Superintelligent Agents for Referee Assignors: Schedules and Payouts
A referee assignor can use agents to flag schedule conflicts and track payouts, while the assignor keeps final say on every assignment.

Superintelligent agents for referee assignors are AI agents with a defined role, memory and tools that watch game schedules for conflicts, travel gaps and availability problems, and that reconcile assignor payouts against the approved fee schedule and the games actually worked. The assignor stays the decision maker. The agent drafts options and flags problems, but nothing changes a crew assignment or releases a payment without a named person approving it. This article explains where that boundary sits and how to set it up.
Key takeaways
- Start with one narrow role, such as a schedule conflict monitor or a payout reconciler, rather than a general assistant.
- The agent proposes and the assignor approves every assignment change and every payout batch.
- Payout tracking needs a versioned fee schedule and clean assignment records before any automation begins.
- Permissions should be read-only by default, with write access limited to a review queue.
- Success is measured with a few named checks, reviewed on a regular schedule, not judged by impression.
Where a referee assignor loses time
Most assignor offices do not lose time on the assignment itself. They lose it in the gaps around it: a referee marked available who is already booked in another league, a travel window that looks workable until the drive between venues is counted, a venue change posted the evening before, or a withdrawal that reaches the inbox but never reaches the schedule. Each gap is small. Together they produce last-minute replacements, double bookings and a long chain of messages to repair them.
Payouts create a different kind of drag. Fees depend on the level of game, the role worked, the structure of a tournament day, travel stipends and adjustments after cancellations or late swaps. When those rules live partly in a fee sheet, partly in a spreadsheet and partly in someone's memory, closing a pay period becomes a manual check that repeats every cycle. Each disputed line costs some of the trust you depend on with the people you assign.
Which role to build first
Start with one role that has a clear owner and a clear output. Use this guide to choose.
- If crews are often notified and then have to be changed, start with the schedule conflict monitor.
- If pay periods close with manual checks and recurring disputes, start with the payout reconciler.
- If both problems are painful, start with the one whose mistakes are hardest to reverse. Once money has moved, a wrong payout is usually the costlier error.
- Do not start with an agent that assigns crews on its own. The assignment is the judgment you keep.
Either way, the first role should be narrow enough that you can describe its inputs, outputs and failure cases on one page. That page becomes the brief for the engagement.
Schedule conflicts: what the agent checks and what it never changes
A conflict monitor reads three things: the game schedule, the availability and eligibility record for each referee, and the rules your office applies. It checks for overlapping windows, insufficient travel time between venues, blackout dates, level or position requirements, limits on games per day and pending withdrawals. The output is a short review queue with one line per flagged game, the reason, and the source record behind it. The agent does not edit the assignment or message referees. It may draft a replacement from the eligible roster, ranked by the rules you wrote, and leave the decision with you.
Travel time is the check most often left out and the one most often wrong. Keep a small table of drive times between your venues, maintained by a person, and have the agent read it instead of estimating. Where the table has no entry, the agent should flag the gap rather than guess. Scheduled-service businesses see the same pattern with event-day changes, as described in the note on DJ bookings, gear holds and event-day changes.
A practical first version looks like this:
- Export the next four weeks of games, the current availability list and the eligibility rules into one folder the agent can read.
- Write each rule as a plain statement, such as a minimum travel time between two venues or a blackout period a referee has requested.
- Define the output as a daily review queue showing the game, the reason, the source record and up to two replacement candidates.
- Grant read access to schedule and availability records, and write access to the review queue only.
- Run the monitor alongside your usual process for two weeks, and record every case where its flag differed from what you caught by hand.
- Review each missed flag and each false alarm, change the rules that caused them, and only then consider a wider rollout.
Payout tracking without guesswork
A payout reconciler needs one source of truth for fees. That means a versioned fee schedule that states which rate applies to which level, role and day type, with the date each version took effect. Without it, the agent can only compare numbers against other numbers. With it, the agent can compare what was assigned, what was worked, what was reported and what is about to be paid.
