that some of their own subgroup results rest on as few as three studies, which makes those specific breakdowns less stable than the headline number suggests. And no meta-analysis reviewed here has measured organizational-level payoff as a primary, independently verified outcome: the question of whether coaching pays off for whoever is writing the check has not actually been tested at the level where it would matter to them.
What Actually Predicts a Good Outcome
The largest study of what drives a good result, rather than just whether coaching works in general, followed 1,895 matched coach-client pairs across thirty-four countries. The strongest predictor of a good outcome was the working alliance: the real, felt quality of the relationship between coach and client, rated directly by both sides. Personality match between coach and client, the premise behind a lot of coach-matching marketing, predicted nothing at all. Two separate meta-analyses tested whether the number of sessions or the length of the engagement predicted outcome, and neither found that it did. No comparative study anywhere in this literature shows one credentialing body's method producing better coaches than another's.
That leaves a buyer with a specific, usable finding: the relationship is the variable to evaluate directly, in a real conversation, and a personality quiz or a certification badge is no substitute.
The Side Effects
A study of 111 coachees found that just over two-thirds reported at least one negative effect from coaching, averaging 3.46 effects each, generally mild. The named categories included reduced job satisfaction, the coach changing the coaching goals without the coachee's explicit agreement, and the coach surfacing a problem they weren't equipped to handle. The same study found a stronger working alliance predicted fewer of these effects: the relationship that predicts a good outcome also predicts less harm.
This comes from a single study. No review has systematically tested coaching's downsides the way the effectiveness studies above have tested its benefits, and no number exists for them.
The ROI Numbers, and Why They Can't Be Trusted
The return-on-investment figures that circulate in coaching sales material, generally somewhere around five to six times the cost, rest on retrospective, self-reported estimates from the executives who received the coaching: the people with the clearest motivation to believe, and report, that their own investment paid off. The most-cited version traces to a 2001 study of forty-three senior executives, reporting their own estimate of the payoff after the fact. None of the studies behind these figures were peer-reviewed or independently replicated, and no meta-analysis in this literature has measured organizational ROI as a verified outcome rather than a self-report. A separate methodological critique goes further, arguing that leaning on a single financial-return number narrows what an organization even notices about coaching's actual effects.
What This Leaves a Buyer to Decide
Coaching has a real, moderate effect at the aggregate level. That evidence has nothing to say about whether it will work for the person deciding right now.
What the evidence doesn't support is a specific ROI multiple, a “right” number of sessions, or a promise that coaching will change the outcome for any one person. What it does support is checking the relationship itself, in an actual conversation, over anything printed on a coach's profile or a certification badge. The Coaching Agreement Checklist, and the self-check “Is Coaching Working?,” cover what to do with that.