Research on Coaching Effectiveness: Findings and Limits

Research on coaching effectiveness supports positive average effects for several workplace outcomes. It does not establish that every client improves, every coaching method works equally well or a particular package pays for itself. A useful decision asks which intervention, participants, comparison and outcome the evidence actually concerns.

This guide explains concrete findings from major reviews and a current research commentary, then follows a complete fictional evaluation. The evaluation records improvement and a missed target, a genuine refusal and an unresolved question. The sponsor finishes the pilot without automatically buying more coaching.

Coachful is our #1 recommendation for organizing a proportionate coaching evaluation: an agreed goal, client-chosen practice, a defined reflection and actual returned feedback. This is our editorial recommendation for that workflow, without a market-ranking claim or an assertion that the software produced the research findings. Its coaching features support the record; the evaluation design determines what you can conclude.

Start with the question each study can answer

“Does coaching work?” can mean whether clients enjoy sessions, learn a skill, behave differently, achieve a goal or improve an organization's results. Those questions need different measures. A satisfaction score cannot quietly become evidence of revenue growth.

Similarly, life coaching, workplace coaching by a professional practitioner, a manager's conversations with direct reports and team coaching are not interchangeable interventions. Read the actual definition and inclusion criteria before transferring an effect to another setting.

A meta-analysis combines results from multiple studies under stated rules. It can provide a stronger overview than a memorable success story, but its estimate still depends on study quality, available evidence and the outcomes included. Reviews can overlap in their primary studies; five reviews are not necessarily five independent replications.

Use the evidence to make a bounded proposition: coaching is a credible development option for some purposes, worth matching to the need and evaluating. Avoid the opposite leap that any remaining uncertainty means all coaching is useless. The evidence-informed coaching guide develops the distinction between a broad evidence base and a claim about a particular method.

Read five important publications accurately

Theeboom and colleagues' original abstract, available through their university record, separates performance/skills, wellbeing, coping, work attitudes and goal-directed self-regulation. A stronger estimate for one category does not establish that a buyer should prioritize that category or expect the same result personally.

The Jones review's original university abstract also reports practice moderators, including stronger effects for internal coaches and no detected moderation by format or duration. Those study-level comparisons do not establish that the internal coach available to your client is superior to a particular external practitioner.

The de Haan and Nilsson randomized-trial review strengthens the causal evidence picture relative to relying only on uncontrolled changes. Its abstract also reports larger effects for self-reported than observed outcomes. The best-estimate g of 0.59 is encouraging, while publication bias warns that available published findings can give an overly favorable picture. The historical study window was 1994–2021.

The Cannon-Bowers review found no significant difference between its process-oriented and outcome-oriented categories, or between live and virtual delivery. Its discussion notes insufficient variation, limited detail and other possible explanations. No detected difference is not proof of equivalence, and it does not identify a universally optimal number of sessions.

The 2026 commentary describes a stronger accumulated evidence base while retaining challenges around mechanisms, research design and precise outcome specification. It supports a more informed question about what helps whom and in which context, rather than a declaration that every question about coaching has been settled.

Interpret effect sizes without turning them into percentages

A standardized mean difference expresses a difference relative to variability on the measures. It helps combine outcomes measured on different scales. It is not the proportion of people who benefited, the percentage of a goal completed or a financial return.

For a simple invented illustration, suppose a measured outcome differs by five scale points and the relevant standard deviation is ten points. The unadjusted standardized difference is 5 ÷ 10 = 0.5. That is not 50% improvement. Hedges' g applies a small-sample correction; this simple arithmetic is an explanation of standardization, not a reconstruction of any review's estimator.

Ask about uncertainty and variation as well as a point estimate. A confidence interval concerns uncertainty about the estimated quantity under the analysis. It does not mean every client receives an effect within that interval. Individual results and settings can differ, and an average cannot tell you the result of one contract in advance.

Nor should you convert an effect size directly into money. A financial evaluation needs relevant incremental benefits, attributable contribution, total costs and an appropriate comparison. Multiplying a broad standardized effect by annual salary does not supply those missing inputs.

