Partly secures $50m from DST Global at $500m valuation for car repair AI beyond GPT-5’s abilities

Partly, a New Zealand-born startup building artificial intelligence for the automotive repair industry, says its specialised model is 12 times more accurate than general-purpose AI when identifying the exact parts needed for a vehicle. The company, which has just secured a $50 million Series B funding round led by DST Global at a $500 million valuation, claims its “Interpreter” foundation model achieves a 60% F1 score on internal benchmarks, compared with just 1–5% for large language models such as GPT-5 and Claude Opus 4.8 that lack domain-specific data.

Funding and valuation

Partly was founded in Christchurch, New Zealand, in 2020 by Levi Fawcett, a former Rocket Lab engineer and four-time entrepreneur, together with Nathan Taylor. The company now operates from a North American headquarters in Austin, Texas, with a second office in San Francisco, and employs 160 people across more than 20 countries. The Series B round was led by DST Global, the investment firm best known for backing consumer internet giants including Facebook, Alibaba, Airbnb, Spotify and Anthropic.

The raise brings Partly’s total funding to $92.4 million, following a $37 million Series A in December 2022 led by Octopus Ventures at a $180 million valuation, with participation from Square Peg, Blackbird, Shasta Ventures and Ten13. Fawcett described the Series B as a partial close and said the company does not expect to raise additional capital until next year. The funds will be directed toward computing power for training the seventh version of Interpreter, expanding the team across Europe, and boosting sales and marketing efforts in the United States.

How Interpreter solves the fitment problem

Partly’s core technology, the Interpreter model, is described by the company as the only foundation model purpose-built for automotive parts. It addresses a problem known in the industry as “fitment”: identifying the precise correct components for a specific vehicle while accounting for manufacturer variations, trim levels, build-plant differences and model-year changes across hundreds of millions of possible configurations. General models were not designed for this level of specificity. “If you ask GPT-5 or Claude Opus 4.8 a question about parts or automotive, they’re going to be very bad — they’ll get it right about 5% of the time,” Fawcett said. “On complex jobs, Interpreter is about four times more accurate than a human, and an order of magnitude more accurate than those models.”

Partly published its own benchmark to support those claims. Tested across 50 Toyota repair jobs, Interpreter achieved a 60% F1 score — a combined measure of precision and recall — while general AI models without catalogue data scored between 1% and 5%. The model was trained on five years of human feedback and synthetic data, supplemented by agreements with more than fifty manufacturers. According to the company, Interpreter now covers 91% of vehicles across the top 58 manufacturers.

The fitment problem is a persistent and costly issue for the automotive aftermarket. Industry estimates suggest that up to 86% of returns in auto parts e-commerce are due to fitment errors. While a number of companies, including GroupBy, WheelPrice and Alhena AI, have developed fitment-focused tools, Partly argues that its foundation-model approach sets it apart. Incumbent players such as Solera, Mitchell1 and ALLDATA offer workflow software and reference databases built on legacy infrastructure, but none operates at the foundation model layer. (Industry comparisons indicate that Mitchell1 is often preferred for user experience and data quality, while ALLDATA is seen as a more budget-friendly option.) Partly’s distinction, the company claims, is that it is building an AI infrastructure layer specifically for automotive rather than layering a new interface onto existing databases.

The broader automotive repair industry is increasingly turning to AI for diagnostics, predictive maintenance and workflow automation. Machine learning algorithms have demonstrated over 90% accuracy in image analysis for damage identification, and AI tools have been shown to reduce human error in repair cost estimations by up to 40%. Companies such as Bosch are developing AI diagnostic assistants, and Bosch recently acquired Uptake Technologies, a specialist in AI-based predictive analytics. Partly, however, focuses specifically on the parts identification and procurement process rather than diagnostics.

Fawcett, who holds a degree in Mechatronics Engineering and previously led the guidance and simulation team at Rocket Lab, also co-founded an e-commerce platform called AllGoods with Taylor before launching Partly. Taylor, who studied computer science and IT management, was named to the Forbes 30 Under 30 Asia list in 2023. Both founders relocated to Austin as Partly established its North American headquarters.

DST Global’s investment rationale

DST Global’s involvement in an industrial B2B company serving collision repair shops marks a departure from its traditional consumer internet focus. However, Fawcett said the firm’s logic aligns with a broader thesis around “physical world AI.” DST Global co-led OpenEvidence’s $250 million Series D in January at a $12 billion valuation — OpenEvidence is described as the most widely used AI platform by doctors in America — and participated in Waymo’s latest funding round, in which the autonomous ride-hailing company raised $16 billion in February 2026 at a $126 billion valuation. “DST have been making a lot of bets around physical world AI — bringing AI into the physical world,” Fawcett told Tech Funding News. “We’re assisting technicians, helping them do better work, taking away the admin.”

Partly also fielded interest from Andreessen Horowitz and Iconiq before DST came in. The scale of the automotive repair market was the decisive factor. “You can build a trillion-dollar business just within this domain,” Fawcett added. The US collision repair market alone exceeds $100 billion annually, spread across roughly 250,000 independent shops — a large, fragmented and historically underserved sector that DST Global has bet on before. Fawcett believes the moment for adoption has arrived. “It really has started working in the last six to nine months. When you talk to the CEOs at any of the large repairers, that is the number one thing they are talking about,” he said.