In-app registration
In-depth profiling & psychology
Goal setting & motivation
Utility Warehouse acquires almost all of its customers through a 60,000-strong partner network, the largest of its kind in Europe. Growth had stalled and the network was leaking partners faster than it could replace them. I joined as Head of Product to rebuild it: new onboarding, a machine-learning activation layer, and the leadership instrumentation the network had never had.
Utility Warehouse is a FTSE 250 multi-service provider that, unusually, builds its customer base almost entirely through a partner network rather than advertising. The network was a genuine competitive advantage with one structural flaw: a 95% attrition rate meant the business ran a leaky bucket, relying on constant new partner acquisition to offset the outflow. Growth was stuck, most output came from a small minority of highly active partners, and the organisation's idea of a "good partner" had been shaped almost entirely by survivorship bias.
My remit was to reinvent the partner channel and reignite customer growth. The strategy I brought was to focus on partner lifetime value rather than acquisition volume, delivered in three phases: a rebuilt mobile-first onboarding journey (a 2× lift in conversion), a machine-learning activation layer that told partners which households were most likely to switch (Switch-o-meter), and leadership instrumentation that gave network leaders real visibility for the first time.
The bets paid off. The harder lessons were cultural: pace matters in transformation work, but trust matters more.
Utility Warehouse is the UK's only genuine multi-service utility provider (energy, broadband, mobile and insurance on a single bill) with over a million customers and a decade of Which? Recommended Provider awards. What makes it unusual is how it grows. UW doesn't advertise. Instead it passes the saved marketing spend to customers and pays a small commission to partners who recommend UW to people they know. That network is 60,000 strong, the largest of its kind in Europe, and it accounts for 96% of all customer acquisition.
The model has real strengths: partners are embedded in their communities, sales are built on trust, and the customers they bring have high lifetime value and low churn. The problem was growth. Customer acquisition was running at a few percent a year, and underneath that number sat a more uncomfortable truth: most of the network's output came from a small, highly active minority. Most of the 60,000 had joined, done a little, and drifted away. I joined at the tail end of a five-year digital transformation, as one of two Heads of Product, with a clear remit: reinvent the partner channel and reignite customer growth.
For every hundred partners who joined, ninety-five eventually left. The business was built around filling the top of the funnel fast enough to offset the outflow at the bottom.
The real cost of the leaky bucket was not the inefficiency. It was survivorship bias in everything: the organisation's understanding of what good partners looked like, its intuitions about what motivated them, its assumptions about what the product needed to do. The tooling, incentives and support were inadvertently designed for the 5% who would have succeeded anyway.
Rather than treat the leaky bucket as a given and optimise the top of the funnel, the goal was to grow the share of partners who activated, retained and grew, making the network more productive without simply making it bigger. That mapped to three lifecycle stages, each carrying three workstreams.
Capabilities arranged along the partner lifetime-value journey. Colour marks the workstream; the highlighted tile is the lead capability per stage.
Matrix reconstructed from the strategy slide. Tile placement is inferred from the deck text, worth a pass to confirm the original grid.
A better first experience would increase how quickly and how many partners reached their first customer acquisition. We rebuilt onboarding inside a new native mobile app, built mobile-first to meet partners where they actually were.
Hypothesis: a user-led rather than partner-led approach to onboarding would improve activation. A mobile-first vehicle letting partners self-serve would move more of them through the key activation milestones. We set the bar at measurable improvement across four activation milestones versus the existing journey.
The app included in-depth profiling, goal setting and a high-touch flow designed to move partners from registration to first sale as fast as possible. The rebuilt journey delivered a 2× increase in customer conversion versus the control group.
Getting partners through onboarding was necessary but not sufficient. We built prospect enrichment and a machine-learning switching-propensity model that told partners which households were most likely to switch provider, so instead of knocking on every door, they could focus on the doors most worth knocking on.
The third phase focused on the leaders who recruit, develop and manage cohorts of partners. We built network-analysis dashboards that surfaced who was active, who was at risk and where intervention would matter most, and redesigned incentives to reward network health, not just raw acquisition.
Leaders had been managing blind. The dashboards gave them a view across their entire book of partners for the first time, turning "fill the hopper" into something closer to portfolio management: spot the at-risk partner early, direct coaching where it would move the needle, and reward the behaviours that built a durable network rather than a one-off spike in sign-ups.
From launch in 2021, the app grew from a simple onboarding tool into a sales platform, each release adding capability and building toward the enrichment work and the Switch-o-meter integration that followed in year two. Pick a release to see what shipped.
The core strategic bets paid off: the mobile platform, the propensity model, and the investment in instrumentation all delivered measurable improvements. Bringing startup energy into a corporate environment moved things faster than the organisation was used to. But pace was a double-edged sword.
Pace, experimentation, OKRs and a genuine build/measure/learn culture surfaced what mattered and gave the team permission to stop working on things that weren't moving the needle. The 2× onboarding lift and the first production ML model both came out of that.
I didn't spend enough time up front winning over the engineering organisation. I moved on the assumption that good ideas would be self-evident and results would bring people with us. They didn't, at least not immediately. It took six months that were harder than they needed to be.
We used experienced contractors to build the native app fast, the right call for momentum. But as we transitioned to internal capability, pace naturally dropped, and that created consternation with management who'd come to expect a certain velocity. The expectation-setting should have been earlier and more explicit.
The switching-propensity model took longer than anticipated to show signal, a familiar problem for ML in an organisation that hasn't built it before. Management expected rapid results; the model needed training and iteration. It worked in the end, but better expectation-setting would have absorbed the pressure.
The partner panels we used for discovery were more early-adopter than we'd realised, so our adoption assumptions were overstated. The product was right; the rollout expectations were calibrated against the wrong users. Being careful about survivorship bias (in organisational culture as much as in research samples) is now a standing discipline for me, not a one-time check.
The UW Partner Network is a study in what it takes to bring product-led thinking into an organisation with real commercial scale that hasn't fully made the transition. The app, the model and the dashboards were real wins: a 2× improvement in onboarding conversion, a data-driven activation layer where intuition had done all the work, and leadership instrumentation that gave the network visibility it had never had.
But the more durable thing I took from UW was a sharper read on the conditions under which product transformation actually works. Pace matters, but trust matters more, and strategic bets need patience as much as conviction. The deeper lesson was about bias: the assumptions that live inside an organisation, in its culture, its data, its mental models of who it serves, will quietly shape everything you build unless you name them and design against them.
The activation work also left more than a model. Switch-o-meter became UW's first MLOps infrastructure: the pipelines, deployment and monitoring behind one production model. That platform became the foundation for later AI initiatives across the business: an NLP read of call-centre transcripts to sharpen customer service, churn modelling in the home-services team, and others since. The most valuable thing I built was not a model. It was the capability to keep building them.