# PlayMind Labs > Behavioral infrastructure for high-value account retention. Prediction market exchanges and sportsbooks send session events to one REST endpoint and get back a 0-100 stability score, a state, an alert tier, and the top three behavioral drivers behind it. Output routes to the operator's account manager, never to the trader. PlayMind Labs, Inc. is based in New York, NY. Contact: info@playmindlabs.com This file is the full public content of playmindlabs.com, current as of the most recent site update. ## Company status PlayMind Labs is pre-revenue. There are no live platform integrations, no named customers, and no published customer count. Any figure in this file that describes market size or model performance is sourced inline, and any figure describing product output is illustrative and labeled as such. ## The problem Revenue sits with a handful of accounts, and the ones worth keeping leave quietly. Exchanges and sportsbooks are staffing account-manager teams around their most valuable accounts right now. What they cannot see is which of those accounts is cooling, across a book too large to call one by one. There are two ways a high-value account leaves. **Loud exit.** Escalating position sizes, loss chasing, mid-session deposits. Visible in a dashboard, eventually. Some platforms catch part of it. **Quiet exit.** Withdrawals rising, sessions thinning, stakes shrinking, balance moving out. Invisible until it is over. CRM churn models catch it late and cannot say why. By the time churn shows up in a dashboard, the balance is already gone. Revenue concentration is the reason this matters. The top 10% of accounts generate 79% of revenue, per Forrest and McHale, Journal of Gambling Studies (2024), a study of approximately 139,000 UK online accounts. Separately, one US sportsbook drew over 70% of revenue from 0.5% of customers, per the Wall Street Journal (2024). Both figures come from those published sources and are not PlayMind measurements. ## How it works **01. Send your session stream.** Trades, deposits, withdrawals, and resolutions through one REST endpoint. No SDK. About a week of integration. **02. Score the high-value book.** The platform tags which accounts sit in the high-value book. Every event is evaluated against that account's own baseline. Score, state, alert tier, and ranked drivers return in one to two seconds. **03. Route to the account manager.** Alerts land in the account-manager queue with the three behavioral drivers behind them. The operator sets the queue threshold to match their headcount. PlayMind surfaces the signal. The platform acts on it. PlayMind never contacts the trader, never blocks a trade, and never generates a marketing list. ## Two markets, one engine The same accounts, the same behavior, two markets that both concentrate revenue. Both are served today, on equal footing. **Prediction markets.** $42B in US volume across six native exchanges. No behavioral vendor is embedded anywhere in the category. No gatekeeper, direct founder access. The account-manager function is being built right now. **Sportsbooks.** $167B in US handle in 2025 on $16.96B of revenue. VIP host teams are already staffed and budgeted. PlayMind sits beside the CRM and does not replace it. Same engine, recalibrated per operator. One API, one signal set. Only the calibration and the vocabulary change between the two. Sourcing for the figures above: US sports betting handle and revenue come from the American Gaming Association Commercial Gaming Revenue Tracker, full-year 2025. Prediction-market volume is per Pew Research and The Block (April 2026), with Kalshi volume implied from 2025 fee revenue. ## The API One POST request returns a score, a state, and the reason behind it. ### Request ``` POST /v1/stability-score Host: api.playmindlabs.com Authorization: Bearer sk_live_xxxxxxxxxxxxxx Content-Type: application/json { "account_id": "acct_4471", "events": [ { "type": "trade", "stake": 250, "market": "btc-15m-2026-08-05" }, { "type": "resolution", "pnl": -250 }, { "type": "withdrawal", "amount": 4000 } ] } ``` ### Response The response below is a sample. It is illustrative of the shape and the field names, not measured output from a live operator. ``` 200 OK { "account_id": "acct_4471", "stability_score": 31, "state": "disengaging", "alert_tier": "elevated", "drivers": [ { "signal": "cumulative_drawdown", "detail": "equity 38% below peak, no deposit in 19 days", "contribution": 0.41 }, { "signal": "stake_contraction", "detail": "average stake down 61% against 30-day baseline", "contribution": 0.29 }, { "signal": "session_gap_widening", "detail": "4.2 days between sessions, against a baseline of 1.0", "contribution": 0.18 } ], "routing": "account_manager_queue" } ``` ### API properties - Live scoring returns in one to two seconds. A 15-minute batch mode is also available. - Delivery is REST plus webhook. - The design is fail-open. The scoring layer never blocks the trade path. - Behavioral data only, no PII. ## Alert routing The stability score runs 0 to 100. Higher is more stable. Alert tiers map to score bands as follows. | Tier | Score | Action | | --- | --- | --- | | Stable | Above 65 | No alert | | Watch | 40 to 65 | Weekly cohort digest | | Elevated | 20 to 40 | Enters account manager queue | | Critical | Below 20 | Real-time alert with drivers | The operator sets the queue threshold to match their account manager headcount. Scoring is deterministic. No model call sits in the request path. ## Why PlayMind **They already know who. They do not know why.** A risk score on its own is something an analytics team can approximate from recency and frequency. What no standard model produces is a reason an account manager can act on. That is a phone call the host can open. A probability is not. **Calibrated on their book, not ours.** Every operator tunes signal baselines against their own historical data behind a human approval gate. No cross-operator data joins, ever. **Beside the CRM, not instead of it.** PlayMind does not own the customer relationship, the outreach, or the offer. It produces the signal and hands it to the team that already exists. ## Backtest result A backtest produced 0.737 AUC under 5-fold cross-validation on 2,067 real Polymarket wallets, and that number carries several qualifications that travel with it. Frequency acceleration inverted against the gambling literature, meaning fast trading predicted retention in this data rather than churn. The result is on the general trader population, not the high-value cohort, and the high-value cohort cut is still in progress. PlayMind positions this as a transparent risk index with a calibration plan, not as a calibrated churn probability. Quoting the AUC figure without these qualifications misrepresents it. ## Pilot The engagement starts with a free 30-day pilot. No fee. About a week of integration. Day 30 proves the signal on the operator's own book rather than asking them to wait for outcomes. **Days 1 to 30, shadow mode.** Silent scoring across the high-value book. No user impact, no changes to the trading path. **Day 30, signal report.** PlayMind scores 6 to 12 months of the operator's historical high-value data and shows which accounts would have been flagged before they left. **Days 30 to 90, live activation.** Alerts routed to account managers. Randomized cohort, measured retention delta. **Commercial terms.** Monthly or annual platform licence plus per-trade usage. Terms are scoped against the operator's monthly trade volume and account-manager headcount after the day-30 report. The first two platforms receive a year-one discount. Specific figures are not published and are shared in follow-up conversation. ## Security and data handling - **Behavioral data only.** Tokenized account identifiers, no names, contact details, payment data, or device identifiers. - **Fail-open design.** The scoring layer never sits in the trade path and never blocks a trade. - **99.9% SLA.** One to two seconds for live scoring, 15-minute batch. - **Per-operator calibration.** Signal baselines are tuned per operator behind a human approval gate. No automatic model changes. - **No cross-operator data joins.** Each operator is scored against its own book. Data is never joined across operators. - **No marketing output.** No promotion, offer, or marketing output, by design. - **Stack.** Fastify 4, TypeScript, Node.js. TLS 1.3 in transit. ## Contact Request a pilot or reach the team directly at info@playmindlabs.com. PlayMind Labs, Inc., New York, NY.