What a Medical App Taught Me About Testing at a Petrol Station

In the world of consumer apps, testing is not an event. It is the operating model.

When I co-founded Your.MD - a digital health platform that eventually reached 30 million users and achieved Class IIa medical device certification - we ran continuous experiments as a matter of course. Different versions of the product deployed simultaneously to different user groups. Real-time analytics showing exactly where users engaged, where they dropped off, and what drove them to return. A change shipped in the morning generating measurable data by the afternoon.

What made this particularly meaningful was the context. Healthcare is one of the most heavily regulated industries in the world. The assumption is often that regulation and rapid experimentation are incompatible - that safety requirements necessarily slow down learning cycles and force conservative, sequential approaches to innovation. Our experience at Your.MD proved the opposite. Medical device certification and continuous experimentation are not mutually exclusive. With the right infrastructure and discipline, you can move fast and maintain rigorous safety standards simultaneously.

That insight stayed with me. And when I created BP's Convenience and EV Labs in 2021 - building the team, the vision, and the methodology from scratch - it became the foundation of everything we tried to do.

The Gap Between Digital and Physical Testing

Petrol retail, like healthcare, is a heavily regulated industry. The presence of flammable materials creates safety requirements that govern everything from the equipment that can be installed on site to the certification standards that hardware must meet. For decades, this created a genuine barrier to the kind of rapid physical experimentation that digital teams take for granted.

There was also a rational risk calculation at play. If an experiment at a live customer site went wrong - technically, operationally, or in terms of customer experience - it could become a public incident for a global brand. Large retailers, including BP, developed testing methodologies appropriate to those constraints: controlled lab environments, vendor-managed pilots, simulated forecourt conditions. Solutions would be evaluated, and results would arrive weeks or months later, filtered through vendor analysis and reporting.

This was not a failure of imagination. It was a sensible response to the realities of physical retail.

What changed everything was the energy transition.

The Shift That Made This Possible

As EV charging began replacing or complementing traditional fuel at forecourt sites, something quietly significant happened. Pure petrol environments require ATEX certification - strict standards governing any equipment installed where explosive atmospheres may be present. EV charging sites operate under different safety parameters. Hardware that would have faced significant certification barriers at a traditional petrol forecourt could be deployed far more straightforwardly at an EV charging location.

The energy transition that posed an existential challenge to the traditional forecourt model simultaneously opened a window for a different kind of innovation. Sites becoming EV destinations could host smart hardware: connected kiosks, intelligent vending, remote-managed systems with software interfaces that could be updated, reconfigured, and experimented with in real time.

This was the moment where the methodology I had developed at Your.MD - and refined further while scaling Talking Tom and Friends to over a billion downloads - became directly applicable to a physical forecourt environment.

What Smart Hardware Actually Changes

The core insight is straightforward: most physical hardware has become software-defined.

A modern vending cabinet is not just a cabinet. It has network connectivity, a software interface, remote management capability, and the ability to present different experiences to different customers. The same is true of EV charging stations, ordering kiosks, payment terminals, and customer-facing screens. Once physical hardware has a meaningful software layer, the principles that govern consumer app development can apply directly to physical retail.

You can deploy different configurations to different sites simultaneously. You can monitor customer interactions in real time. You can push changes remotely without physical intervention. The methodology that digital teams take for granted - adapted for the safety and regulatory requirements of a physical environment - becomes possible.

The challenge was building the infrastructure to make it real.

Building the Capability

What separated genuine real-time testing from the existing model was not primarily the willingness to test with real customers - though that mattered enormously. It was the ability to see what was happening immediately, and to act on it without waiting for a vendor's development cycle.

We built comprehensive analytics from the ground up: not results processed and reported by a third party weeks later, but a live view of customer interactions, transaction data, conversion rates, and operational performance as they happened. If a customer abandoned a process at a particular step, we could see it that day. If hardware went offline, we knew before the site staff did.

We also built remote management capability. Software updates, journey changes, pricing adjustments, hardware reconfiguration - all deployable from our office in London to a live site thousands of miles away, without requiring anyone on the ground to do anything. A hypothesis identified in the morning could be tested by the afternoon, and validated or discarded by the end of the week.

This combination - real customers, real-time analytics, remote deployment - was the infrastructure that made everything else possible.

The Test: South Africa, 2024

In 2024, we launched at three locations in South Africa with bPOWERd - a portable battery rental system designed for markets with unreliable power infrastructure. Customers could rent rechargeable batteries from secured smart cabinets at BP forecourts - a cleaner, more convenient alternative to the diesel generators many were already using.

The initial launch was structured to maximise learning. A team member was present at each site in the first phase - not because the system required it, but because having a human observer captured qualitative detail that analytics alone cannot: what confused customers, what questions they asked, what moments caused hesitation. Every friction point was a candidate for a remote software change in the next iteration.

Several insights emerged that no controlled lab test could have surfaced. bPOWERd attracted customers who were not primarily visiting the site for fuel - a new customer segment that increased foot traffic and created upselling opportunities for convenience items alongside each battery transaction. Specific steps in the customer journey that created friction became visible at a granularity that allowed precise, targeted improvements. Hardware performance in real South African conditions revealed patterns that no simulated environment would have produced.

The product is now expanding to Nigeria - a validation not just of bPOWERd as a product, but of the methodology that produced it.

The Test Store Approach

The most significant outcome of the South Africa launch was not the product itself. It was the proof of concept for a testing methodology that BP can now apply at scale - and that makes the company meaningfully stronger as a retailer.

Major retailers - Amazon, Nike, McDonald's, IKEA, Starbucks - have operated dedicated test environments for years. Locations where new technologies and concepts are validated with real customers before broader rollout. Not simulated conditions, but live sites with real transactions and real behaviour. The learnings from those environments inform product decisions, operational changes, and investment priorities in ways that no internal lab can replicate.

BP now has exactly this capability. What was built in South Africa is not just bPOWERd, and not just the remote management and analytics infrastructure that supports it. It is a replicable, scalable methodology for testing customer-facing innovations safely at live sites - and the demonstrated confidence that it works.

That matters for three reasons. First, it is safe: the combination of real-time monitoring, remote management, and a disciplined phased approach means that live customer testing can be conducted without operational or reputational risk. Second, it scales: the same infrastructure and template that enabled three sites in South Africa can be applied to locations across BP's global network. Third, it produces better customer solutions - because insights gathered from real customers in real conditions are simply more reliable than anything a controlled environment can generate.

The retailers who have understood this longest have built compounding advantages over time. Every test cycle produces better products, more precise operations, and a clearer understanding of what customers actually want. BP now has the foundation to do the same.

The Principle

The gap between digital and physical testing is closing - but it requires deliberate effort to close it.

What the Your.MD experience taught me was that regulation and rapid experimentation are not opposites. A medically certified platform can still learn from its users continuously. A petrol station, transitioning to EV, can still run live experiments safely.

The methodology that consumer app teams have used for years is available to physical retail - the question is whether organisations have built the internal capability to use it, or whether they remain dependent on vendor-mediated cycles that deliver conclusions long after the moment to act has passed.

The retailers who figure this out first will learn faster, iterate more precisely, and build a clearer picture of what their customers actually want - not what they assume they want.

That is a significant advantage. And it compounds.