How to Create Random Teams That Fit the Job

Random teams are useful when the main goal is to divide people quickly without letting self-selection decide the groups. They are not the right answer to every grouping problem.

The first question is: does this activity benefit from randomness, or does it need deliberate team composition?

Cornell’s teaching guidance reflects that trade-off. Random assignment is straightforward and can work well when groups change regularly. Longer-term or more consequential teams may need deliberate decisions about skills, availability, support needs or other factors.

A random team generator can remove one source of human choice. It cannot know which mix of experience, confidence, accessibility needs or personalities will work best.

Start with the size rule

There are two common ways to build teams:

Those instructions are related but not identical.

Eighteen people split into four teams gives sizes of 5, 5, 4 and 4. If the requirement is instead “no more than five people per team”, four teams are also enough. With 17 people and a maximum of four per team, five teams are needed.

When the numbers do not divide evenly, some teams will contain one more person than others. A size-balanced split keeps that difference to no more than one where the requested constraints allow it.

Random does not mean skill-balanced

A random team generator can randomise who goes where. It cannot infer who is the strongest player, most experienced colleague, quietest student or best organiser unless that information is deliberately supplied and used.

GWS does not describe its normal random groups as skill-balanced teams.

The Spread marked people control is narrower. You choose which people are marked, and the generator tries to distribute those marked entries across teams while respecting the hard rules. The mark itself does not mean “high skill” or “low skill”.

If the real aim is to balance ability, role, experience or support needs, use a process built around those criteria rather than treating a random split as equivalent.

Keep Together and Keep Apart

Pure randomness is sometimes too simple.

You may need two people together because they share equipment or are working on the same subtask. In another activity, you may want two people apart because the purpose is to create different pairings.

These are constraints, not weights.

A generator should either satisfy a hard constraint or say that the requested combination cannot be made. Quietly dropping a rule produces a result that looks valid but is not the result the user asked for.

For example, a Keep Together block may be larger than the biggest available team. Two people might also be required both together and apart. Those requests cannot be solved without changing something.

Pairs need an odd-person rule

Pairing is a special case.

With an even number of people, everyone can be placed into pairs. With an odd number, the organiser has to decide what should happen.

One option is a single group of three. Another is one person sitting out for that round.

Neither rule is automatically better. The tool should make the choice explicit because the activity determines which option makes sense.

Add only the constraints you actually need

Every extra rule reduces the number of legal arrangements.

Before adding a long set of Keep Together and Keep Apart rules, ask whether each one is necessary. A heavily constrained request may have very few possible solutions, or none.

For short, low-stakes activities, fewer rules usually make the process easier to understand. For a longer project, human review matters more because the software cannot assess all of the practical context.

When random assignment is a good fit

Random teams can work well for:

They are less suitable when group composition has important consequences for safeguarding, accessibility, assessment, specialist roles or long-term performance.

A practical workflow

  1. Clean the participant list. Include everyone who should take part and remove anyone who should not.
  2. Choose the size rule. Decide whether the number of teams or the maximum size matters more.
  3. Add only necessary constraints. Avoid rules that exist only out of habit.
  4. Generate the teams.
  5. Read any impossible-rule message. Do not keep regenerating in the hope that an impossible request will eventually work.
  6. Inspect the result. Check practical issues the software cannot know.
  7. Adjust openly if needed. A manual change for a legitimate reason is a human adjustment, not a random result.

Random assignment is valuable because it handles the part of grouping that genuinely benefits from chance. It should not hide the parts that still require judgement.

Sources and further reading

These sources discuss teaching-team formation. They do not establish that random grouping is best for every classroom, workplace or activity.

← Back to the wheel