North Star
₹ recoveredper truck per month
Urja wins when owners recover money they’d otherwise lose. Leading signals: share of mornings the owner opens the brief; share of flags acted on within 24 hours.
A concept for Bytebeam · Tushar Pathak · Product Manager · September 2026
Fleet owners learn where their money leaked at month end. Urja tells them the next morning.
Urja is an AI munshi for Indian truck owners. It reconciles every trip’s diesel, tolls and kilometres from data Bytebeam-class devices already send, and explains each rupee that doesn’t add up — in Hindi or English, on WhatsApp, with the driver’s side attached.
01 · How it started
It started with a lunch stop.
A few months ago I was riding my bike from Bengaluru to Mysore. Halfway, I stopped at a small dhaba for lunch and got talking to a truck owner sitting nearby. I asked how many trucks he owned. “24.”
That changed the conversation. I asked: “How much money did your fleet actually make yesterday?” He couldn’t answer straight away. He depended on his munshi, registers and phone calls, and he would often find out something was wrong at month end, when the money was already gone.
That was one conversation, not a study. But it pointed to a trust and visibility problem, not a dashboard problem.
He doesn’t want another complicated system. He wants to be told every morning what happened yesterday, how much he earned, where the money doesn’t add up, and the evidence. That’s where Urja came from.
Illustrative quotes, not from interviews: composites written to show what fleet owners commonly describe. Real field notes will replace them.
I find out a trip lost money when the diesel bill comes, three weeks later. By then the driver is on another trip and nobody remembers anything.
Every evening I sit with the fuel slips, the FASTag messages and the drivers’ diaries. Two hours, and I still can’t say which trip the missing litres came from.
I don’t want to fight with my drivers. I just want to know which trip to ask about, so I don’t have to ask about all of them.
The GPS app tells me where the truck is. It doesn’t tell me whether the trip made money.
What published research says, to be tested against the field: fuel is about 45–55% of a truck’s operating cost, and diesel leakage is often cited at about 8% of fuel filled (a soft figure; vendor blogs claim more). Those numbers are directional, not proven.
02 · The owner
1 to 20 trucks, run from a phone.
The owner doesn’t sit in a control room. They think in rupees per trip, live on WhatsApp, and often prefer Hindi. Today a munshi reconciles trips by hand from paper slips, days late. Every competitor built a dashboard for an enterprise fleet manager. Nobody built for this person. Bigger fleets describe the same pain, but the first segment is the owner with 1 to 20 trucks.
03 · Why not another dashboard
The market has tools. The owner still finds out too late.
| Product | What it does well |
|---|---|
| Fleetx | Fuel-sensor alerts for diesel drops, FASTag, trip P&L, TMSWhat the small-fleet owner still lacks: Needs an extra fuel sensor; built around a dashboard |
| Intangles | Predictive maintenance, digital twinWhat the small-fleet owner still lacks: Answers “will it break?”, not “where did my money go?” |
| LocoNav | GPS, video telematics, driver scorecardsWhat the small-fleet owner still lacks: Scores drivers without asking their side |
| Samsara | AI agents and daily fleet summaries (US)What the small-fleet owner still lacks: Not built for Indian roads, Hindi, or FASTag |
| Urja | One morning answer in rupees, with evidence and the driver’s side |
04 · Built on Bytebeam
Most of the stack already exists. The new part is the reconciliation.
- existsCAN / DBC parsersfuel level, odometer
- existsStreams30-second telemetry
- existsGeofencespumps, plazas, yards
- newLeakage rules5 rules + confidence
- newBrief + Ask UrjaHindi / English
- existsWhatsApp alertsdelivery at 7 AM
No new hardware: fuel level comes from the vehicle’s CAN bus, and FASTag records come from the toll statement.
05 · How we’ll know
One metric that proves the promise, one guardrail that keeps trust.
Guardrail
< 10%wrong flags
Wrong flags under 10%: flags the driver disputes and the owner accepts as innocent. If Urja cries wolf, trust disappears, and good drivers leave. Watch driver 90-day retention too.
For the SuprFleet bet, the layer above is verified truck-months: the record a lender can finance against.
06 · First 90 days
Can we reliably find money owners didn’t know they were losing?
Days 1–30
Sit with 15 fleet owners and their munshis. Map how a trip is reconciled today, and what they do when the money doesn’t add up.
Days 31–60
Run the five rules on 3 fleets’ real CAN/telematics data. Measure the wrong-flag rate, and improve the rules before any owner sees a flag.
Days 61–90
Ship the morning brief and Ask Urja to the pilot owners. Then go or no-go, on ₹ recovered per truck and the wrong-flag guardrail.
07 · About this prototype
What’s real here and what isn’t.
Sharma Roadlines is fictional, and its 24 trucks and 30 days of trips are simulated. That’s a little above the 1-to-20-truck owner in chapter 02; the rules check each trip the same way at any fleet size. The leakage rules really run on that data, every rupee on screen is computed, and Ask Urja is a live model answering only from it. Nothing here is a Bytebeam product.
08 · The bet
From closed books to credit, with the owner’s consent.
Once the owner trusts the morning answer, the same verified record becomes credit.
Urja is the munshi core: it closes a fleet owner’s books every morning. The SuprFleet bet, my answer to the SuprFleet 2030 brief, builds on it in steps: from closed books to verified truck-months a lender can finance against, then to loans through a lending partner, made only with the owner’s consent.
This is a prototype on simulated data, and the research behind the bet is unverified until each source is opened and quoted.
See the bet
The bet, tab by tab
- Where we playThe jobs we chose, the ones we dropped, and what is structural versus hype.
- ProductWhere it is 5–10x better, and how far each data stream lets it act on its own. Try it in the Flag lab on trip 0926-04.
- TiersWhat each tier does, and who pays for it.
- Lender viewOne truck’s verified record, as a lender would read it.
- PlanThe roadmap, the metrics, and the hypotheses H1–H7 still to test.
- ArtifactsThe strategy doc, the PRD, the deck, the pitch script and the research report.