How to Create Balanced Teams

Decide what “balanced” means for you

Balance is not one thing. Equal headcount matters for a relay race; equal skill matters for a competitive match; a mix of experience matters for a project group; and for a friendly kickabout, unpredictability may matter more than either.

Pick the mode that matches your goal rather than reaching for the most sophisticated option. Score balancing on a list where nobody has a meaningful score just adds noise.

Fully random teams

In fully random mode each person is assigned to a uniformly chosen team, so sizes vary — sometimes noticeably. With twelve people and three teams you might get 5/4/3.

This is the right choice when variation is acceptable and unpredictability is the point: informal games, icebreakers, rotating pairs. It is the wrong choice when the activity depends on equal numbers.

Equal headcount

Equal mode shuffles the roster and deals each person to whichever team currently has the fewest members. Sizes end up differing by at most one, and because the shuffle happens first, which team gets the extra person changes every time.

This is the default for a reason: most of the time, equal numbers is what people actually mean by fair teams.

Score-based balancing

Give each person a score — “Ava: 8” — and the generator sorts them strongest first, then hands each one to the team with the lowest running total. It is a greedy algorithm, and it works well: on typical rosters it lands the team totals close together in a single pass.

Two details matter. First, people with equal scores are shuffled before sorting, so equally rated players do not get locked into the same team every draw. Second, greedy is not optimal — finding the genuinely best split is a partitioning problem that becomes expensive fast, and the generator does not attempt it.

Use scores that mean something on a consistent scale, and remember that the totals shown are the sum of numbers you supplied. They are only as good as your ratings.

Spreading leaders

Mark leaders with an asterisk — “*Ava: 8” — and tick “spread leaders”. Leaders are shuffled and dealt one per team before anyone else is placed, so each team gets at most one until they run out.

If there are more leaders than teams, the extras are placed like everyone else and the tool tells you. If there are fewer, some teams have none, which is also stated rather than quietly patched over.

Keep together, keep apart

Keep-together groups are merged into a single unit before the deal, so they always land on the same team. This is the constraint most likely to skew sizes: a keep-together group of four in a five-person team leaves very little room.

Keep-apart rules filter which teams are eligible for each person as they are placed. When no eligible team remains — which happens easily on small rosters with several rules — the person is placed anyway and the unmet rule is listed with the results. The generator will not spin forever trying to solve an impossible constraint set, and it will not silently drop a rule either.

The limits worth stating out loud

No algorithm can promise balanced teams, because balance depends on things a list cannot express: who plays well together, who had a bad week, who has been in this group before, and how the activity itself rewards different strengths.

Treat the generated split as a good first draft. Look at it, move two people with the arrows or by dragging, and run with it. The value of the tool is that it removes the awkward negotiation and the accusation of favouritism — not that it produces a mathematically optimal answer.