iGAMING TRENDS
Support trends in iGaming 2027
an agent grants a bonus or closes an account inside an approved policy, is that decision stored in a form a regulator would accept?
Agentic AI was the easy prediction, 2027 will be about proving it works. Harpo Lilja, CEO of Tugi Tark, explains what actual changes we will see in iGaming support next year.
I
n this magazine last year, I argued that 2026 would be the year AI agents stopped answering questions and started taking action. Applying an eligible bonus, clearing a failed payment before the player gets angry, triggering a responsible gaming intervention while the session is still live. That shift is underway. It was also the consensus prediction. What the industry underestimated is how much of the work sits somewhere else entirely.
THE BOTTLENECK IS NOT THE MODEL
The gap between operators getting real value from agentic support and operators running an expensive pilot has almost nothing to do with which model they chose. It comes down to plumbing. Can the AI write to the back office, or only read from it? A failed deposit is a good test. Reading the error is easy. Retrying the payment, updating the wallet, and closing the ticket only works if the agent has authenticated write access to the payment provider and the PAM, and if that action is logged against the player. Does the tagging exist to identify a VIP before the first message lands? When
None of that is interesting to talk about. All of it decides whether the project works. My prediction for 2027 is that the operators who pull ahead will be the ones who did the unglamorous integration work in 2026, not the ones who ran the most impressive proof of concept.
THE YEAR SUPPORT AI GETS MEASURED HONESTLY The second thing that changes in 2027 is scrutiny. Boards funded AI support on the promise of cost reduction, and they are starting to ask for the number.
The industry is not ready for that question, because too many of the figures in circulation measure the wrong thing. Deflection rate counts conversations that ended without a human, which includes every player who gave up in the bot and left. Automation rates often divide AI-handled work by every ticket in the operation, including the ones that were never in scope for AI.
Underneath both sits a structural error. If you count escalations that happened this month but divide by every ticket that happened to be open this month, you are comparing apples and pears, as we say in Sweden. One side is a count of events inside a period. The other is a snapshot of everything sitting in the queue, most of it inherited from previous months. The result drifts with the size of your backlog rather than with the quality of the AI, and it always drifts in the flattering direction. With resolution times measured in weeks, that distortion is large.
The KPI that survives scrutiny is AI resolution rate, calculated on closed tickets that were eligible for AI in the first place. Of the in-scope tickets we
closed last month, what share did the AI resolve without a human at any point in the ticket’s life? Both sides of the ratio describe the same tickets over the same full lifespans. A ticket the bot ‘resolved’ that came back and was finished by a person does not count. A legal or VIP-only ticket the AI was never asked to handle does not belong in the denominator, and scope should be agreed with the operator up front and frozen, not adjusted after the results are in. The number lags a live dashboard, which is the point. It is a quality figure, not a marketing title.
In 2027, expect procurement teams to start asking how that rate is calculated. That is a good development for the industry and an uncomfortable one for parts of it.
A DATE IN THE CALENDAR: 2 DECEMBER 2027
Under the EU’s Digital Omnibus, high-risk obligations for standalone systems listed in Annex III of the AI Act now apply from 2 December 2027, deferred from August 2026. Transparency obligations and the AI literacy duty were not deferred and already apply.
Most operators classified their support AI as low risk when it was a chatbot answering questions about withdrawal times. That classification deserves a second look now that the same system can grant bonuses, close accounts, and initiate responsible gaming interventions. Annex III is a closed list, not a judgment call. Support AI does not become high risk because it is capable. It may become high risk because of what it is allowed to decide. The question is not whether the label was convenient. It is whether it still describes the system you are running.
There is a related shift worth planning for either way. Support transcripts are becoming compliance evidence, not just service records. If an AI recognised a risk signal in a chat and acted on it, that conversation is part of the audit trail, with the retention, explainability, and data protection obligations that it implies.
TRENDS FOR 2027
Three things to plan around. Integration quality will separate the winners from the pilots. AI resolution rate will become the number that matters, and vendors will be asked to show their working. And anyone still treating agentic support as just a chatbot should re-run the classification before counsel does it for them.
The operators who treat all three as engineering problems rather than marketing problems will be the ones still talking about results this time next year.
12 OCTOBER 2026 GIO
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36
