PartPakka — fitment concierge for India’s spare-parts counters

1. One-liner

WhatsApp bot that turns a mechanic’s messy “Swift 2015 diesel” into the exact part number before dispatch.

2. Trend signal — why now?

The auto aftermarket has the highest return rate of any industry — 22.6% of all auto parts get returned, and 86% of those returns are wrong-fitment, not damage or shipping. That’s not a US-only number; the cause — too many variants under one model badge — is worse in India, where a single “Swift” spans petrol/diesel/CNG/AMT across a dozen production batches with different part specs. Indian aftermarket trade blogs now openly call wrong-part orders “one of the biggest time killers in Indian workshops”: the job stalls, the lift stays blocked, the customer calls every two hours, and the shop eats the courier return or absorbs a sealed-item it can’t restock.

What changed in the last 12 months: (1) multilingual LLMs now parse Hinglish voice notes and free-text (“Splendor ki clutch plate, 2018 wala”) reliably and cheaply; (2) the Vahan number-plate → make/model/variant API now covers ~200M Indian vehicles via resellers; (3) US fitment-AI startups (Visual Fitment shipped image-based vehicle ID in Mar 2026, AutoPartsAgent.ai opened Cohort 2 in May 2026) proved conversational fitment cuts returns 20–30% — but every one of them assumes a structured VIN and a clean ACES/PIES catalog that the Indian counter does not have.

Provenance:

Signal 1 (Demand): Indian workshop trade content documents wrong-part orders as a top time-killer driven by intra-model variant confusion; Brazilian Reclame Aqui shows the same wrong-part complaint pattern in another emerging market — https://autodukan.com/blog/why-spare-parts-do-not-fit-even-for-the-same-car-model-in-india/ — 2026-06-20
Signal 2 (Feasibility): Vahan RC/number-plate → make-model-variant API live with pan-India coverage (~200M vehicles) via Eko / carregistrationapi, plus cheap Hinglish-capable LLMs and vision-based vehicle/part ID — https://eko.in/developers/eps/rc-verification-api/ — 2026-06-20
Signal 3 (Economic): US conversational-fitment startups funded and scaling (Visual Fitment Mar 2026; AutoPartsAgent.ai Cohort 2 May 2026); industry data confirms fitment AI cuts returns 20–30% on a $16.4B India aftermarket growing 8%+ CAGR — https://visualfitment.com/blog/how-ai-vehicle-identification-works — 2026-06-20 Category: Geographic arbitrage

3. The opportunity

The US fitment-AI wave is real but it’s built for e-commerce checkout: a buyer who knows their VIN, filtering a clean ACES/PIES catalog. India’s spare-parts trade is the opposite world — 60–80% of orders arrive by phone and WhatsApp, in Hinglish, from a mechanic who knows the car is “white Swift, diesel, 2015-ish” and nothing more. The counter guy guesses the variant from memory and a dog-eared printed catalog. When he guesses wrong, the shop pays: courier both ways, a blocked job at the garage, and a mechanic who next time calls a competitor.

Incumbents (Vyapar, Marg, VasyERP) sell GST billing and inventory. VasyERP has manual “vehicle-wise part mapping,” but nobody resolves a fuzzy natural-language request into the correct variant + part number automatically. That resolution step — the thing that actually causes the wrong part — is unowned. We own it, sit on top of whatever catalog the shop already has, and answer inside WhatsApp where the order already lives.

4. Target market

Primary customer: Independent auto and two-wheeler spare-parts retailers and distributors in Indian Tier-1/2 cities — 1–4 counter staff, ₹10–80L monthly turnover, taking most orders by phone/WhatsApp from mechanics, garages, and fleet operators.
Why they buy: “Galat part chala gaya” is a daily event. Each wrong dispatch = courier round-trip (₹100–400), a sealed part they may not restock, an hour of staff back-and-forth, and erosion of the mechanic relationship that drives repeat orders. They feel it in cash, not abstraction.
Rough TAM reasoning: India aftermarket is $16.4B (2024) and highly fragmented across “unorganized local retailers, informal distributors, and independent garages.” Even a conservative count puts spare-parts counters in the low hundreds of thousands. Capturing 5,000 paying shops at ₹1,500/mo is ~~₹9 Cr (~~$1.1M) ARR — a fraction of a fraction.
Why now for them: WhatsApp is already their order channel; service (customer-initiated) conversations are free in India. The only missing piece — turning messy input into a confirmed part — just became cheap.

