Team building changes how teams feel more than how they perform

The first meta-analysis of team building found no significant effect on performance. Published in 1999, it went further: on objective measures, team building showed a slight and nonsignificant tendency to make performance worse. Ten years later an update from the same research group found a moderate positive effect, strongest on how members feel about the team and how they work together.

Two reviews with an author in common, a decade apart, with different answers. We read both abstracts side by side, then set them against the teamwork training meta-analyses, which ask a related question with stricter designs. As of September 2026 the picture is consistent, and more specific than "team building works".

What team building means in this research

In both meta-analyses, team building is an intervention on a team as a whole, aimed at how the team functions, and both sort it into the same four components. Goal setting agrees what the team is trying to achieve. Role clarification settles who is responsible for what. Problem solving works through the team's actual obstacles. Interpersonal relations works on trust, conflict and communication between members.

The four components are the reviews' own coding scheme. An activity sold as team building that contains none of them, a day of kayaking with no discussion of goals or roles, has no evidence base in this literature.

Teamwork training is a separate category. It teaches specific teamwork skills, such as briefing, monitoring each other's work or speaking up, and it is usually tested in controlled designs with a comparison group. The two categories overlap, and the evidence for each is different in quality.

In 1999, no significant effect on performance

Eduardo Salas and colleagues pooled the studies of team building that measured performance and published the result in Small Group Research. The headline was null.

Overall, there was no significant effect of team building on performance.

The detail underneath matters more than the headline. On subjective measures of performance, team building had a small significant positive effect. On objective measures, the tendency ran the other way and did not reach significance. The effect also shrank as teams got larger.

Of the four components, only role clarification was more likely to raise performance. Interventions built around goal setting, problem solving or interpersonal relations were no more likely to raise it than to lower it. The component closest to the popular image of team building, getting people to like and trust each other, was among those with nothing to show.

In 2009, a moderate effect, strongest on attitudes and process

Cameron Klein, Deborah DiazGranados, Salas and four others updated the 1999 review with a larger database and a wider set of outcomes, again in Small Group Research. They looked at cognitive, affective, process and performance outcomes, and at the same four components.

Based on 60 correlations, the abstract reports that team building has a positive moderate effect across all team outcomes. It was most strongly related to affective outcomes, meaning members' attitudes, and to process outcomes, meaning how the team works together.

The 2009 paper sits behind a paywall and the open repository copy at the University of Central Florida holds only the abstract. We could not read the tables, so this article does not report the paper's component rankings or its performance figure. The role clarification finding is stated plainly in the 1999 abstract. The 2009 abstract does not rank the components.

Read together, the two contradict each other less than their headlines suggest. The second review widened the outcomes to include feelings and process, which is where the effect is largest. On performance alone, the one abstract that states a result says there was none overall.

Teamwork training has the stronger case for performance

A second line of meta-analyses asks whether teaching teams specific teamwork skills improves results. Salas, DiazGranados, Klein and colleagues pooled 93 effect sizes representing 2,650 teams in Human Factors in 2008 and report moderate, positive relationships between team training and each outcome type, including performance. Training content, the stability of team membership and team size all moderated the effect.

The most rigorous review is the most recent. Desmond McEwan and colleagues at the University of British Columbia searched 16,849 articles and kept 51 that used controlled experimental designs: 72 interventions and 8,439 participants. Before outliers were removed, the effect of teamwork training on team performance was 0.92. After five outlying interventions were removed, it was 0.58, with a confidence interval from 0.47 to 0.69. The effect on teamwork behaviour itself fell from 0.68 to 0.55 on the same treatment.

Three of the five performance outliers, from two studies, were team building in the classic sense. One was the team-building condition of a 1986 field experiment, with an effect of 3.72. The other two were both arms of an appreciative inquiry team-development study from 1995, at 4.57 and 5.96. The reviewers flagged them in their sensitivity and residual checks and removed them. The largest numbers in this literature come from the studies least like the rest.

