evoke | Our Experience

Across regulated gambling and gaming, machine learning and generative AI have quickly moved from pilot projects to business infrastructure. However, this usually only happens one use case at a time, each with its own pipeline built from scratch. It’s an approach that works until the backlog of candidate use cases outgrows the teams available to build them.

evoke, whose brands include William Hill, 888 and Mr Green, reached that point in 2025. In response, CreateFuture designed and built a single AI platform that brought machine learning, generative AI and automation onto one shared foundation, connected to evoke's existing Snowflake data platform.

The result is any team can now start a new use case from something already built instead of from scratch.

Meet evoke

evoke is one of the world's leading betting and gaming companies, running brands including William Hill, 888 and Mr Green across sports betting, casino and poker. More than 1.7 million players use its platforms every month, backed by over 1,300 betting shops on the UK high street. It's listed on the stock market, so its AI ambitions sit inside a Value Creation Plan that shareholders can see and measure it against.

The challenge

evoke had set out a public Value Creation Plan partly built on using technology to run its sports betting and gaming business more efficiently. AI sat at the centre of it. Individual teams had already started building their own machine learning and generative AI for forecasting revenue, reading customer sentiment, and recommending bets to players. Each one built its own pipeline into evoke's data from scratch.

evoke had already brought its data together on a Snowflake-based Smart Data Platform. What it didn't have was an equivalent foundation for AI. It needed a shared way to move data into a model, train and monitor it, then put it safely in front of customers. Without this, only the best-resourced teams could move, and the rest of evoke's growing backlog of use cases queued behind them.

The approach

CreateFuture designed the platform in two layers: a shared foundation any team could build on, and a small number of priority use cases to prove it in production.

The foundation runs on AWS SageMaker and Bedrock, connected directly into evoke's Snowflake data platform. An MLOps framework handles ingestion, training, deployment and monitoring end to end, so a new use case plugs in rather than starting over.

An eight-person team, spanning data engineering, cloud infrastructure, MLOps and security, was embedded inside evoke's own technology function. This was alongside AWS as evoke's cloud partner, so evoke's own engineers learned the platform as it was built rather than inheriting it afterwards.

evoke's existing cloud landing zones stayed in place, with a move to a new strategic landing zone planned separately. The platform was designed with transition states so it could go live now and move across later, instead of waiting for that migration to finish first.

Delivery focused on three use cases evoke had already prioritised: sentiment analysis of customer feedback, revenue and cash flow forecasting, and personalised recommendations for players. Each was chosen because it proved the platform's core capabilities without a bespoke rebuild.

AI-native delivery

The generative AI layer runs on AWS Bedrock, giving access to Claude, GPT-4 and Qwen behind one integration. Prompt routing sends each request to whichever model gives evoke the best balance of cost and quality, and model distillation. This cuts cost and response time further where a task doesn't need a model's full size.

On the machine learning side, SageMaker handles reinforcement learning and fine-tuning on evoke's own data, with MLflow tracking experiments as models are built. Every generative AI output that reaches an evoke customer, or informs a decision the team makes, goes through human review before it's acted on, backed by Bedrock's own content guardrails.

What was built

evoke now has one AI platform, built on AWS SageMaker and Bedrock and connected directly to its Snowflake data platform. This covers the whole lifecycle a machine learning or generative AI use case needs: bringing data in, preparing it, training and deploying a model, and monitoring it once it's live.

The platform sits inside evoke's existing cloud infrastructure today, with a defined path to a future landing zone already built in, and supports multiple large language model providers so evoke isn't tied to one as the market moves.

"The test for us was never whether a model worked in a demo. It's whether the next team can start from something built instead of queuing behind the last one. Three use cases are live, and the backlog is moving. Working with CreateFuture means 70,000 recommendations reached real players last year, from a platform that didn't exist eighteen months ago."

Kanda Kumar
Director of AI and Intelligent Automation

The result

evoke's own engineers, embedded with the CreateFuture team from day one, now run and extend the platform themselves. CreateFuture continues to work alongside evoke and AWS, migrating further model suites onto the platform and adding new use cases. This means evoke's backlog keeps moving rather than waiting on the next phase of work.