DieselSaaf — fuel-slip checker for small Indian transporters

1. One-liner

Snap every diesel slip on WhatsApp; we flag the inflated, fake and duplicate ones against the truck’s GPS distance.

2. Trend signal — why now?

Fuel is 50–60% of an Indian transporter’s running cost, and a documented 15–20% of it leaks. Three things changed in the last 12 months that make a software-only product viable where it wasn’t before:

1. Vision-model inference collapsed in price. Self-hosted VLM-OCR became “167× cheaper than vendor APIs as of October 2025.” Reading and structuring a crumpled, thermal-printed Hindi diesel slip — at a price that survives a ₹/truck/month plan — is now economic. It wasn’t 18 months ago.
2. The fraud is industrialised. “FreeFuelBill.in — India’s #1 free fuel bill generator” exists and is popular. Drivers and munshis can mint a clean, GST-looking diesel bill for a fill that never happened. The counterfeit side is now a product; the detection side is not.
3. Money is moving here. India transportation & logistics tech raised $1.6B in H1 2025, +104% YoY (Porter hit unicorn). Hardware fuel sensors sell at ₹9,000–₹22,500/truck with a sub-90-day payback claim — proof transporters already pay real money to stop this exact bleed.

The wedge is the gap between (1) and (3): hardware vendors catch siphoning but not paper fraud, and the under-40-truck long tail can’t afford the hardware anyway.

Provenance:
  - Signal 1 (demand): Indian fleets lose ₹15,000–₹25,000/truck/month to fuel theft; 15–20% of fuel purchases leak; owners reconcile slips by hand. — https://www.sahajgps.com/how-gps-tracking-is-stopping-fuel-theft-in-indian-fleets-real-stories-and-proven-savings/ , https://heavyvehicleinspection.com/blog/post/fuel-theft-in-india-the-complete-fleet-playbook-2026-tata-motors-crane-case-prevention-strategy — 2026-05-17
  - Signal 2 (feasibility): Self-hosted VLM-OCR became 167× cheaper than vendor APIs as of Oct 2025; LLMs now cross-check documents for inconsistencies and produce explainable fraud alerts. — https://modal.com/blog/8-top-open-source-ocr-models-compared , https://taktile.com/articles/llms-investigative-partners-fraud-detection — 2026-05-17
  - Signal 3 (economic): India transport/logistics tech raised $1.6B in H1 2025 (+104% YoY); hardware fuel sensors sell at ₹9K–₹22.5K/truck with <90-day payback — proven willingness-to-pay against this exact loss. — https://qubit.capital/blog/india-mobility-logistics-tech-fundraising , https://www.indiamart.com/proddetail/ultrasonic-fuel-sensor-2852381979791.html — 2026-05-17
  Category: Tech-unlock

3. The opportunity

Hardware fuel-monitoring vendors (Pictor, Yatis, Roadcast, Aditi, TrackoBit) sell a ₹9K–₹22.5K sensor that detects tank-level drops — siphoning. Their margin is the hardware, so they have zero incentive to ship a no-install, software-only product. And the sensor is blind to the dominant SMB fraud:

• Inflated slip: slip says 200 L, 150 L went in the tank, ₹/L of diesel in cash split between driver and pump attendant.
• Ghost fill: a slip with no fill at all, generated by a free bill app.
• Rate padding: slip charges ₹3–5/L above that pump’s actual board rate that day.
• Duplicate: same slip submitted across two trips.
• Off-route fill: a fill logged 80 km from where the GPS says the truck was.

Broad TMS players (Fleetx, Fleetable) sell wide enterprise platforms at “price on request” to fleets big enough to have a fleet manager. Nobody is selling a sharp, single-job, ₹-tier forensic tool to the owner of 8 trucks who today gets a rubber-band bundle of slips from his munshi and has no way to know which are fake. That owner is the customer. AI is load-bearing: without vision + cross-referencing there is no product, just another expense register.

4. Target market

• Primary customer: Owner of a 5–40 truck road-transport business in India (FTL, parcel, contract carriage, market-load) — ₹50L–₹5Cr annual turnover, Tier-2/Tier-3 base (Indore, Nagpur, Ludhiana, Hosur, Vijayawada), runs operations from a phone plus a munshi with a paper register. Often the owner-driver-turned-fleet-owner.
• Why they buy (in their words): “Diesel kha jaata hai driver, slip dekh ke kuch samajh nahi aata.” The loss is the single biggest controllable line item and it is invisible to them today.
• Rough TAM reasoning: India has ~12 million goods vehicles; the long tail of 5–40 truck operators numbers in the low hundreds of thousands of businesses. Capturing 8,000–15,000 of them at ₹150–₹250/truck/month is a ₹15–45 Cr ARR business — comfortably inside the $1–5M band, far below VC-scale, ideal bootstrap.
• Why now for them: Diesel at record nominal levels makes the % leak hurt in absolute rupees; smartphones and WhatsApp are now universal even with drivers; FASTag/GPS data exists on most trucks already for e-way/tracking reasons, so the cross-reference inputs are sitting there unused.

