By scoping thoroughly upfront and building on their own SDK abstraction, Dunelm completed what they feared would be a year-long migration in weeks, and used the move as a clean slate to cut active flags from 125 to 80.
Where the previous platform was opaque, Unleash's impression data and dashboards let the team see exactly what flags are doing, settling a timeout mystery during rollout, and giving product teams the visibility to make decisions without filing a ticket.
Code goes behind a flag and engineering steps back. Product owns the rollout, the experimentation team analyzes the data, and independent non-overlapping flags let Dunelm test multiple features on the same page without interference.
Dunelm is one of the UK’s largest home furnishings retailers, grown from a single market stall into hundreds of stores and a large ecommerce business that drives 30 to 40% of sales. Unlike many teams adopting FeatureOps for the first time, Dunelm already had an established practice, trunk-based development, progressive rollouts, the works. Their challenge was different: the plane was already in the air, and they needed to change engines mid-flight.
This is the story of a fast, clean migration to Unleash, told by tech lead Tom Lightfoot.
Dunelm’s previous feature flag provider had been the right fit when they chose it. By the end of the contract, the relationship had drifted.
“Their business had outstripped ours. We weren’t able to use all the features they offered, or we were using features that overlapped with other services we’d already procured. It didn’t make sense to have such a large piece of software for something we wanted to be more targeted and focused on feature flagging, or, as I got used to with Unleash, FeatureOps.” — Tom Lightfoot, Tech Lead, Dunelm
They wanted a platform sized to what they actually did, without paying for surface area they would never use.
The first criterion was the migration itself. A switch that ate a year of engineering time would have erased the savings that justified it.
“We didn’t want to spend a year migrating from one service to another, because the cost savings from reducing our contract wouldn’t be worth it if we had to spend a whole year of developers’ time.” — Tom Lightfoot, Dunelm
Beyond that, Dunelm needed to keep everything it already relied on, rollout strategies and infrastructure as code with Terraform, and it valued the open source community around Unleash. Where the default did not cover a need, they could extend it.
“Unleash doesn’t have the biggest Terraform provider, but our friends at Philips Labs extended a third-party one, which we’ve now contributed to. It’s great that there’s an open source community around Unleash. That was a massive appeal.” — Tom Lightfoot, Dunelm
The team had built their own abstraction around the API, using React with custom hooks rather than depending on a single SDK. That paid off. They started with the basic JavaScript SDK and plugged its results straight into their existing hooks, with light reshaping to match what they had before.
“It was essentially lift and shift, really easy to migrate sideways. The API made it really easy to migrate all the feature flags we already had, and the mappings just seemed to work.” — Tom Lightfoot, Dunelm
The migration finished so quickly it surprised the Unleash team, who heard about onboarding one week and completion the next. Tom credits the preparation.
“One thing we did well was scope everything out, so there were no blind spots. That helped massively.” — Tom Lightfoot, Dunelm
The team also treated the move as a chance for a clean slate, taking active flags from roughly 125 down to 80. That part was harder, and it taught them something about lifecycle.
“It’s much harder to remove a flag that’s lived for a year. There’s always uncertainty about removing code, you don’t know whether you’ll miss an edge case. Doing it sooner, when people have more context, is easier.” — Tom Lightfoot, Dunelm
Crucially, Tom’s takeaway wasn’t to push harder on the teams. It was that the platform has to make the right thing easy.
“It wasn’t just the teams’ fault. We have to do better and offer an easier way to manage feature flags.” — Tom Lightfoot, Dunelm
Visibility was a pain point with the old platform and a deciding factor with the new one. When Dunelm saw timeouts during rollout that Unleash wasn’t seeing, the shared data settled it quickly.
“Because of the transparency of being able to see that data, we could pin down the issue, which was in our own system. That constant transparency and openness made the migration a lot easier.” — Tom Lightfoot, Dunelm
Tom is candid that the non-technical side increasingly decides these things. Snappy support, often within the hour, and a responsive technical account manager mattered as much as the platform.
“As much as I’m a technical lead, the technical part, especially in the world of AI, is becoming easier and easier. So the communication between companies and people is becoming even more important.” — Tom Lightfoot, Dunelm
For an ecommerce business, the biggest payoff is experimentation. Dunelm runs controlled tests to present products differently and cross-sell between categories, and the ability to run independent, non-overlapping flags is central to that.
“Maybe you want to test two features on the same page at the same time without having them overlap and affect each other. When the data shows a feature is adding value, more to the basket, more revenue, landing on more product types, that’s a big win.” — Tom Lightfoot, Dunelm
FeatureOps also drew a cleaner line between teams. Engineers ship the code behind a flag, and the product and experimentation teams take it from there.
“Engineers worry about the code they’re delivering and putting it behind a feature flag, and that’s where their involvement ends. The product team determines when to roll out, the experimentation team analyzes the data, and engineering doesn’t have to worry about it until a decision is made.” — Tom Lightfoot, Dunelm
That separation shows up in daily life as fewer hand-offs. Using Unleash’s impression data, the team reports only on the flags they care about instead of pushing everything into their Google Tag Manager layer, and product can make rollout decisions without filing a ticket for every small change.
Asked what he’d tell other tech leads, Tom named three things: transparency, which you only learn by talking to the vendor; developer experience, including documentation quality and the AI docs that help engineers find the right page fast; and the unglamorous discipline of scoping a migration thoroughly so blind spots don’t bite later. In his words: observability, ease of use, and blind spots.
Next, Dunelm wants to push experimentation further, targeting specific pages and user groups and building consistent journeys across related flags, as a large business moves from broad optimization into the finer detail.
