The cost of full-cycle CS automation for betting operators

Raphie co-founders Justin Heath and Alex Jones ran support operations for betting brands before they built AI for them. They speak with SBC News and argue the industry is automating player support faster than it understands what support is for. Every operator in the market is now automating custome…
Raphie co-founders Justin Heath and Alex Jones ran support operations for betting brands before they built AI for them. They speak with SBC News and argue the industry is automating player support faster than it understands what support is for. Every operator in the market is now automating customer support. The pitch is familiar: cut cost-to-serve, clear the queue, keep players out of a human agent’s inbox. With gambling taxes climbing and margins under pressure, the logic is hard to argue with. It also misses what a support desk is for. The desk is the first place an operator sees a broken deposit flow, a bonus nobody understands, a withdrawal stuck in review, a player who has started to chase losses. Support is where the business finds out what is wrong, usually before any dashboard does. The problem is not automation itself, but what betting operators automate for. Most AI support is built to contain a conversation and close it without a human stepping in, and the figure that goes back to the business is how often that happened. A player who gives up and shuts the chat scores the same as a player whose problem was solved. “An abandoned conversation is categorically a negative outcome,” says Justin Heath, co-founder and CEO of AI player-support firm Raphie. In a regulated market it can be worse than negative. The signals that matter most, financial distress, confusion, the early markers of harm, are the ones a deflection-first system is built to swallow. Operations first, automation second Heath and co-founder Alex Jones reach the problem from the other end. Before Raphie was an AI company, it was the team running the support and payments operation for betting brands, at one point fielding hundreds of agents across multiple betting operators. They built the floors, trained the agents and watched which tickets came in. The AI layer arrived later, on top of eight years in the business and more than a million real player interactions. “We ran support for operators before we ran AI for them,” Jones says. “That order matters. You cannot automate a process well if you have never had to run it during the peak events and address issues specific to the industry.” The floor taught them that iGaming support looks nothing like generic customer service. Around 90% of contacts need account-specific information: this player, this bet, this pending withdrawal. A model trained on help-centre articles and generic chat logs handles almost none of it. That is the gap betting operators keep falling into, where a tool advertised at 70% or 80% automation delivers 5% or 10% once it meets live tickets. “Operators tell us they already have an AI tool inside Salesforce or Zendesk,” Heath says. “Then they show us the automation rate, and it is nowhere near what was on the box.” Resolution as the operating standard Raphie’s answer starts with what it counts. The company reports on true resolution rather than deflection. A contact only lands as resolved when three things hold: The player is satisfied, judged on their own rating of the exchange rather than on whether the chat simply closed. No human had to step in, so the outcome counts as genuinely autonomous and not a case handed to an agent at the last moment. The same player does not come back about the issue within 24 hours, which screens out the “resolved” ticket that reopens the next morning. On that stricter definition Raphie puts true resolution at roughly 74%, with customer satisfaction around 88% across its book of business. Jones argues it is the main number worth managing to. “Containment flatters everyone and fixes nothing,” he says. “Resolution is the number that shows up later in retention.” The model is built to earn autonomy rather than assume it. Deployments open with a narrow scope and widen category by category as the system proves itself, much the way the founders once brought human agents up to speed. When a case needs judgment, a compliance flag, a vulnerable player, a disputed bet, it escalates to a trained specialist instead of improvising. Raphie treats that escalation path as core product, not a failure state, which is the part regulated operators tend to test hardest. Handled well, players rarely notice the handover. “We see people wishing Raphie a good night and signing off,” Jones says. Proof in a regulated deployment The case the founders point to is an OSB operator in Arkansas, where Raphie built the human support and payments operation first and automated it in stages. Within the first eight weeks of full automation on selected ticket types, the operator reported a 77% autonomous resolution rate, response times down by more than half, and player satisfaction above 90%. Across its clients, Raphie says it takes cost-to-serve down by more than 60%, with a typical go-live inside roughly ten weeks. The common thread is sequence: the betting operators seeing the strongest numbers tend to be the ones who got the support operation right before they asked a model to run it. Scaling into European markets The company is now taking the model into Europe, where rising gambling taxes and tighter cost bases are pushing operators toward automation on a shorter timeline than their US counterparts. The payments, bonus and KYC knowledge transfers; Raphie rebuilds the compliance and escalation layer market by market to match local player-protection rules and audit expectations. Heath reads the cost pressure as the opening. “The appetite to move quickly is higher in Europe than anywhere we have operated,” he says. The founders’ argument comes back to where it began. Support is not a cost line to minimise. It is the part of the business that sees the player clearest, and the question is whether automation sharpens that view or blanks it. “The floor was always the best early-warning system an operator had,” Jones says. “The job now is to make the AI listen as well as the floor did, and act faster.” On that reading, the betting operators who win with AI support will not be the ones who automated first. They will be the ones who never forgot what their support desk was for.