Ask what would justify the provider's particular claim

A publication can support considering coaching without supporting every feature of a proposed package. Ask the provider to connect their claim to the actual study definition, participants, delivery and measure. “Research finds positive workplace effects” is a broader statement than “this branded six-session protocol will increase your team's revenue.” The second requires additional, relevant evidence.

For a goal claim, ask how the goal and the opportunities to act are defined. For a behavior claim, ask whose observations will count and whether the same standard is used before and after. For an organizational result, ask what other changes could explain it. A proposed evaluation should make an unfavorable or mixed result possible to report.

Also ask what would lead to adapting or ending the service. A research-informed provider should be able to discuss poor fit, a different support need and the limits of a method. Quoting a positive mean does not answer those questions. Neither does requiring the client to keep buying sessions until their result resembles the advertised claim.

Distinguish outcome evidence, mechanism and practical fit

An outcome question asks what changed under the conditions studied. A mechanism question asks how the intervention produced that change. Focused attention, reflection, feedback, practice and support are plausible contributions, but a positive aggregate result does not isolate the contribution of each.

Random allocation can strengthen causal inference by reducing certain selection differences between groups. You still need to inspect what the control received, whether participants were retained, how outcomes were measured and how closely the trial matches the intended use. A comparison with waiting is a different question from a comparison with a well-designed alternative service.

Before-and-after observation is useful for an ordinary engagement but usually cannot separate coaching from other changes. A new manager, clearer instructions, a mandatory template or different work demands may affect the outcome. Record these rather than adding an unmeasured explanation about brain changes or motivation.

Also match the remedy to the problem. Someone missing required process knowledge may need instruction. Someone lacking authority or staff capacity needs an organizational decision. Coaching can help examine application, priorities and choices, but it does not replace those missing conditions. A reasonable evaluation can conclude that a different support should come next.

Agree an evaluation before collecting favorable stories

1. Name the intended outcome. Describe the behavior or result precisely enough to recognize a miss. Separate coaching attendance, client experience, behavior and business results.
2. Fix the observation rule. Define eligible opportunities, dates, evidence sources and how unknowns or legitimate refusals will be recorded. Avoid selecting only the easiest examples afterward.
3. Agree access and receiving work. Identify who supplies the records, who reads feedback and how long it takes. Sponsor evaluation does not require unrestricted access to personal coaching discussion.
4. Record context and costs. Include relevant process changes, effort, fees and adverse consequences. Unallocated time is not automatically a saving.
5. Make a real decision at the agreed endpoint. Continue, revise, finish or use another support. A pilot does not create an obligation to renew, and a positive average in research does not override an unsuitable local result.

Complete a fictional workplace evaluation

All people, records and results here are fictional. Nell leads a small cultural programme. Omar coaches her individually. Ivy sponsors a finite pilot; Tess and Bo participate in the programme's regular decision meetings. The measured pilot runs from 3–21 March 2070, Europe/Amsterdam.

The practical goal is clearer acceptance of work, not merely more names in a record. The pre-agreed indicator is a named owner, a proposed task deadline and explicit acceptance within twenty-four elapsed hours after the meeting ends. The target is at least five of the next six eligible decisions. A refusal or unanswered question is recorded honestly and does not become consent.

Set the denominator, boundaries and starting account

Nell, Omar and Ivy have a twenty-minute kickoff on 3 March. They agree to use all six decisions in the three preceding meeting records as the baseline and all six eligible decisions in the next three meetings. This is a small descriptive series, not a randomized study or a representative sample of the organization.

Nell's baseline coding finds explicit acceptance documented for two of six decisions; the remaining four are unconfirmed in the available records. Missing documentation is not proof that those people refused or did nothing. The baseline result is therefore “2/6 confirmed,” not a diagnosis of their commitment.

The pilot includes two thirty-minute coaching sessions, three existing twenty-minute team meetings, the agreed record work and a final evaluation. Ivy authorizes and pays the fictional £320 fee within her existing £350 pilot allocation. Her preparation time includes the agreement and payment handling. The fee is a pilot cost, not evidence of value created.