5. Product sketch (MVP)

Mechanic messages the shop’s WhatsApp (text, voice note in Hindi/Hinglish, or a photo of the worn part / number plate / chassis sticker).
Bot resolves the vehicle: number-plate → Vahan make-model-variant, or parses “Swift 2015 diesel manual” into a specific variant, asking one clarifying question only when fitment genuinely forks (petrol vs diesel, AMT vs manual).
Maps the request to the shop’s own catalog/inventory (uploaded Excel, Marg/Vyapar export, or printed-catalog photos OCR’d at onboarding) and returns the exact SKU + a confidence flag.
Shows the counter staff a one-line “confirm before dispatch” card: part number, fitment basis, in-stock qty, last price.
Flags cross-reference / supersession (“this OEM no. replaced by X”) and obvious mismatches before the bill is cut.
Daily log of resolved vs guessed orders and a running “wrong-dispatch avoided” counter — the ROI receipt the owner sees.

6. AI angle — what’s load-bearing

Remove the AI and there is no product. The entire value is collapsing unstructured, code-switched, voice-and-photo input into a structured variant + part number — exactly the task that defeats a printed catalog and a busy counter. LLM language understanding (Hinglish, abbreviations, mechanic slang) + vision (read a number plate, a chassis sticker, or an old part) + fuzzy-match against a messy local catalog is the whole engine. A dumb keyword search over the same catalog reproduces today’s wrong-part rate; that’s the control we beat.

7. Localization angle

This is the localization play — it cannot be a generic global product. Hinglish/regional-language voice, no-VIN-culture (number plate and chassis sticker instead), Vahan as the canonical vehicle DB, WhatsApp as the order channel, ₹1,500/mo pricing where a $49 US tool can’t land, and catalogs that live in Excel and printed books rather than ACES/PIES. A US fitment tool dropped into this market resolves nothing on day one.

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

Pricing: ₹1,499/mo per shop (single counter), ₹2,999/mo multi-counter/distributor tier with team seats + analytics. Annual prepay discount.
ACV: ~₹18,000–36,000/shop/year.
Rough math to $1M ARR: ~4,600 shops × ₹1,500/mo × 12 ≈ ₹8.3 Cr ≈ $1M.
Rough math to $5M ARR: ~12,000 shops blended at ₹2,700/mo (distributor mix) — requires winning distributor accounts that pull their downstream retailers on, plus a parts-supplier-funded tier (suppliers pay to keep their cross-reference data accurate in the resolver).
Expansion path: seats → distributor analytics → supplier-side data revenue → a marketplace nudge (“you don’t stock this; nearest distributor does”) without becoming a marketplace dependency.

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

Distributor-led pull: sign 3–5 regional parts distributors; each has 50–200 downstream retailer customers. Onboard the distributor free, give their retailers PartPakka at a discount — the distributor benefits from fewer wrong-order returns flowing back upstream. One distributor partnership = a cluster of shops.
Auto-parts market walk: physically work the dense spare-parts bazaars (Kashmere Gate Delhi, Sardar Patel/Kurla Mumbai, Pudupet Chennai). Live demo: ask the owner for yesterday’s worst wrong-part WhatsApp, run it through the bot. Close on the spot with a 14-day free trial.
Mechanic/garage WhatsApp groups & YouTube: vernacular auto-repair creators and city-level garage WhatsApp groups already exist; a 60-second “send a voice note, get the right part” clip is shareable and lands the demand side that pressures shops to adopt.
Billing-software piggyback: integrate as an add-on for Marg/Vyapar/VasyERP users (export their catalog in, results back) and co-sell through their reseller channels.

