Fitment copilot for collision-repair groups that turns estimates into exact parts orders before supplement delays hit.

1. Horizontal copilots are not credible in this workflow because general AI still scores only 1-5% on automotive parts fitment, leaving a clear opening for software built around a domain-specific model.
2. The jump to 60% F1, combined with five years of human-feedback training data from 50+ manufacturers, indicates fitment interpretation has crossed the minimum quality threshold needed for production exception-routing software.
3. A $100B collision-repair market spread across 250,000+ independent shops is large enough to build a venture-scale company even before expansion into distributors, insurers, or adjacent repair workflows.
4. DST Global's unusual investment into an industrial-vertical AI company signals that both capital and customer attention are moving toward proprietary workflow AI in sectors long ignored by horizontal software.

Catalyst. Partly's reported jump from 1-5% generic-model accuracy to 60% F1 on fitment, backed by five years of human feedback and 50+ manufacturer datasets, suggests this workflow just crossed from "too risky to automate" into "narrow enough for production pilots."

The product is a workflow layer that sits between estimating software and parts ordering. It pulls the VIN, build context, and line items from a repair estimate, proposes exact-fit part candidates across OEM, aftermarket, and recycled options, and flags ambiguity before the order is sent. High-confidence recommendations flow directly into the buyer's existing ordering process, while edge cases go to a parts specialist with the model's reasoning and supporting attributes attached. Every accepted, rejected, returned, or supplemented part becomes feedback that improves future recommendations at the shop-group level and creates a defensible operational dataset over time.

What's different. This is not another catalog database or a broad "AI for auto repair" assistant. The wedge is the transaction-control layer at the moment an estimate becomes an order: confidence scoring, exception routing, and feedback capture on returned or supplemented parts. That closed-loop dataset compounds faster than static catalog content and makes the product more useful with every shop group deployed, while shops keep their existing estimating and supplier systems.

Jobs to be done

flowchart LR
  Estimate[Repair estimate + VIN] --> Engine[Fitment copilot]
  Engine -->|High confidence| Order[Order-ready parts recommendation]
  Engine -->|Low confidence| Queue[Human exception queue]
  Order --> Outcome[Fewer supplements and returns]
  Queue --> Outcome

Executive takeaways

• This is a pre-order decision-control problem, not just an e-commerce problem: Partly's benchmark and MSO proof point tie better parts interpretation to fewer supplementaries, fewer returns, and shorter key-to-key time.[5][8]
• Vertical AI is now credible in this niche: Partly's reported 60% F1 on complex fitment tasks versus 1%-5% for general AI implies a usable quality gap between domain-trained and horizontal models.[7][111]
• ADAS is compounding urgency: CCC says calibration incidence climbed from 0.9% of repairable appraisals in 2017 to more than 23% in 2025, and Caliber projects up to 60% of collision repairs will require mandated calibrations in 2025.[40][92]
• Regional MSOs are the right beachhead because the market is consolidating but still fragmented: Focus counts 800+ independent MSOs even as the Big Five continue to take share.[95][96]
• Incumbents validate spend but leave whitespace: Orderly, CCC, Mitchell, and CollisionLink all help transact or document parts workflows, yet the clearest gap remains confidence-scored fitment gating before an order is placed.[15][21][26][49][58]

Market definition

The relevant market is pre-order collision-parts decision software: tools that ingest VIN and estimate data, resolve the correct part path across OEM, aftermarket, and recycled options, and route uncertain lines to a human before a PO is sent. It sits between estimating systems, supplier networks, OEM repair data, and downstream procurement platforms, and excludes generic shop management or pure post-order AP tools.[2][7][26][49][60][63][69]

Customer and buyer

Daily users are repair planners, estimators, and centralized parts coordinators who must translate a VIN and estimate lines into orderable parts under time pressure. The economic buyer is usually the VP of operations, director of parts procurement, or regional MSO leader centralizing process across 10-50 shops. External market structure points to the middle market rather than tiny independents or national consolidators: Focus tracks 800+ independent MSOs, while FenderBender's survey shows the average shop profile is already operationally substantive.[42][95][96][99]

Buying triggers

• Supplement volume rises after centralizing buying or taking on newer, ADAS-heavier vehicle mix. [37][40][92]
• Supplier scarcity and backorders make estimate cleanliness and electronic ordering more important to get the right part quickly. [38][45][103]
• An MSO wants to standardize procurement across locations without replacing the estimating system its shops already use. [9][21][95]