Tess and Bo each spend ten minutes reading and accepting the defined evaluation participation, record audience and their current time allowance. That agreement includes regular note handling and the brief feedback request. It does not pre-accept any operational task or make a yes response compulsory. The five-of-six target is a workflow indicator, not permission to pressure a colleague into ownership.

A new mandatory proposed-owner field is introduced in the meeting form at the same time. They record that simultaneous process change before observing results. It prevents later attribution of every difference to coaching. The business records contain task states, not private coaching interpretations.

Deliver the coaching and correct an actual false acceptance

Omar and Nell meet for thirty minutes on 3 March and thirty on 17 March, with ten minutes preparation per person for each. The first session distinguishes naming someone from obtaining their agreement. Nell prepares a short question: “Can you accept this defined task and proposed date, or what needs to change?”

In the 4 March meeting she nevertheless records Tess as accepting a photography-permission task without checking. Tess replies after reading the note:

I did not accept that task. I do not hold the permission decision, and being named in the form does not give me that authority. Please remove the accepted status. I can explain the boundary, but I am not taking ownership of obtaining the permission.

Nell's fifteen-minute correction exchange covers reading the response, checking her note, correcting the status and sending a substantive answer: “You are right. I recorded acceptance that you did not give. The record now says proposed to Tess, declined, no assigned owner. I will not treat that as a delivery commitment. The permission question remains unresolved.” Tess's ten-minute allocation covers writing her correction, reading Nell's answer and checking the amended record.

The second coaching session examines this actual failure and the other emerging records. It does not redefine refusal as poor participation. Identifying an unauthorized assignment is useful information even though it does not meet the chosen acceptance indicator.

Finish the complete six-decision record

The team meetings run on 4, 11 and 18 March, each 10:00–10:20. The acceptance cutoff for each pair is therefore 10:20 the next calendar day. All proposed task-delivery dates are 24 March at noon, after the evaluation. The pilot measures agreement to responsibility; it does not observe those later task deliveries.

Six proposed-owner names appear, but only four decisions have the required acceptance evidence. The result is 4/6, or about 66.7%, against a target of 5/6, about 83.3%. The original target is missed by one confirmed decision.

Compared with the earlier 2/6, about 33.3%, the recorded proportion rises by about 33.3 percentage points. That descriptive increase is not a 100% causal coaching gain. The periods may differ, the mandatory field changed and undocumented earlier acceptance could have occurred. No control group isolates coaching.

Include participant experience and make the actual decision

Tess and Bo each spend five minutes providing the agreed feedback. Tess says, “The corrected record is useful, but naming someone before resolving their questions creates pressure. Leave time for clarification before calling it agreement.” Bo says, “The last wording question is still unanswered. A name in the form does not settle who can approve it.”

Nell rates usefulness at four out of five on a single original feedback item. That is her experience, not a validated outcome scale or evidence that the six-decision target was met. Omar has delivered both agreed coaching sessions; attendance is also a separate fact.

At the thirty-minute evaluation on 21 March, Nell, Omar and Ivy review the complete record, costs and participant concerns. Ivy decides to finish the paid pilot without automatic renewal: “There is a clearer account and a real correction, but the target was not met and two decisions remain without acceptance. I am not buying further coaching now. Any revised meeting process or additional work needs a separate authorized arrangement.”

Nell then spends ten minutes issuing the business closure; Ivy spends five minutes reading it. The completed summary states:

Pilot closed after two agreed coaching sessions. Confirmed ownership within the defined clock increased from 2/6 recorded baseline decisions to 4/6 pilot decisions; the 5/6 target was not met. One false acceptance was corrected after Tess's refusal. Bo's authorization question remains outstanding. The proposed-owner field changed concurrently, so no isolated coaching effect is estimated. Nell reported usefulness of 4/5, while participant feedback identified pressure and unresolved clarification. The £320 fee is a cost; no financial benefit or ROI was measured. Further coaching is not renewed. The four accepted tasks have future delivery dates; this evaluation does not establish completion, permission or publication approval.

Omar separately spends six minutes returning individual coaching feedback to Nell: “Your record now distinguishes a proposal, acceptance, refusal and an unanswered question. Keep the correction visible. The useful local learning is compatible with a missed target and a decision not to renew. It does not establish why the proportions changed or predict your next six decisions.” Nell spends four minutes reading it. No additional acknowledgment event is claimed.

