Understanding the challenge
What we saw on the surface. A small, committed operations team doing hero work: onboarding new events, configuring workouts and divisions, generating heat schedules, coordinating judges, fielding frantic last-mile questions. Payments added complexity. Stripe where available, fallback collection and later bank transfer in some markets, plus refunds or chargebacks to triage.
What sat underneath. Sensitive personal data meant GDPR obligations were real, not theoretical. DOBs, phone numbers, emergency contacts, socials. DSARs, retention rules, access controls. These were handled with best intentions and shared docs, but little automation. During event peaks, people worked around process because the show had to go on. Risk crept in through manual work, duplicated spreadsheets, brittle handoffs.
Our approach
We started with an AI Readiness & Efficiency Audit to map how work really happened. Who touched what, when, with which data. Then we designed a narrow, high-yield automation layer around first-party systems and structured handoffs. No "magical agents" scraping the open web, no rewrites of the product. Just better rails for the work people already did. In other words, we kept our promise to be useful, not breathless.
We broke the problem into three loops: pre-event (onboard, configure, preflight), live-day (runbook, triage, updates), post-event (reconcile, report, retain). For each, we set clear owners, guardrails and a plain-English definition of "done."
Technical innovation
Ops Command Center. One intake form, one source of truth. Each organiser request spun up a templated project: checklists, docs, SLAs, a Slack channel. A lightweight routing bot flagged risk (new country, no Stripe, over 500 athletes) and assigned an owner. Where Stripe wasn't available, the flow generated the right bank-transfer paperwork and an escrow ledger, then triggered a controlled post-event settlement.
Support Copilot, human in the loop. All conversations (email, in-app, social) landed in one helpdesk. A small language model, grounded in Circle21's policies, FAQs, event data, drafted replies and suggested next actions. Refund? Heat change? Judge access? Agents approved with a click. Ambiguous cases went to a senior triage queue. Less heroics, more rhythm.
Finance Ops & reconciliation agent. Stripe balance transactions matched to event orders produced per-event P&L, including fees and currency effects. Exceptions (chargebacks, late refunds, non-Stripe settlements) were flagged and assigned. A close checklist kept approvals clean and produced an immutable summary for audit.
Governance & data hygiene. Role-based access, PII redaction in tickets, a guided GDPR flow (verify, locate, export/redact, confirm). Every action logged, so the story is there when someone inevitably asks.
The Impact
- Support resolution dropped 83% from 18 hours to 3 hours, with first-contact resolution up 24 points.
- Organiser onboarding shortened by 58%, from 7 business days to 3, through templates and risk-based routing.
- 92% of transactions auto-reconciled. Month-end close fell 72%, from 3 days to 8 hours.
- 1.4 FTE capacity freed across ops and finance, redeployed to organiser success work.
- Zero SLA breaches on peak weekends in the final two pilot months.