Willingness to pay

Willingness to pay is credible because this workflow sits next to large existing costs. CCC says parts are the largest portion of repair cost, HLDI reports collision claim severity of $8,739, FenderBender reports $5,000+ average repair orders and 11-day key-to-key cycle time, and AAA shows ADAS components alone can add $1,540 to a minor front collision repair. A product that prevents even a small share of wrong-part returns or supplement delays can pay for itself quickly. [45][90][91][99][103]

Category dynamics

Growth signal 0.8% CAGR to 2030 for the U.S. collision repair market (top-down cross-check)

Validation signals

• Partly's MSO use case claims 2.7x fewer supplementaries, 2.4x fewer returns, and 20% lower key-to-key time after standardizing procurement.
• Crash Champions rolled Orderly across 650+ repair centers, proving large MSOs will standardize procurement software when workflow value is clear.
• CCC says shops write nearly 20% more supplements now than they did six years ago.
• CCC reports calibration incidence above 23% of repairable appraisals in 2025 and nearly one-third in DRP claims.
• FenderBender's survey puts the average shop around 11 days key-to-key with $5,000+ repair orders, confirming the operational value of fewer wrong-part loops.

Regulatory & technical constraints

• Fitment logic must normalize VIN data and aftermarket fitment taxonomies across vPIC, ACES, and VCdb rather than rely on free-text descriptions alone.
• State disclosure and consent rules require accurate OEM versus non-OEM labeling in estimates and invoices.
• ADAS-sensitive components must respect OEM repair and calibration guidance before a recommendation is treated as order-ready.
• The product has to coexist with incumbent estimating and ordering stacks, including parts code tables and OEC-linked order flows.

The competitive set splits into four buckets: AI-first fitment infrastructure (Partly), open procurement platforms (Orderly/PartsTrader), OEM-direct procurement networks (CollisionLink), and broad estimating/workflow suites (CCC and Mitchell). All validate demand for better parts workflow, but most start after the part choice is already mostly known. The whitespace is a confidence-scored decision layer that resolves ambiguous lines earlier, documents why a part was chosen, and learns from accept/reject/return outcomes across shops.[2][12][15][21][26][49][58][110]

Why incumbents do not win by default

• Estimating and workflow suites. CCC and Mitchell are embedded systems of record, but they do not win by default because fitment reasoning is one module inside a broad platform rather than a dedicated decision-control layer with closed-loop learning from returns, supplements, and exceptions.
• Procurement marketplaces and networks. Orderly and CollisionLink are strong at live quoting, supplier routing, DRP compliance, and OEM-direct ordering, but they assume the shop can already identify the right line-item path more often than this startup does.
• Domain-specific AI platforms. Partly validates that a specialist model can outperform generic AI, yet its platform scope spans estimation through procurement and broader network infrastructure, leaving room for a lighter-weight pre-order gating product inside existing North American workflows.
• Manual expert workflows. The default substitute remains senior estimator or parts-manager judgment layered over supplier portals, spreadsheets, and phone calls; it is trusted, but it does not scale well across regional MSOs facing staffing pressure and rising vehicle complexity.

Collision supplement fitment copilot is a pre-order decision layer for regional U.S. collision-repair MSOs that centralize parts buying and suffer recurring supplement and wrong-part delays. The first product ingests VIN and estimate lines, ranks order-ready parts for bumper, lighting, and ADAS-adjacent repairs, and routes ambiguous lines into a human exception queue before a purchase order is sent. The company should not start as a full procurement marketplace or shop-management suite; the fastest proof comes from showing measurable reduction in supplementaries, returns, and key-to-key cycle time inside the systems shops already use. The first customer is a 15–30 shop MSO whose centralized parts desk is overloaded by newer vehicle complexity and weekly supplement loops. Pricing should combine a location- based platform fee with a per-reviewed-order fee so ROI tracks the workflow being improved rather than user seats. If the product can achieve shadow-mode proof and then production gating without heavy custom integration, it can build a durable dataset of accept, reject, return, and supplement outcomes that incumbents do not capture cleanly today. Research supports the workflow pain and the incumbent gap, but it does not yet show the minimum confidence threshold buyers will trust or which estimator and BMS stacks are easiest to standardize first. The main strategic risk is whether this narrow beachhead is large and defensible enough before incumbents or a model supplier absorb the feature, so expansion beyond regional collision MSOs must be validated early rather than assumed.

Problem

• Centralized collision parts teams still translate VIN, trim, and estimate lines into exact OEM, aftermarket, or recycled parts through fragmented catalogs, phone calls, and senior staff judgment.
• Wrong-part decisions create supplementaries, return freight, and longer key-to-key cycle time, and the penalty is rising as ADAS-sensitive and disclosure-sensitive repairs become a larger share of the mix.

