Same deployment, three numbers: Why iGaming AI support rates don’t compare

Vendor figures built on different definitions, ticket sets, and timeframes cannot be compared until the industry agrees a shared standard with per-category reporting.

Every AI support vendor in iGaming now leads with a percentage. Seventy per cent automated. Eighty per cent resolved. Ninety per cent handled without a human. Operators are being asked to choose between vendors on numbers that, in most cases, measure different things.

I run a company that sells AI player support, so I have an interest here. That is also why I think the industry needs a shared way to read these figures before procurement teams start treating them as comparable.

Three numbers that get called ‘resolution’

Containment counts conversations that never reached a human agent. A player who gives up and closes the chat counts as contained. So does a player who asks the same question again the next day.

Automation rate usually counts conversations where the AI produced every reply. It says nothing about whether the player’s problem was fixed.

Resolution should mean the issue was fixed and the player did not come back about it within a set window, typically three to seven days. Very few published figures use this definition, and fewer still state the window.

The same deployment can report 85% containment, 70% automation, and 55% true resolution, depending on which of these is quoted.

The denominator problem

The second source of noise is which tickets are counted. Some figures cover chat only and leave out email, where harder cases tend to land. Some exclude tickets routed straight to payments or compliance teams. Some cover the first eight weeks of a deployment, when the AI is handling a hand-picked set of simple intents.

None of that is dishonest in itself. But a figure without its denominator cannot be compared with another figure without its denominator.

Why this matters more in iGaming than in retail

In e-commerce, a falsely ‘resolved’ ticket usually means a slightly annoyed customer. In regulated gambling, the tickets that matter most are withdrawals, verification, bonus disputes, and responsible gambling signals.

A player told that their withdrawal is ‘being processed’ when it is actually blocked on a source-of-funds check will come back, angrier, and in some markets that becomes a complaint to the regulator.

The UK Gambling Commission’s own complaints data has shown withdrawals and financial transactions as the largest single category of complaints made to the regulator. That is exactly the category where an AI that contains without resolving does the most damage.

The number is not only about the technology

There is a further reason two vendors’ figures rarely transfer from one operator to another: the automation rate is not a property of the model alone. In the deployments I have seen, the result depends just as much on three other things.

The first is the provider’s own work after the contract is signed. Someone has to map the operator’s bonus terms, know your customer (KYC) rules, payment flows, and escalation paths into procedures the AI can follow, then keep correcting them as promotions and policies change.

Vendors that send engineers and iGaming specialists to work inside the operator’s operation for the first weeks, a forward-deployed model, get to high resolution far faster than those that hand over a login and a knowledge base upload.

The second is domain expertise. An agent that understands why a withdrawal is held, what a wagering requirement means for a specific bonus, or when a message is a responsible gambling signal will resolve more, and more safely, than a general-purpose bot trained on the same documents.

The third is the operator. Teams that give the AI access to the back office, commit their support leads’ time to reviewing conversations, and set an ambitious target from the start tend to reach rates that cautious, read-only deployments never approach. The same software can produce very different rates at two operators.

So when a vendor quotes a figure, it is worth asking not only how it was measured, but what the vendor and the operator did to reach it.

Five questions to ask any vendor, including us

1. Which definition is this number? Containment, automation, or resolution, and what re-contact window?
2. Which channels and ticket types are in the denominator, and which were excluded?
3. Over what period, and how many weeks after go-live did measurement start?
4. What happens to the number on withdrawals and KYC specifically, not the blended average?
5. Can you show us the conversations behind it, on a sample of our own tickets?

A vendor that is confident in its figure should be able to answer all five in writing.

A shared standard would help everyone

I would like to see operators, platforms, and vendors agree on one public definition of AI resolution for player support, with a fixed re-contact window and per-category reporting for payments, KYC, bonuses, and responsible gambling. Trade bodies and the larger platforms are well placed to set it.

Until then, the safest assumption for any operator is that two percentages from two vendors are measuring two different things. Ask for the definition before you compare the numbers.

The post Same deployment, three numbers: Why iGaming AI support rates don’t compare appeared first on European Gaming Industry News.