Software That Watches for Harm

A Leeds-based gambling analytics firm has raised £6.4 million to scale a harm-detection tool it sells to UKGC-licensed operators. Harbour Analytics closed its Series A this week, led by a Manchester growth fund, with existing backers following on. The company employs 31 people, most of them data scientists and engineers, and counts six licensed UK operators among its clients. The money pays for a bigger team and a push into more casino lobbies before the end of next year.

"Operators can spot a problem account in a report three days after the fact," co-founder Priya Raghavan said. "We want to move that window down to minutes and hand the decision to someone who can act on it."

Harbour builds monitoring software. Its models read a live session, weigh stake size against deposit history, session length and the pace at which a player taps through a slot, then rank the accounts that look like trouble. Operators get a dashboard and a queue of names.

Judging trouble starts with the maths behind the games. A player who never learns how slot RTP works, or the volatility rating attached to a title, can misread a bad run of spins as a rigged machine rather than ordinary variance. Harbour's pitch is that the same numbers tell an operator when someone has stopped playing for fun. Four hours on a high-volatility title with stakes climbing is a pattern the software treats as a red flag, and a pattern a support agent misses until the next morning.

Take a customer who deposits £30 on a Friday, plays for 40 minutes and logs off. The model leaves her alone. A second customer deposits £400 across six payments between 11pm and 2am, spins through 900 rounds on one title and loses £310 of it. The dashboard puts her at the top of the queue with a suggested action: a call, a limit offer, or a cooling-off period.

Why operators buy monitoring software

Gambling Commission rules already require operators to monitor play and step in when they spot harm. Updates to the Licence Conditions and Codes of Practice tightened that duty: firms must intervene earlier and show the Commission what they did. Fines for falling short have reached seven figures, and the regulator has funded its own research into how operators use the data they hold. Casework on firms that failed gives compliance officers a template for what to avoid.

Manual review does not scale. A mid-sized casino with 200,000 UK accounts cannot put a person on every session, so most sites triage with static triggers: a deposit over £500, a session past two hours. Those rules miss the customer who deposits £50 twelve times in a week. They also flag the high-roller who can afford every penny of it, which wastes an agent's afternoon and annoys a profitable account holder. Static triggers stayed popular for a decade because they were cheap and easy to explain in an audit.

Machine learning sorts some of that mess, though the sector still argues about what works. The push towards AI tools for safer gambling has produced false alarms alongside genuine catches, and the Commission wants evidence that a model's output changes what an operator does. A model that flags accounts and changes nothing is a document, not a control.

The false-positive problem

Any harm-detection tool that fires too often gets switched off. Support teams drown in alerts, players get asked for documents mid-session, and the software loses credibility with the people expected to use it. Agents who stop trusting an alert stop reading it. Harbour says its false-positive rate sits under 12 per cent across its live operator base, a figure the company will need to defend when its clients face Commission audits.

Models drift as well. A title that trends in March behaves differently in December, and a training set from last year misreads a new game mechanic. Data creates a second problem. Operators keep deposit records, game logs and interaction notes in systems that do not talk to each other, and a model trained on slot data struggles on a sportsbook where the same customer bets in five-minute bursts on a Saturday afternoon. Cleaning up that plumbing costs more than the software licence.

What a player sees

If you play at a UKGC-licensed site, you meet this software through a nudge. A pop-up after a long session, an email offering a deposit limit, a call from a safer gambling team. Sites that handle this well keep the conversation short and the controls easy to find. An NV Casino review shows what a UKGC-licensed operator puts in front of a new player: deposit limits set at sign-up, a link to GamStop, and a reality check that interrupts play after a set period. Fewer sites bury those behind three menus than they did five years ago.

The tools work when someone uses them. Deposit limits, time-outs and GamStop self-exclusion do their job once a player sets them, and software can prompt an operator to offer those options without making a customer accept them. GamCare and BeGambleAware take the calls that follow. A player who sets a monthly limit at sign-up sees no prompt at all, which is the point.

A crowded field

Harbour competes with open banking providers, payments firms and platform suppliers that bundle player-protection analytics into wider contracts. Two rivals give it away with a deposit-processing deal. The specialists argue their models are sharper because they do nothing else, and that argument holds until a client compares invoices. Harbour charges per monitored account, which keeps the entry price low for a small operator and steep for a large one. No contract runs past 12 months, so a client can leave after a bad quarter.

Investors see a market with a regulator behind it. The Commission's demand for evidence turns compliance work into a budget line, and operators that need a paper trail will pay for software that produces one. Harbour's model runs on data its clients already hold, which shortens a sale to weeks rather than the year a full platform migration takes.

Detecting harm early does not remove it. A ranked list still leaves an operator deciding whether to ring a customer at 11pm, and either call carries a cost: upset a good customer, or miss a bad night.

The founders now have a year to prove their models travel between products and markets. Harbour says it will publish detection benchmarks from its live operator base and let clients audit the results before the next funding round. Operators will buy on those numbers, and the Commission will read them too.