More than half of the operators in a 2026 SBC Media and Comm100 survey had already deployed or were piloting AI agents in player support. At Coinbase, the company-reported figure is larger: 65% of customer-support interactions were fully automated by October 2025.
The useful question for a Head of Player Support is no longer whether to use AI. It is which conversations to leave to it. The operator benchmark points at two ticket types the industry is not ready to hand over.
Where operators draw the line
The SBC Media × Comm100 benchmark reports a steep trust gradient:
- 65.2% trust AI for proactive engagement.
- 60.9% trust it for moderate reactive help.
- 21.7% trust it with a complex responsible-gambling query.
- 17.4% trust it with a VIP interaction.
Those percentages describe what operators say they trust AI to handle. They are not measured failure rates. Even so, the gap is useful: confidence drops precisely where judgment, regulatory exposure and player value converge.
1. The responsible-gambling escalation
Every operator in the benchmark trains support agents on responsible-gambling behaviours. The operating system around those agents is less complete:
- 55.2% cite limited integration with responsible-gambling systems as their biggest compliance challenge.
- 51.8% still rely on manual supervisor review.
- 30% give players no direct route to a responsible-gambling specialist.
The weak point is not detection alone. It is the handoff between a signal, the support desk and the person authorised to act.
Regulators have made the cost of a missed handoff concrete. In October 2025, Ontario's AGCO issued a C$105,000 monetary penalty against theScore, alleging that clear warning signs—including distress disclosed to a VIP host—did not lead to meaningful intervention.
In Great Britain, the Gambling Commission fined Spreadex £2.022 million after anti-money-laundering and social-responsibility failures. One cited case involved a customer who hit a daily deposit limit 12 times in 14 days; the operator used four pop-up messages and did not carry out a human interaction.
Neither action proves that AI mishandled a ticket. Both show why an automated alert is not an intervention. The escalation path needs a trained person, enough context to make a judgment and a record of what happened next.
What a safe handoff requires
- A trigger the support team can see. A risk signal trapped in a separate responsible-gambling system cannot change the conversation.
- A named owner. “Send to compliance” is not a queue design. The case needs a role, a response target and a fallback.
- A complete context packet. Recent contacts, behaviour flags, account actions and the player's own words should move with the escalation.
- A closed loop. The support record should show who reviewed the case, what was decided and when the player was contacted.
2. The VIP conversation
Only 17.4% of surveyed operators would trust AI with a VIP interaction on its own. A separate study shows what is at stake.
The Playa, Gaming Operations Academy and WarriorLab surveyed 70 VIP professionals in 2026. iGaming Business reported that:
- 46% often or very often see VIP players become inactive without warning.
- 71% have VIP data spread across multiple systems.
- 38% worry that AI could reduce the personal, human touch.
For a support operation, that means a high-value player can drift away without opening a ticket, while the signs are split across systems no single agent sees at once.
Coinbase offers a useful comparison from crypto. On its Q3 2025 earnings call, the company reported 65% support automation while also describing a white-glove service for high-value advanced traders: concierge-level support, a personal account manager and commitments around time to resolution.
That is not automation versus people. It is automation for routine volume and named human ownership where the relationship carries more value or risk.
What these numbers do not show
The SBC/Comm100 figures are stated trust, not observed error rates. The Playa study has 70 respondents and is vendor-produced. Coinbase's 65% is a company-reported aggregate that does not reveal which ticket types were automated.
Treat the numbers as operating signals, not universal performance benchmarks. They support a clear question: does your automation rate hide the mix of work being automated?
A desk can look efficient overall while routing its most sensitive conversations through the least appropriate channel.
A 30-minute check for your own queue
Take last month's closed tickets and add two tags.
- Resolved by AI or by a person? This gives you the overall automation rate—the number most boards already see.
- Routine, or sensitive? Mark any conversation you would not want a regulator or a top VIP to read back without human review.
Then compare the rates. If your sensitive-ticket automation rate is not tracked separately, the aggregate cost figure is hiding the boundary that matters.
The review should answer four practical questions:
- Which responsible-gambling and VIP signals can the support team actually see?
- Which automated flows can stop and hand a case to a person?
- Who owns the case after the handoff?
- Can you reconstruct the decision later from the record?
Common questions
Does this mean AI player support does not work?
No. Operators report much higher trust for proactive engagement and moderate reactive help, and Coinbase says most of its support interactions are fully automated. The issue is ticket selection, not a blanket rejection of AI.
Does “can't close” mean AI fails on these tickets?
Not as a measured claim. The available research reports where operators stop trusting AI to work alone; it does not publish a failure rate for responsible-gambling or VIP tickets.
Why is trust lower for responsible-gambling cases?
These cases combine regulatory exposure, incomplete context and human judgment. Recent enforcement actions show that identifying a warning sign is not enough: an operator must make a timely, proportionate intervention.
What should a Head of Player Support track?
Track two numbers instead of one: the overall automation rate and the automation rate on responsible-gambling and VIP tickets. Add handoff time and human-review completion, and the dashboard starts measuring control rather than cost alone.
Where risk sits in player support
Responsible-gambling escalations and VIP conversations are where a missed signal can become a regulatory action or a lost relationship. They need specialists who know the rules, understand the player and can leave an auditable record.
That is the work Toeshee does for iGaming, fintech and digital-asset platforms: specialised support run as a risk-management function, with automation for scale and accountable people at the moments that should not be left to a model.
We've got your back. Human where it matters.