Emerging-market commerce · India
Reaching the unreachable garage: how local-language AI converts a $32 billion aftermarket
India's auto-parts demand runs through millions of independent mechanics who order in Kannada, Tamil, Telugu, and Malayalam, and who buy a part by sight rather than by part number. Working with an industry-leading automotive parts company, we are rebuilding the buying interface around how these mechanics already work, and the market follows.
India builds and repairs vehicles at a scale few markets match, yet the channel that keeps those vehicles running is the one digital commerce has never managed to enter. The country's auto-component industry turned over roughly $78.7 billion in FY25 and is on track to reach $200 billion by 2030. The aftermarket inside it, the parts that flow to repair shops every day, sits near $11.6 billion today and is projected to approach $32 billion by 2026. The demand is enormous, growing, and almost entirely offline at the point where it is decided: the workbench of an independent mechanic.
That mechanic is the buyer no platform has reached. Conventional parts marketplaces are built for someone who can type an English query, recognize a SKU, read a catalog, and complete a web checkout. The mechanic ordering a wheel bearing in Hubballi does none of those things. The gap between how parts commerce is built and how the buyer actually operates is the entire reason this market stays unconverted, and it is the gap we are helping an industry-leading automotive parts company close.
Exhibit 1
India's auto-parts aftermarket is on track to roughly triple, and most of it runs through an unorganised garage channel.
2024
FY25
2026, projected
Aftermarket sized within a broader auto-component industry of ~$78.7B in FY25, projected to reach ~$200B by 2030. The aftermarket is served largely by the unorganised sector of independent garages. Source: IBEF / ACMA, Indian auto-components industry (FY25); McKinsey, "Shaping the future of India's auto component industry."
The ProblemWhy this market resists every interface built for it
The wall is not affordability or willingness to buy. Mechanics buy parts constantly. The wall is language and interface, and it runs three layers deep.
India transacts in its own languages, and parts platforms do not speak them. English is the primary language of under 1 percent of Indians and a comfortable second language for only about a tenth. Meanwhile the Indian-language internet user base climbed from 234 million in 2016 toward 536 million by 2021, far outpacing English users. Roughly nine in ten of the country's new internet users consume content in a local language, and about 70 percent trust local-language content more than English. A mechanic in Kerala or Karnataka is online and transacting, in Malayalam, in Kannada, on platforms that meet them there. Parts commerce has not.
Exhibit 2
Nine in ten of India's next internet users transact in a language today's parts platforms don't speak.
Source: KPMG-Google, "Indian Languages: Defining India's Internet"; Statista, India internet user base by language, 2011 to 2021.
Mechanics identify parts by sight, rather than by part number. A worn alternator, a cracked bush, a specific oil seal, the mechanic knows it in the hand, not as an alphanumeric SKU buried in an OEM catalog. Asking that buyer to translate a physical object into a typed English search string is asking them to do the one thing the platform should be doing for them.
The point of sale is the shop floor, rather than a browser. Orders are decided mid-repair, on a low-end Android phone, often inside WhatsApp, sometimes on patchy connectivity. A standalone app that demands a download, an onboarding flow, an account, and a stable connection loses the buyer before the first query.
The market is not unwilling to convert. It has simply never been addressed in the language, the modality, or the place where the buying decision is actually made.
Coded to SolutionMapping each wall to a capability that removes it
The design is a direct, one-to-one response to those three layers of friction. Every capability we have built exists to dissolve a specific barrier that has historically lost the mechanic.
Exhibit 3
The new platform removes friction at the exact points where conventional platforms lose the mechanic.
The wall
Part identification. The mechanic recognizes a worn part by sight and has no SKU, part number, or English name to type.
Coded to
Reverse image search. Snap a photo of the old part; the vision model matches it to the exact catalog SKU and compatible alternatives.
The wall
Text and English literacy. Typing a query, in English or in a regional script, is slow, unfamiliar, or simply not possible.
Coded to
Voice in local languages. Speak the request in Kannada, Tamil, Telugu, or Malayalam; on-device speech models capture intent, dialect, and automotive vocabulary.
The wall
Checkout complexity. Multi-step carts, address forms, variant pickers, and payment flows in English break the transaction.
Coded to
Custom agents and actions. An agent confirms the part, quantity, and price by voice, then places the order, applies pricing, and schedules delivery end to end.
The wall
Access and onboarding. Standalone apps demand downloads, accounts, data, and connectivity the shop floor doesn't reliably have.
Coded to
Meet them where they work. Lightweight and WhatsApp-native, running on low-end Android and tolerant of intermittent connectivity through local inference.
Platform architecture developed with the company, mapped to observed adoption barriers in India's independent-garage channel.
The EngineWhy local AI is the enabler, rather than a feature
Cloud-only AI cannot serve this buyer at this scale and price. Local, on-device models are what make the entire approach viable, for four concrete reasons.
Latency that matches the workbench
A mechanic mid-repair needs a part match in seconds. On-device inference answers in the shop, without a round trip to a distant server, so voice and image queries resolve at the speed of the conversation.
Unit economics that survive millions of low-ARPU users
Per-query cloud inference across millions of high-frequency, low-value orders would erase the margin. Running matching and intent on the device pushes the marginal cost of a query toward zero, which is the only way the math works at the bottom of this market.
Tolerance for the connectivity that actually exists
Garages sit in basements, in dense markets, in towns with uneven signal. Local models keep core identification and voice working through dead zones, syncing the order when the connection returns.
Dialect and domain coverage generic models miss
Automotive vernacular, the colloquial Kannada or Tamil name for a specific bush or seal, is precisely what general cloud models handle worst. Local models fine-tuned on this domain and these dialects close the comprehension gap that would otherwise reintroduce friction.
Exhibit 4
Collapsing four English-first steps into one voice-and-image flow lifts completed orders sharply.
Conventional platform
AI-enabled flow
Indexed to 100 mechanics who begin a parts search. Illustrative model for proposal purposes; conversion rates to be validated in field pilots.
The FlywheelWhy adoption compounds once it starts
The wedge is a single mechanic placing one order by photo and voice. What turns that into a market position is the loop it sets off.
- 1More mechanics adoptEach mechanic served in their own language and modality brings the next, through the dense word-of-mouth networks of local repair clusters.
- 2More part-photo and voice dataEvery query enriches the catalog of real-world part images and the vernacular automotive vocabulary the models learn from.
- 3Higher match accuracy, lower frictionBetter local models mean faster, more confident matches, which raises completion and trust at the workbench.
- 4Demand pull from brands and distributorsA converted, measurable channel of independent garages becomes a distribution asset that parts brands and distributors want to reach through the platform.
Each turn lowers the cost of acquiring the next mechanic and widens the moat against any platform still asking this buyer to type an English SKU.
The WindowThe market is addressable now, on terms a first mover sets
The conditions have converged. The aftermarket is large and compounding. Vernacular and voice are now the default mode of India's internet, rather than a frontier. Local AI has reached the point where on-device vision and speech are good enough, cheap enough, and fast enough to run in a roadside garage. The buyer has been ready for years; the technology to address them has only just arrived.
The platform that meets the mechanic in their language, in their workflow, and at their workbench will define how parts commerce works for the next hundred million vehicles on Indian roads. We are building it with an industry-leading automotive parts company to be exactly that. The market that no one could convert was waiting for an interface designed around it, and that interface is now buildable.