5. Product sketch (MVP)

• WhatsApp slip capture: driver or munshi sends a photo of every diesel slip to a number; bot replies in Hindi/regional language with a confirmation and the truck it was tagged to.
• Vision extraction: litres, rate, amount, pump name, date/time, vehicle number pulled from the slip — including handwritten and faded thermal prints.
• Cross-check engine: each slip scored against (a) the truck’s GPS/FASTag distance since last fill vs. its known mileage band → expected litres; (b) that pump’s prevailing diesel rate for that date/state; (c) duplicate-slip hash; (d) fill location vs. truck’s actual route.
• Red/amber/green verdict per slip with a one-line reason in plain language (“Slip 210 L but truck ran 380 km since last fill at 4 km/L → expected ~95 L. Overstated by 115 L (₹10,400).”).
• Monthly fraud statement per truck and per driver — ranked leakage, repeat-offender pumps, trend.
• Owner dashboard (mobile web): total verified vs. flagged spend, money recovered estimate, driver scorecard.
• Munshi mode: bulk upload of a day’s slips; tie-out report the owner can act on at month-end settlement.

6. AI angle — what’s load-bearing

Remove the AI and there is no product. Two AI jobs are core: (1) vision — reliably reading a greasy, handwritten or thermal Hindi/regional slip into structured fields is the thing that was impossible-at-this-price until late 2025; (2) reasoning over inconsistency — combining slip data with GPS distance, mileage priors, and pump rate tables to produce an explainable “this is why this slip is suspect” verdict the owner trusts enough to confront a driver with. A dumb expense app that just stores slips already exists and nobody small uses it. The defensible work is the judgement, in the owner’s language, that turns a pile of paper into a list of names and rupees.

7. Localization angle

This is India-first by construction, not by decoration:

• Language/script: slips are Hindi/Marathi/Telugu/Punjabi mixed with English numerals; capture and the verdict reply must be vernacular over WhatsApp — a generic global expense tool cannot do this.
• Distribution rail: WhatsApp is the only channel a truck driver will reliably use; no app install.
• Pricing: ₹150–₹250/truck/month works where a $20/truck Western telematics SKU is a non-starter.
• Data rails: FASTag transaction data and Indian state-wise daily diesel price tables are the cross-reference backbone — India-specific public/semi-public data.

8. Business model — path to $1M–$5M ARR

• Pricing: ₹199/truck/month (annual) — anchored well below the ₹9K–₹22.5K hardware sensor and framed as “recover one inflated slip and it’s paid for the year.”
• ACV: a 12-truck fleet = ₹199 × 12 × 12 = ₹28,656/year (~$340).
• To ₹1 Cr ARR (~$120K): ~350 fleets averaging 12 trucks.
• To ₹8 Cr ARR (~$1M): ~2,800 fleets — reachable via transporter associations and fuel-card/GPS resellers.
• To ₹40 Cr ARR (~$5M): ~14,000 fleets, plus expansion SKUs: tyre/AdBlue/repair-slip fraud, driver-trip settlement automation, a per-flagged-rupee success fee tier for larger fleets.
• Expansion path: ACV grows by (a) more trucks per fleet as they grow, (b) adding the tyre/maintenance/AdBlue slip modules, (c) a recovered-money success-fee tier for 40+ truck fleets.

9. Go-to-market wedge — first 100 customers

• Transporter associations: India has dense district-level lorry/transport associations (Bombay Goods Transport Assn, Indore, Namakkal, Sangli). Pitch the office-bearers a free 30-day pilot for 20 member fleets; one credible “we caught ₹1.4L in a month” testimonial inside an association WhatsApp group of 400 owners sells itself. Target 3 associations in month 1–2.
• Fuel-card & GPS-tracker resellers as a channel: thousands of local GPS/FASTag resellers already sit in front of exactly this customer and have nothing to upsell that isn’t hardware. Revenue-share them ₹50/truck/month — they make the intro, we close on WhatsApp.
• The Quora/YouTube complaint surface: there is a standing, searched body of content (“How does a tank lorry driver cheat a petrol pump owner in India?”, petrol-bunk-scam YouTube channels with large audiences). Sponsor/seed 3 vernacular trucking creators with a 90-second “catch your driver” demo; their comment sections are the exact buyer.
• Cold WhatsApp to scraped fleet directories: TCI/transport directories and IndiaMART transporter listings give phone numbers; a 20-second vernacular video of a flagged slip → expect 3–6% pilot uptake on a high-pain, free-trial offer.

If I can’t get 5 of the first 20 association pilots to convert at ₹199/truck, the idea is wrong — and I’ll know in 60 days.