10. Build complexity — justification

Medium. Off-the-shelf: WhatsApp Business API (Indian service conversations free), multilingual LLM, vision OCR, Vahan reseller API. Custom work is the messy-catalog ingestion + fuzzy resolver and the variant-disambiguation logic — that’s real engineering but bounded, not research-grade. A 2–3 person team ships a credible v1 in ~3–4 months; the hard yards are catalog onboarding per shop, which is ops, not R&D.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Vahan data via licensed resellers; standard B2B SaaS.
Ethical — no harm / dark patterns	✅	Reduces waste and customer friction; transparent confidence flags.
Market exists (evidence above)	✅	$16.4B aftermarket, documented wrong-part pain, free WhatsApp channel.
1–5 person team can build this	✅	2–3 people, ~4 months.
Launchable with <$50K / ₹40L	✅	API + cloud + field sales; well under cap.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	16/20	Daily, costs real cash and relationships; not hair-on-fire existential, but felt every day.
Demand evidence	15	12/15	Strong industry data (22.6% returns, 86% fitment) + India-specific trade documentation; weaker on direct verbatim shop quotes.
Build feasibility	15	11/15	All inputs off-the-shelf; catalog ingestion + resolver is the bounded custom work.
Distribution clarity	15	12/15	Distributor pull + physical bazaar sales are concrete and cheap; conversion math still to prove.
Revenue mechanics	15	11/15	Price fits wallet, free WhatsApp helps margins; ACV modest, needs volume.
Time to first revenue	10	8/10	Trial-to-paid in a market walk can close in days; onboarding adds friction.
Defensibility	10	6/10	Moat is accumulated catalog/cross-reference data + distributor relationships + workflow lock-in, not tech.
Total	100	76/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy (multilingual + vision + fuzzy catalog resolver) · operations-heavy (per-shop catalog onboarding, field/bazaar sales).

Key assumptions to validate (3–5)

Assumption: Shops will pay ₹1,499/mo to cut wrong-dispatches. How to test: 30 in-person interviews across two parts bazaars; pre-sell 10 annual plans before building beyond a demo.
Assumption: The resolver hits ≥90% correct-part on real messy WhatsApp orders. How to test: collect 500 historical wrong/ambiguous orders from 5 shops, measure resolution accuracy offline.
Assumption: Catalog onboarding per shop is fast enough to scale (<1 day). How to test: time-box ingestion of 10 real shop catalogs (Excel, Marg export, printed-book photos).
Assumption: Distributors will champion adoption to cut upstream returns. How to test: sign one distributor pilot; measure retailer activation rate.

Risk flags

Platform dependency: WhatsApp API policy/pricing changes and Vahan API access via resellers are both single points of failure. Keep a voice/IVR and web fallback.
Data-quality drift: Garbage shop catalog → wrong answers that erode trust fast; onboarding QA is existential, not optional.
Market timing / incumbent move: Vyapar/Marg/VasyERP could bolt on a fitment resolver; speed and distributor lock-in are the only defense.
Low ACV grind: ₹1,500/mo means thousands of small accounts; churn and support load can swamp a tiny team without the distributor-cluster motion.

14. Structured verdict

Score:                  76/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder fluent in Hindi/regional language + an ops co-founder who can work the parts bazaars
Time to revenue:        6–10 weeks (market-walk trial-to-paid)
Capital to launch:      ₹8–15 lakh ($10–18K)
Top 3 assumptions to validate first:
  1. Willingness to pay ₹1,499/mo — 30 bazaar interviews + 10 pre-sold annual plans
  2. ≥90% correct-part resolution on 500 real messy historical orders
  3. Catalog onboarding under 1 day/shop across 10 real catalogs
Kill criteria:
  - Abandon if resolver accuracy stays <85% on real messy orders after two iterations
  - Abandon if <3 of 30 interviewed shops pre-pay an annual plan
  - Abandon if Vahan API access or WhatsApp service-conversation economics change such that per-shop cost exceeds ₹500/mo

15. Next step — 1-week validation sprint

Day 1–2: Sit in 2 spare-parts shops in one bazaar; collect 200–300 real historical WhatsApp/phone orders, tag each as resolved-clean vs wrong/ambiguous. Quantify the true wrong-part rate and per-incident cost.
Day 3–4: Run those messy orders through a throwaway LLM+Vahan resolver against the shops’ actual catalogs. Measure correct-part accuracy and how often a single clarifying question fixes ambiguity.
Day 5: Pitch ₹1,499/mo to 10 owners with their own resolved orders as proof. Go if resolver ≥85% accurate AND ≥3 owners verbally commit to an annual prepay; no-go otherwise.

The result is falsifiable: a hard accuracy number on real orders and a hard count of prepay commitments — not vibes.

Interested in a detailed proposal?

Get a deep-dive with market research, competitive analysis, and implementation roadmap.

info@startupbasket.ai