The table carries its own warning. More than half the interventions behind the 0.58 were laboratory experiments, and 39 of the 45 used newly formed teams. Only three came from industry and only six from intact teams, which is what a company actually has. The large figures for those two rows rest on so few studies that we would not quote either as a result.

What 0.58 would mean for 40 teams

Take a firm of 400 people in 40 teams, considering a programme at 1,500 euro per team, 60,000 euro in all. The firm and the price are an invented example.

If the controlled-study effect of 0.58 transferred to this firm, the average trained team would outperform about 72 percent of untrained teams. Put another way, about 29 of the 40 teams would end up above the performance of the median untrained team, against 20 without the programme. This is the optimistic case.

If the programme is team building in the older sense, built around interpersonal relations, the 1999 abstract finds interventions of that kind no more likely to raise performance than to lower it. The same 60,000 euro buys a measurable improvement in how the teams feel, which may well be worth having, and no demonstrated change in their output. The one component with a performance effect in the 1999 review is the least theatrical, working out who does what.

A parallel pattern appears in the evidence on leadership training, where programmes that combine information, demonstration and practice outperform those that only inform.

No review has tested online team building

We searched for a meta-analysis of online, remote or virtual team building and found none. Neither team-building review reports a moderator for delivery mode. The McEwan review includes, by their titles, at least four studies involving virtual or distributed teams, among them a 2014 study of self-guided training for conflict management in virtual teams. It does not analyse virtuality as a moderator, so no pooled figure for online delivery exists.

The two neighbouring literatures point in different directions. In the meta-analyses of coaching, remote and face-to-face delivery performed the same. In the leadership training meta-analysis, face-to-face programmes produced stronger transfer than virtual ones on one of four outcomes. Neither concerns a team working on itself.

The review that would settle online team building has not been done. Whether the effect on attitudes survives a screen, and whether role clarification works as well when the roles are negotiated in a shared document, are open questions with no data behind either answer.

Common questions

Does team building improve team performance? The evidence is weak. A 1999 meta-analysis found no significant overall effect on performance, with a slight nonsignificant negative trend on objective measures. A 2009 update found a moderate positive effect across outcomes, strongest on attitudes and team process.

Which kind of team building works best? Role clarification. In the 1999 meta-analysis it was the only component more likely to raise performance. Goal setting, problem solving and interpersonal relations were no more likely to raise it than to lower it.

What is the difference between team building and teamwork training? Team building works on a team's goals, roles, problems and relationships. Teamwork training teaches specific skills such as briefing, monitoring and speaking up, and has a meta-analysis restricted to controlled experiments.

How large is the effect of teamwork training? A 2017 meta-analysis of 51 controlled studies found an effect of 0.58 on team performance after outliers were removed, and 0.55 on teamwork behaviour. By the usual convention that is a medium effect.

Does online or virtual team building work? No meta-analysis has tested it. A few individual studies in the teamwork training literature involve virtual teams, but none of the reviews reports a pooled figure for online delivery.

Where do the very large effect sizes come from? The largest effects in this literature come from a handful of studies that reviewers treated as outliers, with values between 2.8 and 6.0. Once they are removed, the effect of teamwork training on performance falls from 0.92 to 0.58.

Sources

1. Salas, Rozell, Mullen & Driskell (1999), The effect of team building on performance: An integration, Small Group Research 30(3), 309-329
2. Klein, DiazGranados, Salas, Le, Burke, Lyons & Goodwin (2009), Does team building work?, Small Group Research 40(2), 181-222
3. Salas, DiazGranados, Klein, Burke, Stagl, Goodwin & Halpin (2008), Does team training improve team performance? A meta-analysis, Human Factors 50(6), 903-933
4. McEwan, Ruissen, Eys, Zumbo & Beauchamp (2017), The effectiveness of teamwork training on teamwork behaviors and team performance: A systematic review and meta-analysis of controlled interventions, PLOS ONE 12(1), e0169604