10. Build complexity — justification

Medium. Off-the-shelf: WhatsApp Business API, a vision model for slip OCR, standard web stack, public state diesel-price tables. Custom work: the cross-check engine (mileage priors per vehicle class, GPS/FASTag distance reconciliation, duplicate detection) and robust vernacular vision on genuinely bad slip images — the long pole. A 2–3 person team ships a credible v1 in ~12–16 weeks; the model accuracy on real grimy slips needs an iterative data loop with pilot fleets, which is why this is Medium not Low.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Analysing a customer’s own fuel slips and their own trucks’ GPS. No regulated data.
Ethical — no harm / dark patterns	✅	Surfaces verifiable inconsistency; owner decides action. Driver-facing transparency recommended in onboarding.
Market exists (evidence above)	✅	Hardware vendors at ₹9K–22.5K with sub-90-day payback prove paid demand.
1–5 person team can build this	✅	2–3 people, ~12–16 weeks to v1.
Launchable with <$50K / ₹40L	✅	Inference + WhatsApp API + small team; well under ₹40L to first revenue.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	17/20	Largest controllable cost line, leaking 15–20%, felt every month-end, no current visibility. Hair-on-fire for the owner.
Demand evidence	15	12/15	Hardware vendors charging ₹9K–22.5K with payback claims = hard paid signal; standing complaint corpus. Docked: no direct evidence the software-only cut converts yet.
Build feasibility	15	11/15	Mostly off-the-shelf, but vernacular vision on bad slips + the reconciliation engine need a real iteration loop.
Distribution clarity	15	11/15	Associations + reseller channel are concrete and warm; conversion math still unproven.
Revenue mechanics	15	11/15	Pricing well-anchored below hardware; ₹1Cr needs only ~350 fleets. Churn/retention on a “set and forget” tool is the risk.
Time to first revenue	10	7/10	Pilot-to-paid in ~6–8 weeks; needs a 30-day proof window before money.
Defensibility	10	4/10	Execution + vernacular vision data loop + association relationships. Copyable in 12 months; bet is speed and channel lock-in.
Total	100	73/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy (vision + reconciliation engine is the product) · operations-heavy (association/reseller channel and pilot hand-holding are how it sells).

Key assumptions to validate (3–5)

1. Assumption: Owners will act on a flagged slip — confront/dock a driver — rather than ignore it to avoid losing the driver in a tight labour market. How to test: 15 in-person owner interviews; ask what they did the last time they suspected fuel fraud and whether they’d act on a system’s flag.
2. Assumption: Vision extraction hits ≥90% field accuracy on real grimy/handwritten Indian diesel slips. How to test: collect 300 real slips from 5 pilot fleets, measure extraction accuracy before any pricing conversation.
3. Assumption: GPS/FASTag distance is obtainable for the target fleet’s trucks (many small fleets’ trackers are cheap/closed). How to test: audit what tracker/FASTag data 20 pilot fleets can actually export; design a manual odometer fallback if not.
4. Assumption: ₹199/truck/month converts at ≥25% of association pilots. How to test: run 3 association pilots, measure paid conversion at day 30.

Risk flags

1. Retention risk: once a fleet “cleans up” its drivers, perceived value may drop and they churn. Mitigation: shift to ongoing driver scorecard + expand to tyre/maintenance slips so it stays a standing control, not a one-time cleanup.
2. Data-input dependency: value collapses if drivers/munshis stop sending slips (the people being policed control the input). Mitigation: tie capture to settlement — no slip submitted, no reimbursement at month-end; owner enforces.
3. Labour-market risk: driver shortage means owners may tolerate known fraud rather than lose a driver — caps willingness to act, and thus willingness to pay.
4. Platform dependency: WhatsApp Business API policy/pricing changes are a single point of failure for capture.

14. Structured verdict

Score:                  73/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder (vision/ML) + an ops/sales partner wired into Indian transporter associations
Time to revenue:        8–10 weeks (30-day pilot → paid)
Capital to launch:      ₹8–15 lakh ($10–18K)
Top 3 assumptions to validate first:
  1. Owners will act on flags despite driver-shortage — 15 in-person owner interviews
  2. ≥90% vision accuracy on 300 real grimy slips — measure before pricing
  3. ≥25% paid conversion from 3 association pilots at ₹199/truck — day-30 metric
Kill criteria:
  - Abandon if vision accuracy on real slips stays <85% after a 300-slip iteration loop
  - Abandon if <3 of 20 association-pilot fleets convert to paid at ₹199/truck within 60 days
  - Abandon if >40% of target fleets cannot export any GPS/FASTag distance data and reject manual odometer entry

15. Next step — 1-week validation sprint

• Day 1–2: Sit in 2 district transporter-association offices in one Tier-2 city. Interview 12–15 fleet owners (5–40 trucks): “Last time you suspected diesel fraud — what happened, what did you do, would you pay ₹199/truck to be told which slips are fake?” Collect 100+ photos of real diesel slips on the spot.
• Day 3–4: Run those 100+ slips through an off-the-shelf vision model with a hand-built prompt. Manually score field-extraction accuracy. Build the reconciliation logic for 10 slips by hand using whatever GPS/odometer data those owners can produce — see if the math actually surfaces a real fraud the owner agrees is fraud.
• Day 5: Go / no-go on a single falsifiable bar: ≥8 of 15 owners say “yes, I’d pay ₹199/truck for this” AND vision extraction ≥85% on the 100 real slips AND at least 2 reconstructed cases flag a discrepancy the owner confirms is real. Anything less → revise the cut or kill.