Solution

• Insert a pre-order fitment gate between estimating and ordering that proposes exact-fit candidates, scores confidence, and routes ambiguous lines into a human exception queue before a PO is sent.
• Capture accept, reject, return, and supplement outcomes at the line-item level so each MSO builds an auditable rule and feedback layer without replacing its estimator or supplier network.

Why we win

• The wedge is the estimate-to-order control point with the clearest ROI, so the startup can prove value faster than a broader procurement or shop-management product.
• Closed-loop outcome data, insurer and supplier rule libraries, and auditable handling of ADAS and OEM disclosure edge cases can compound into a moat that static catalog tools do not have.

Milestones

flowchart LR
  Wedge[Shadow-mode fitment gate] --> MVP[MVP on bumper, lighting, and ADAS lines]
  MVP --> Proof[Lower supplements, returns, and cycle time]
  Proof --> Expansion[Group rollout plus adjacent repair and network workflows]

Founding team

Experiment roadmap

Risk assessment

What must be true

• Regional 10-50 shop MSOs will fund a pre-order decision layer now instead of waiting for CCC, Mitchell, OEC, or supplier networks to add enough similar functionality.
• A shadow-mode pilot can reduce targeted supplement or return events by at least 15% within one quarter on bumper, lighting, or ADAS-adjacent categories.
• The first estimator and BMS connector path can be deployed without replacing the system of record or turning the company into a services-heavy integrator.
• Buyers will trust confidence thresholds and human exception routing enough to let the product block or reroute some line items before order submission.
• Adjacent expansion into insurers, distributors, or other repair workflows is real enough to grow beyond a small collision-only software market.

Open diligence questions

• Which estimator and BMS stacks dominate 10-50 shop regional MSOs, and how different are their integration paths?
• What KPI threshold convinces an MSO to convert from a pilot to a multi-shop annual rollout?
• How much licensed OEM or supplier data is required to reach production-grade coverage on the first part categories?
• How defensible is the workflow if Partly, CCC, Mitchell, or OEC launches a similar confidence-scored gate?
• Do state disclosure and OEM procedure rules create purchase urgency or mainly raise adoption liability concerns?

Model sanity

• Revenue engine. Base revenue is driven by turning 4 paying accounts at the end of Y1 into 20 rolled-out MSOs by Q4Y3 while mature account revenue rises toward the research-backed per-order spend level.
• Must go right. The implementation playbook has to stay repeatable enough that one founder-led GTM motion and two solutions-oriented hires can support 10 accounts by Q4Y2 without crushing gross margin.
• Model breaks if. If pilot-to-rollout timing doubles or mature account ARPU stalls near the low end of the pricing range, the downside case burns toward a sub-$200K cash floor before the seed story is ready.
• Next-round proof. The next financing case is 15 rolled-out MSOs by Q2Y3 with one adjacent-workflow design partner and quarterly burn already below $100K.

Scenarios

Sensitivity

flowchart LR
  Pipeline[Qualified MSO pipeline] --> Pilots[Paid pilots]
  Pilots --> Rollouts[Rolled-out MSOs]
  Rollouts --> ReviewedOrders[Reviewed repair orders]
  ReviewedOrders --> Revenue[Subscription + usage revenue]
  Revenue --> GrossProfit[Gross profit]
  GrossProfit --> Cash[Cash and runway]

Flags: The base case effectively reaches the researched 20-account SOM rollout by the end of Y3, so venture upside still depends on the adjacent expansion path described in the business plan. · Gross margin does not hit the stated 70% target until Y3, so repeatable connectors and disciplined scope control are mandatory. · customersEop includes both paid pilots and rolled-out accounts in Y1, so true production-logo count trails the headline customer number early in the model. · Cash is modeled as EBITDA with no working-capital timing, deferred revenue benefit, or capex drag, which is reasonable at this stage but not exact.

• Workflow adoption friction. Parts teams may resist any tool that interrupts ordering speed if early recommendations are noisy or exception queues are slower than current habits. Mitigation: Launch as a shadow-review mode that measures missed errors and confidence quality before turning on ordering gates, then enable intervention only for high-risk line items.
• Dependency on external fitment intelligence. If the startup depends too heavily on a third-party model or catalog source, margin and roadmap control could be constrained. Mitigation: Design the product as a model-agnostic orchestration layer, retain all customer feedback data, and build internal evaluation sets so providers can be swapped as accuracy and economics evolve.
• Integration bottlenecks. Estimating and supplier-ordering systems may not expose clean interfaces, slowing deployment and making pilots expensive to support. Mitigation: Start with MSOs willing to pilot via estimate exports and lightweight workflow overlays, prove ROI, then prioritize repeatable connectors for the two or three systems most common in the beachhead.