PayPoint | Our Experience

PayPoint's Platform Engineering team had a GitHub Copilot licence for every one of its 13 engineers but almost no usage to show for it.

Over an eight-week first phase, CreateFuture embedded an AI Principal Engineer in the team, combining structured coaching with workflow automation and hands-on pairing. The result was a move from isolated, occasional AI use to a habit most of the team relied on.

77%

of the team using Copilot weekly from week 3 through week 8, up from 1 engineer in 13 at baseline

500+

Copilot requests a week within the first three weeks, up from 3

100%

of the team using the new Azure DevOps integration in some form by week 6

Meet PayPoint

PayPoint opened for business in 1996, allowing customers to top up prepaid gas and electricity meters with cash at their local convenience store. The idea was tested in Northern Ireland, reached London the following year, and grew into one of the UK's best-known payments networks.

Listed on the London Stock Exchange, PayPoint now works through more than 28,000 retail outlets, handling millions of transactions a day. They range from bill payments and parcel collection to the Love2shop gift card network and services that let people without a bank account access cash. As the range of services keeps growing, so does the engineering behind them.

The challenge

Buying AI licences is the easy part of enterprise adoption. What takes longer is getting engineers who change live infrastructure to trust the tools, and then use them every day.

PayPoint's Platform Engineering team sits underneath the group's payments and technology infrastructure. This handles millions of transactions a day, and every other engineering team builds on top of it. In April 2026, all 13 of its engineers had a Copilot licence, but only one used it regularly.

The work spanned a large legacy estate, from infrastructure as code and CI/CD pipelines to manual scripting and point-and-click operations, so skill levels varied widely inside the same team. The engineers were candid that they didn't trust AI-generated changes to infrastructure without clear guardrails. Some expected that reviewing AI-written infrastructure code would take as long as writing it themselves.

The team also sits at the end of PayPoint's delivery chain, absorbing outages, incidents and last-minute business requirements that other teams never see. That unplanned load displaced planned work sprint, after sprint. Some engineers had begun reaching for tools outside PayPoint's governance to get the productivity the enterprise tooling wasn't yet delivering. The licences were in place. The missing piece was a habit the team trusted.

The approach

CreateFuture embedded a single AI Principal Engineer full-time in the Paypoint team, rather than running a one-off training session.

Two weeks of discovery came first, pairing with engineers, shadowing their day-to-day work and mapping pain points, architecture and skill spread. That revealed gaps a generic rollout would have missed, including no defined golden path for using AI safely against the team's Terraform repositories. One-to-one sessions with every engineer then set each person's starting point, and group workshops built from absolute-beginner level upward.

Two things turned the coaching into a habit:

Workflow automation: GitHub Copilot connected to Azure DevOps through Model Context Protocol (MCP) servers, so engineers could work with tickets and code from their terminal and IDE instead of switching between browser tabs.
Engineering practice, from limiting work in progress to writing tickets clear enough for anyone on the team to pick up.

AI-native delivery

GitHub Copilot did the work throughout, connected to Azure DevOps through MCP servers. Engineers had a governed way to handle tickets and code without leaving their terminal or IDE.

Engineers reviewed and approved every change themselves, and nothing in this phase automated a decision an engineer hadn't made. This mattered for a team that told CreateFuture directly it wouldn't trust anything else.

Copilot's Plan mode proved the most useful technique, working through a change collaboratively with the AI instead of asking it to write the change outright. Even an engineer who already considered themselves a heavy AI user picked up new ground in an early workshop, using subagents to hand off tasks and cut the context bloat that builds up in orchestration workflows.

77%

of the team using Copilot weekly from week 3 through week 8, up from 1 engineer in 13 at baseline

500+

Copilot requests a week within the first three weeks, up from 3

100%

of the team using the new Azure DevOps integration in some form by week 6

What was built

By the end of the first phase, PayPoint's Platform Engineering team had GitHub Copilot connected to Azure DevOps through MCP servers. Every engineer was using the integration in some form. An engineer on the team had begun prototyping a similar integration for Halo, PayPoint's IT service management tool, extending the same approach to change requests.

An agentic documentation drive produced 37 pull requests by week 6, fixing missing or outdated README files across the team's repositories. A Terraform repository in the team's Access Packages estate, which had needed refactoring for 18 months, was done in about two hours with Plan mode, paired with the engineer who owned it. It removed 1,500 lines of redundant code and produced zero change to the deployment plan.

The result

This was phase one of a wider AI enablement engagement, and what PayPoint's team has now is a working habit and a foundation, not a finished programme. The engineers run the Azure DevOps integration themselves, and one is already extending it to Halo.

“PayPoint had already paid for AI tooling and needed proof that coaching, not another licence, was what turned it into daily practice. We showed that a single embedded AI Principal can change adoption in a way a rollout of licences and a training day doesn't.”

Martin MacDonald
Cloud Lead

There's more where that came from

Building a MoneySavingExpert app took brand and engineering excellence, and complex stakeholder management. But the result has helped over a million people spend money.

View Case Study