Typical findings include a game with an assignment but no game report, a stipend with no matching travel record, a rate that does not match the schedule version in force on the game date, and a duplicate line created when an export ran twice. Each finding goes into the review queue with the source lines attached. The assignor or the person who runs payroll decides whether to correct the line, request information or accept it.
Duplicates deserve particular care. Payment APIs often protect against retries with an idempotency key, a unique identifier that makes a repeated request count only once. Stripe documents this pattern for its own API. Your reconciliation should work the same way: a batch identifier for each pay period, so that a re-run updates the same batch instead of creating a second one. Disputes over money follow a similar logic in other settings, as covered in the note on marketplace trust and disputes.
Illustrative scenario: On a tournament weekend with a full slate of games, the reconciler finds one referee with no report for a midday game and a travel stipend entered twice in the export. It places both items in the review queue with the source lines attached. The assignor confirms the stipend is a duplicate, asks for the missing report, and approves the corrected batch. Only that approval moves the batch forward to the person who releases payment.
Permissions and approval boundaries
Give the agent the smallest set of permissions that lets it do its job, and deny everything else by default. The OWASP authorization guidance makes the same point for any system: access should follow an explicit rule, not a convenient shortcut. In practice, the split looks like this:
- The agent may read schedules, availability, eligibility rules, fee schedules and reports.
- The agent may compute conflicts and reconciliations and write notes to the review queue.
- The agent may draft a message to a referee or a finance contact, which a person sends.
- Only an assignor may change an assignment, and only a named approver may release a payout batch.
- No one delegates decisions about certification, safety or league-specific eligibility to the agent alone.
Referee messages, league emails and report notes are written by other people, and they may contain instructions unrelated to your process. Treat that content as data to read, never as commands to follow. The OWASP prompt injection guidance covers this risk in detail, and it is one reason the agent's tools should not act directly on the content of a message.
How Be Human's managed approach handles this
Be Human designs this kind of role as a first engagement: one defined role, the memory it needs (the rules, the fee versions and the travel table), the tools it may call, the approval boundaries, and an observable success criterion agreed with the assignor. Managed operations then cover monitoring of the review queue, recurring evaluations of how often flags were correct, permission reviews, integration maintenance when the scheduling or payroll system changes, and controlled updates to rules. The work is remote, in English or Portuguese, and the assignor keeps final say on every assignment and payout throughout.
Measuring whether it helps and choosing what comes next
Choose two or three measures before the agent runs. Useful ones include how many flags the assignor confirmed as real, how many real conflicts were found only after notification, how many payout lines were disputed after release, and how long the assignor spends closing a pay period. Review these on a regular schedule rather than once. The NIST AI Risk Management Framework treats ongoing measurement and human oversight as part of managing AI risk, and that approach fits a small office well. A missed flag is useful information: record it, understand it, and change the rule or the data that caused it.
Take the free Be Human AI diagnostic to find the first task worth handing to an agent.
Frequently asked questions
Can an agent assign referees on its own? It should not, and a well-run office keeps that step with a person. The agent can flag conflicts, rank eligible replacements and draft a proposed assignment. The assignor reviews the proposal, changes it if needed, and makes the assignment in the scheduling system, which keeps accountability clear when a league or crew member asks why a game was assigned as it was.
What information does an assignor need before an agent can flag conflicts? At minimum, you need the game schedule with venues and times, the availability and eligibility record for each referee, and the rules your office applies, written as statements rather than remembered practice. A travel-time table between venues is especially useful. If a piece is missing, the agent should report the gap instead of filling it with an estimate.
How does payout tracking work without the agent moving money? The agent compares assignments, game reports, stipend records and the fee schedule version in force on each game date, then places any mismatch in a review queue. It does not release payments. A named person approves each corrected batch, and the payment system or payroll process performs the transfer under its own controls.
How long before you know whether the first engagement is working? You learn sooner if the success criterion was chosen before the agent started. The early signals are whether the flags the assignor reviews are mostly correct and whether conflicts that used to surface late are now caught earlier. Expect to adjust rules in the first weeks, because a first rule set always reveals cases nobody wrote down, and Be Human reviews those results with you rather than promising an outcome in advance.
Your next step
See where an AI agent fits in your company first.
Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.
Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.


