- iGaming Retention Day, a free online conference on player retention organised by Handbox.io with GGRBoost.io, took place on 17 November 2026, in Russian, with ten talks.
- Arctura's talk at 13:55 CET argued that an operator without its own platform, CRM or analytics should not start AI with deep personalisation, which needs about six months of clean history.
- The first piece should be player support: the data already exists, the effect shows in weeks, it scales to new brands and markets, and it rehearses data access, ownership, measurement and team adoption.
- An agent reading every chat sends CRM three signals it never gets otherwise: withdrawal complaints, stated intent to leave and early VIPs.
- First projects break on team fear, no business owner and no agreed success criterion: one launch looked like '70% unhappy with the bot' until human agents turned out to be rated negatively in about 90% of cases.
- The talk ended with a 90-day plan: knowledge base, agent on part of the traffic, signals into CRM.
iGaming Retention Day took place online on Tuesday 17 November 2026, 10:00 to 16:00 CET. It is a free conference on player retention, organised by Handbox.io, a CRM-marketing agency for iGaming operators, with GGRBoost.io as partner, and it ran in Russian. Ten speakers each had a 25-minute talk and five minutes of questions from the chat. Arctura's founder, Viktor Andriichuk, spoke at 13:55 CET on where an operator without a data team should start with AI. This page is a recap of that talk in English.
Disclosure: Arctura is not paid by the organiser, and this page is our own editorial work. Last updated: 17 November 2026. We will add the recording and the results of the audience poll here once they are available.
Key facts
| Date | Tuesday 17 November 2026, 10:00–16:00 CET |
|---|---|
| Format | Online, one stream, nothing in parallel; streamed from a studio |
| Price | Free; registration on retentionday.io |
| Language | Russian |
| Organiser | Handbox.io; partner GGRBoost.io |
| Talks | 25 minutes plus 5 minutes of questions from the chat |
| Recording | Promised to registered attendees; we will link it here if it is published |
Programme
| Time | Talk | Speaker |
|---|---|---|
| 10:00 | Opening: why this conference, and what the market says about retention | Vladimir Andrienko, CEO, Handbox.io |
| 10:15 | The next tab: what a player does when they leave you | Max Tesla, CEO and co-founder, Blask |
| 10:50 | Communications infrastructure: must-haves for 2026–2027 | Nadezhda Sinenko, Head of Technical, Handbox.io |
| 11:25 | From click to return: how data changes decisions in iGaming | Daniil Lebedev, Product Manager, Playerly |
| 12:00 | Segmentation that pays: splitting the base so every email earns its cost | Dmitry Matua, Team Lead CRM, Handbox.io |
| 12:35 | Break | |
| 12:45 | Retention as a product: what keeps a player when everyone's bonuses are the same | Andrey Beyerbakh, CEO and co-founder, Gamixter |
| 13:20 | VIP portfolios worth millions are still run in Excel | Alexandra Voronetskaya, CCO, The Playa |
| 13:55 | AI for an operator without a data team: where to actually start | Viktor Andriichuk, founder, Arctura |
| 14:30 | From game lobby to communications: raising retention by personalising the game experience | Bogdan, Lead Partnership Manager, MICo |
| 15:05 | CRM marketing in LatAm: what really retains players in Brazil, Mexico and Peru | Ekaterina Belaya, iGaming Research Lead, GGRBoost.io |
The talk: AI for an operator without a data team
The talk did not explain what AI is or how models work; by 2026 every conference has covered that. It asked a narrower question: where does AI start for an operator that has no data team of its own, runs on someone else's platform and uses someone else's CRM? The answer was eat the elephant one piece at a time, and make player support the first piece, because support is already retention.
Starting point: our own mistake
Before Arctura, Viktor spent several years building AI systems inside a large iGaming operator, where the AI department grew from two people to twenty-five and shipped around ten products: support automation, retention, anti-fraud, telemarketing and quality control. After years in one company, it is easy to assume every operator works at the same level. So when Arctura started, the plan was to offer what had worked there: deep personalisation that decides which player gets which bonus, which game and which channel.
Conversations with owners, CEOs, operations directors and CRM teams showed a different market. Operators where AI is part of the product and earns measurable money are a handful, usually large ones with their own platform, data and a team that has done this for years. Everyone else is watching. Some teams that were thinking about AI did not yet have basic player segmentation at the start of 2026. Everyone agrees they cannot fall behind; few know where to start.
Why personalisation is not the first step
The talk took the hardest case: player data spread across three or more systems, and nobody who can pull a year of deposit history in a day. Such an operator usually does not own three things:
- The platform. Player data is available only as far as the provider allows: some exports work, some go through the provider's support, some do not exist.
- The CRM. Retention scenarios live in a third-party system, and plugging in an outside model that decides bonuses in real time is a separate, difficult project.
- The analytics. Nobody collects player history, labels it and watches that the data does not drift.
A personalisation system needs roughly six months of preparation: collect history, account for seasonality and big sporting events, train the model. Without that, the model learns from noise, gives noisy recommendations, and the team concludes that AI does not work. Personalisation is not a bad idea; there is simply no foundation for it yet.
How it started for us: the first piece
The AI department Viktor worked in did not start with AI in the heart of the product either. It had no reputation: other teams did not know it existed. It was given a free hand to find where AI would help, went from team to team, and heard from marketing that acquisition was getting more expensive while conversion stayed low. The first project was game recommendations in the lobby. In three months it gave roughly a 15% uplift: players stayed in the lobby and started a game noticeably more often. With that number the department went to other teams, and they started bringing their own problems. Eighteen months later there were around ten products and twenty-five people. The detailed numbers from that period, and what each one measures, are in our iGaming AI case study.
