FACTS ON BUSINESS SEPTEMBER 2026

Evaluating AI Vendors in a Crowded iGaming Market

New gambling-specific and enterprise research point to the same lesson: measurement discipline matters as much as the technology.

Domain Knowledge Shapes the Outcome

The gambling-specific research makes this particularly relevant. The State of AI in Gaming 2026 found substantial appetite for the technology but uneven readiness to use it. More than 80 percent of companies surveyed had adopted generative AI for tasks such as content creation and analytical insight, while adoption of more autonomous, agentic systems remained far lower.

The report suggests that this gap may partly reflect the high stakes nature of gambling operations, where regulatory compliance, player safety and operational risk raise the bar for handing a decision to a system that can act without a person in the loop.

As Kasra Ghaharian, the report's editor-in-chief and director of research at UNLV's International Gaming Institute, put it, gambling has lacked a “rigorous, independent baseline for understanding where the gambling industry stands”, which is precisely what this first edition sets out to establish.

A provider can be technically sophisticated without necessarily understanding the environment in which its technology will operate. Whether it understands player lifecycle management, gross gaming revenue (GGR), net gaming revenue (NGR) and contribution margin, and the economics of bonusing, and whether its system can function within responsible gambling processes, differing jurisdictional rules, data permissions, audit trails and human intervention, are all reasonable questions for an operator to put to a vendor before signing.

That does not mean specialist providers are automatically better than generalist ones. It means an operator should test whether a given provider understands the specific workflow and commercial problem it has been asked to address.
Financial services offers a useful comparison, if an imperfect one.

Deloitte's research into generative AI adoption in financial services found that organisations with greater self-assessed expertise, what the firm calls “pioneers”, were further along in implementation and more likely to say their returns exceeded expectations. Forty-seven percent of pioneers estimated that the return on investment from their most advanced generative AI initiative exceeded expectations, compared with 17 percent of self-described followers, a measure of self-reported perception rather than independently verified return.

Banks and insurers operate under different commercial and regulatory pressures to gambling operators, but the underlying lesson travels: technical capability is only one part of the equation.

Measure the Value, Not Just the Output

This may be the most important distinction for operators evaluating AI providers. Domino Data Lab's Fifth Annual Enterprise AI Report found that 93 percent of the 639 senior AI leaders it surveyed reported improved AI production capability, up from 88 percent in 2025. Yet 57 percent said their AI returns were growing no faster than their investment, the same rounded share as in 2025. Production capability, in other words, is improving faster than the share of organisations reporting returns that are outpacing what they have invested.

That has an obvious parallel in iGaming CRM. Suppose an AI system identifies a player at risk of churn. If that finding remains inside an analytics dashboard, the system has produced an insight. If it reaches the CRM team at the right point in the player lifecycle, generates an approved intervention and the operator can subsequently measure the result, it has become part of a business process. The model has to do more than produce an accurate prediction; its output needs to reach the person or system responsible for the decision while that decision can still be changed, which makes integration and workflow design part of the AI proposition rather than an implementation detail.

Enterprise buyers are changing how they define success accordingly. In Futurum’s survey of 830 IT decision makers, 21.7 percent selected either revenue growth or profitability as their primary measure of AI success. Those two options replaced one broader financial performance option in the previous survey, which limits direct comparison between the two surveys. The share selecting productivity as the primary measure fell from 23.8 to 18.0 percent. PwC's 29th Global CEO Survey found a similar picture: 30 percent of CEOs reported additional revenue from AI over the previous year and 26 percent reported lower costs, but 56 percent reported neither.

“A small group of companies are already turning AI into measurable financial returns”
Mohamed Kande, Global Chairman, PwC

The figures show why operators need to understand exactly what a vendor's return on investment claim represents, rather than accept the headline figure at face value. A useful discipline is to ask what the baseline was, what counts as incremental, which commercial metric is actually being measured and over what period, since a claim about improved engagement is not the same as a claim about incremental net gaming revenue, and a reduction in handling time is not necessarily a reduction in the underlying cost base. Vendor case studies can be useful evidence, but operators are better served establishing their own baseline and measurement methodology wherever possible.

None of this supports a simple conclusion that AI is failing to deliver value in gambling, or that vendors are systematically overpromising. It shows something more specific: the market is moving from a phase in which experimentation and productivity were sufficient measures of progress towards one in which operators, like enterprise buyers generally, increasingly want measurable financial and operational outcomes. The gambling-specific evidence is consistent with a similar shift, though a single benchmark report cannot on its own show how the industry is changing over time. Adoption is growing, but maturity and governance remain uneven, and meaningful returns are not yet universal.
For operators, that puts more weight on how a provider is evaluated than on which one is chosen. Starting with a defined business problem, testing a provider's domain knowledge, understanding what happens once the model produces an answer, and establishing a measurement methodology before deployment are not guarantees of a positive return.

But they narrow the gap between the two questions that matter: whether a provider uses AI, and whether it can show, in a way the operator can independently verify, what changes when it is introduced, why that change should occur, and whether the resulting value exceeds the cost of achieving it.