How to run a contest your agents will trust

Kouros Jelveh, EKPK LLC · October 23, 2026

A contest that agents suspect is rigged, lucky or pointless is worse than no contest. It costs the prize, the supervisor's hour, and some of the trust the next one would have needed. The rules below are the ones Arena enforces in its challenge engine, and each exists because of a contest that went wrong somewhere.

1. Measure more than one thing, and weight it

A race on contacts handled is a race to rush. A race on evaluation score alone can be won on a single lucky review. Arena's races take one or two measures with weights, for example evaluation average weighted two and customer sentiment weighted one, and rank agents on a normalised score from both. Genesys allows up to ten measures in its contests for the same reason. Two well chosen ones are usually enough.

2. Set a minimum qualifier

An agent who handled three contacts and scored 100 on one evaluation should not beat an agent who handled sixty and averaged 93. A minimum number of contacts to be ranked, three or five for a short race, removes the one-call wonder without excluding anyone who did the work. Arena shows unranked agents their own figures with a note on what they need to qualify.

3. Prizes by place, not winner takes all

Gold, silver and bronze keep the middle of the board engaged in the last two days, when a single prize has already been decided. Arena pays tiered prizes as points, once, when the challenge ends, with a line in each winner's feed.

4. Hide the names until the end, when it helps

Anonymised standings sound odd until you have watched a race where the leader is obvious from day one and everyone else stops trying. Arena can show each agent their own position and the rest as ranks only, with names revealed when the race closes.

5. Freeze the results and let a supervisor override

Results must be final the moment the challenge ends, computed from the same data the whole race used, and visible afterwards exactly as they were. Arena freezes standings on close and pays prizes from the frozen copy. A supervisor can disqualify an agent before the end, for a reason that is logged, and reinstate them. Nobody edits a finished result.

6. Give the prizes a budget

Points become rewards, and rewards cost money. A monthly reward budget per team, with the console refusing an approval that would exceed it and a note to the channel at 25, 50, 75 and 100 percent, is what lets finance say yes to the programme in the first place. Spot awards from supervisors, capped per award and per day, fit inside the same budget.

The rule behind the rules

Scores come from platform data and nothing else. No self-reporting, no manual adjustments, no points for activities the system cannot see. The moment a leaderboard can be influenced by anything other than doing the work, the floor knows, and the contest is over whether or not it is still running.

Arena's challenge templates, contest rules and reward budgets are described in the help center. The console also suggests the next contest from the team's own numbers, so the first one you run is the one the data says you need.