Reconcile the pilot's actual effort and cost

The ledger includes coaching, implementation meetings, record handling and evaluation. Earlier meetings are baseline context. Future task delivery and any later process revision are not counted as completed or treated as free work.

Total effort is 740 person-minutes against 770, leaving thirty unallocated. The three team meetings alone use 180 person-minutes. Regular note handling includes the recipients' routine answers or questions and excludes the separately counted correction exchange. Ivy's £350 allocation less the £320 fee leaves £30 unspent, not a measured return. Neither spare minutes nor satisfaction can substitute for a financial-benefit estimate.

The 740 minutes are the observed effort in the defined pilot, including regular meetings that would otherwise have happened. They are not an estimate of 740 additional minutes caused by coaching. Estimating incremental cost would require an appropriate account of the alternative, changes in ordinary work and the value assigned to time. The fee alone is also not the total economic cost of the engagement.

Similarly, an unallocated minute is not a measured minute saved. No evidence here establishes that clearer records reduced rework or generated income. The sponsor has enough information for the bounded renewal decision while leaving those financial questions unanswered. A practical decision can be justified without inventing a precise return.

Use Coachful without mistaking activity for evidence

1. Record the agreed purpose and review rule. An original assigned worksheet can hold the goal, opportunity definition, indicators, other influences and review decision. Clients use Reflections, Forms & worksheets; verify the actual route with a synthetic response.
2. Track actions with precise states. A goal-linked task can record preparing a request or completing the evaluation. Completion does not establish an external owner's acceptance, a business result or a causal effect.
3. Return client feedback deliberately. Questionnaire Coach review is private to the authorized coach team and excluded from the response PDF. It is not client delivery. Use the individual client profile's message icon for the actual private return, as Omar does.
4. Separate evaluation audiences. Keep the business summary distinct from personal coaching material. Journal pages explicitly shared with the coach differ from unshared pages. Agree actual access and purpose rather than assuming sponsor funding permits every disclosure.
5. Check recap settings. Automatic recaps are the default. Approval can hold new unsent recaps, session-type overrides matter and already sent content is not recalled. Check handling before recording sensitive material.

The progress and accountability guide develops how an ordinary action record can support a useful review while retaining its limits.

Troubleshoot common evaluation errors

Only the favorable examples are retained. Return to the original opportunity and selection rule. Include refusals, unknowns and missed targets with appropriate explanations. Do not silently shrink a denominator to make the programme appear successful.

A null comparison becomes proof of equal methods. Check subgroup size, intervention descriptions and what the analysis could detect. No significant difference does not establish that formats or approaches are interchangeable for every purpose.

Positive feedback becomes a renewal requirement. Treat experience as one source of evidence. A client can value coaching and decide that the next need is instruction, authority clarification, a process change or simply an ending.

The buyer wants a guaranteed return. Explain the actual evidence and missing inputs. Offer a proportionate review where suitable, without promising that a small observational pilot will establish causality or financial payback.

Frequently asked questions

Does research support workplace coaching?

Several meta-analyses find positive average effects under their inclusion criteria. That supports considering coaching for relevant development needs, with attention to fit, measures, study quality and individual variation.

Does g of 0.74 mean 74% improvement?

No. The cited historical estimate concerns goal-directed self-regulation and is standardized relative to variability. It is not a universal effect, a percentage success rate or a financial return.

Do more sessions necessarily produce a better result?

The cited moderator findings do not establish a universal dose rule. Neither a guaranteed benefit from more sessions nor a claim that additional sessions never matter follows. Agree a proportionate service and review the actual need.

Can a small client pilot prove coaching caused the change?

Usually it supports a descriptive account and a practical decision. Isolating causality requires an appropriate comparison and design. Record other changes and avoid giving a local series more inferential weight than it carries.

Can a useful evaluation end without renewal?

Yes. Completing the agreed work, identifying a limitation and deciding to stop can be responsible outcomes. Report the actual decision and what remains unknown rather than converting encouraging research into an obligation to continue.