How the Field Research Started
I arrived in Nairobi in March 2024 with an introduction from a startup accelerator that had worked with several boda-boda cooperatives on digital payments infrastructure. The introduction got me a two-hour meeting with Samuel Wachira, who manages a 40-vehicle electric boda-boda cooperative in Kasarani. He agreed to let me observe operations for a month in exchange for helping him document the cooperative's maintenance records in a spreadsheet.
One month became six. Samuel introduced me to two other cooperative managers. I ended up with access to 114 vehicles total — three cooperatives, three neighborhoods, three different configurations of charging infrastructure and maintenance practices. The field notes from those six months became the design specification for Stima.
What follows are the observations that shaped the product most directly. Not the numbers — I've published the data in other posts — but the specific situations and decisions that made the problem visible.
Observation 1: The Driver Knows First
Every cooperative manager told me the same thing: the drivers know when a battery is going bad before any inspection catches it. They feel it in the throttle response. The bike feels "lazy" — takes longer to accelerate from traffic lights, struggles on the Ngong Road hill that most experienced drivers use as a calibration point for how strong a pack is feeling that day.
The problem is the information doesn't travel reliably. A driver who tells his cooperative manager "this bike feels lazy" might be ignored — the manager is handling six other things, the bike just came off a charge showing 95%, and there's no objective record to check against. The driver takes the bike out. It cuts out in Westlands at 11am. Now there's a tow and a missed shift and a driver who is owed income he didn't earn.
This observation led directly to one of Stima's features: the driver app includes a one-tap "battery concern" report that is automatically surfaced in the dispatch dashboard alongside telemetry data. When a driver flags a pack and the telemetry shows declining peak power delivery, the combination is treated as a high-confidence degradation signal. The driver's intuition becomes data.
Observation 2: Maintenance Records Don't Exist in the Form You Expect
Before my first week ended, I asked Samuel to show me his maintenance records. He pulled out a school exercise book — the kind with a green cover that costs 30 Kenyan shillings at any stationery shop — and flipped to a handwritten page. Each row was a date, a vehicle number, and a brief description: "battery swap," "charger fault," "tyre." No battery serial numbers. No record of how many cycles a pack had seen before it was swapped. No indication of what happened to the pack that was removed.
This wasn't negligence. It was a rational response to the available tools. Writing down a 20-character battery serial number in a school exercise book is unreliable — the handwriting degrades, the book gets wet, the information can't be searched. Samuel had no system that made capturing battery-level detail worthwhile, so he captured vehicle-level events only.
Stima's data model stores telemetry at the pack level, not the vehicle level. Each SEM-1 module reads the pack serial number from the BMS during initial pairing. When a pack moves between vehicles — which happens regularly in cooperative fleets — the telemetry follows the pack, not the vehicle frame. The history of a pack's discharge curves is preserved across reassignments. This design decision was impossible to make from first principles; it required watching a cooperative manager swap packs between vehicles for six months.
Observation 3: Charging Infrastructure Is the Real Variable
Of the three cooperatives I observed, one had a dedicated charging shed with four commercial-grade AC chargers installed by an NGO as part of a clean mobility grant. One charged from a generator running on petrol during evening hours, with a long extension cord and a consumer-grade power strip. One relied on individual drivers charging at home from whatever outlet was available.
The battery degradation profiles across these three setups were different in ways I hadn't anticipated. The cooperative with commercial chargers had the slowest degradation — expected. The generator-charging cooperative had degradation that was 30–40% faster than the commercial setup, which aligned with what I knew about voltage instability. The surprise was the home-charging cooperative, which showed the most variable degradation profile: some packs that performed almost as well as the commercial setup, others that degraded faster than the generator setup.
The variable turned out to be whether the driver's home outlet was on a dedicated circuit or shared with other appliances. Drivers whose packs were charged on a dedicated circuit showed near-commercial-grade degradation profiles. Drivers whose packs competed with a refrigerator and a water pump for current showed the fastest degradation of any vehicle in the study. This finding shaped Stima's charging quality detection feature, which monitors voltage and current stability during each charging event and flags packs that are consistently being charged on unstable power.
Observation 4: The Economics of Downtime Are Asymmetric
When a boda-boda driver's bike goes offline for a day, the cooperative loses the daily vehicle lease fee (typically 400–700 KES per day for an e-bike). The driver loses their income for that day — usually 1,200–2,000 KES in fares. The cooperative manager spends 2–4 hours coordinating the swap, dealing with the driver, and potentially arranging a tow.
The total economic cost of a single unplanned battery failure, when you add up direct lost revenue, manager time, and occasional tow costs, runs between 3,000 and 6,000 KES per event. Samuel's cooperative was experiencing 3–4 such events per month before the first prototype. That's a monthly drag of 9,000–24,000 KES — on a 40-vehicle cooperative with daily revenue probably in the range of 20,000–28,000 KES total.
Battery failure wasn't a maintenance inconvenience. It was a meaningful portion of operating cost. This asymmetry — where a $30/month monitoring tool could prevent $250–$600/month in direct losses — is why the willingness to pay for Stima is high relative to the price point. The problem is expensive enough that the solution almost always pays for itself.
Observation 5: What Managers Actually Need From a Dashboard
I spent time watching Samuel interact with the early prototype dashboard on his phone. The visualizations that made sense to a data engineer did not map to his mental model of fleet management. A line chart of state-of-health over time required him to interpret a concept he'd never encountered. A percentage number (SOH: 73%) required him to know what 73% meant relative to normal versus failing.
What Samuel could interpret immediately: colors (green/yellow/red for each vehicle), a count of vehicles that needed attention today, and a ranked list of "most likely to fail this week." The ranked list format came directly from watching him scan his two laminated sheets each morning. He was already doing a mental ranking — our dashboard just had to surface a better-informed version of the same mental model he was already using.
Stima's dispatch view leads with the vehicle health ranking. The numerical details are available one tap deeper for operators who want them. This information architecture isn't a UX preference — it's a decision made by watching a cooperative manager try to extract operational meaning from data visualizations at 6am before his drivers started arriving.
What Six Months Taught Us About Building for This Market
The central lesson from the field period: the right product for this market is not a stripped-down version of what an enterprise fleet management tool would be. It's a different product entirely, designed around the specific constraints of cooperative fleet operators in high-growth, infrastructure-variable environments.
Those constraints include: intermittent connectivity, price sensitivity that demands sub-$50/month total cost including hardware amortization, operators who manage fleets via WhatsApp and phone calls, mixed vehicle models with different diagnostic protocols, and charging infrastructure that ranges from commercial-grade to a generator and an extension cord.
Every feature in Stima exists because of something we saw in those six months. None of it was theoretical. That's the only building method that produces software that actually gets used.
Filed under: Field Notes, Product Design · Back to Blog