The lesson of the talk: the big retention system was the end of a path, and the path went piece by piece. Each operator should walk it again, starting with a small project whose result can be measured and shown.
Why support is the first piece
- A support chat is already retention. Players often decide whether to stay when something goes wrong: a withdrawal that did not go through, a stuck deposit, unclear bonus terms. How fast and how humanly the chat answers decides whether they come back.
- It scales to new markets and brands. The usual pain of a support team is hiring, training and churn, with payroll growing alongside the player base and a team hired in advance for every new market. With a ready-made AI core, brand tone, knowledge base, jurisdiction rules and payment methods are set up and tested in four to six weeks. The integration with the platform and the helpdesk is built once, so each following brand launches faster, and night shifts and rare languages stop being a problem.
- It is a rehearsal for every later AI project. Four things get settled on a project where a mistake is still cheap: a data layer between the operator's systems and AI, a business owner for the result, a metric agreed before the start, and how the team takes to AI. Personalisation needs all four, and there a mistake costs much more.
Three signals a CRM never gets
A person on the first line cannot turn a thousand chats a day into analytics for CRM; an agent that reads every conversation can. It sees three signals: a complaint about a withdrawal from a player who is still playing but already annoyed and will leave within a week if nobody talks to them; a stated intention to leave ("I'm closing my account", "others pay out faster"), which only chat agents ever see; and an early VIP, a player who asks about limits, large withdrawals or the VIP programme. Each signal goes to the CRM team as an event: who, what they said, when. The existing retention scenarios then run as usual, with no code changes. We describe these signals in casino CRM strategies: use the signals support already has and spotting VIP players early in support.
Players have plenty of choice, so even one who can deposit hundreds of thousands often starts with 50 dollars to test a brand. A VIP team sees them only after a large deposit, when every competitor is already chasing them. Support can spot them earlier, from what they write in the chat and how they behave in the chat widget, by comparing them with players who became VIPs before. The difference is timing: pick the player up at 50 dollars instead of fighting for them at a million.
Where first projects break
Three problems come up again and again, including in the AI support rollout Arctura is doing now with a European operator.
- The team is afraid. Of change in general and, in support, of being replaced. Announce "we are automating support" and the project gets slowed down quietly: the knowledge base is not handed over, "this needs empathy" becomes the standard answer, testing drags on. What worked, twice: position AI as an assistant that takes night shifts and typical questions while people keep the hard cases, conflicts and VIP players. And the message has to come from management, not from the vendor.
- Nobody owns the project. In the operator where Viktor worked, the first support project happened because the head of support came with the pain of endless hiring and training and became its champion. In the current rollout the champion is the CEO, which is why the project kept moving when the support team pushed back. A pilot needs an owner on the business side with authority and a budget, or it hangs between IT, support and CRM.
- Nobody agreed what success means. A few weeks after the first AI support launch, the head of support came with "by my estimate, 70% of players are unhappy with your bot". The launch had been careful: one market, part of the players, people on standby. Going through the ratings together showed that human agents were rated negatively in about 90% of cases in the same data. Players come to a support chat with a problem and with emotions, and low ratings there are normal; against that baseline the bot did better. The mistake was not agreeing beforehand what counts as success and what the bot is compared with. Since then, every pilot starts with a single page: metric, threshold, deadline and baseline.
A 90-day plan without a data team
The talk closed with a plan for an operator without a data team. The fastest route is not to build from scratch but to start from a ready-made solution.
| Days | Step | What happens |
|---|---|---|
| 1–30 | Knowledge base | Export at least the last month of chats and analyse them with AI: typical requests, languages, times of day. Decide what goes to the agent and what stays with people. Apply a QA scorecard, configure the agent, agree the success criterion |
| 31–60 | Agent on part of the traffic | Pilot on clear topics: deposit status, verification, bonus terms. Measure the share of conversations closed without a person and the player rating, against the agreed criterion |
| 61–90 | Signals into CRM | Withdrawal complaints, intentions to leave and early VIPs go to the CRM team as events, and the team starts working with them |
After 90 days the operator has savings on the first line, a system it can roll out to new brands and markets, and its own first data channel for retention: the first piece of the elephant. The same approach, for operators with no AI team at all, is on our page no AI team? Start where the data already is. How we prove each step worked is in our measurement approach.
After the conference
During the talk the audience answered two questions in the chat: how many systems hold player data at their operator, and whether they have an analyst who could pull a year of deposit history in a day. We will publish the results here, together with the recording, once the organiser shares them. If you are an operator and want to go through the 90-day plan on your own support data, the booking link is below. Questions about the talk can go to hello@arctura.eu.
Frequently asked questions
When was iGaming Retention Day 2026?
On Tuesday 17 November 2026, from 10:00 to 16:00 CET, online, in a single stream with no parallel sessions.
Is there a recording of iGaming Retention Day?
The organiser promised the recording to registered attendees. We will link it on this page if it is published.
What language was the conference in?
Russian. This page summarises the programme and our talk in English.
Who organises iGaming Retention Day?
Handbox.io, a CRM-marketing agency for iGaming operators, with GGRBoost.io as partner.
What was Arctura's talk about?
Where an operator without a data team should start with AI. Deep personalisation needs about six months of preparation and a foundation most operators do not have yet, so the first piece should be player support: the data already exists, the effect shows in weeks, and an agent that reads every chat sends the CRM team signals it never gets otherwise. The talk ended with a 90-day plan.
Why should AI in iGaming start with player support rather than personalisation?
Because support chats already exist at every operator whatever platform it runs on, results can be measured within weeks, and the project settles the data access, business ownership, success metric and team adoption that any later AI project, including personalisation, will need.
Last reviewed 17 November 2026. Figures marked as industry ranges are ranges, not Arctura claims; operator-specific numbers are only quoted after validation on the operator's own baseline.


