Automotive AI agent · Parts
Parts-Lookup & Fitment AI Assistant
Turn a VIN, a symptom or half a part number into candidates with the build data, supersessions and catalogue conflicts behind each — ordering and the fitment call stay with your parts department.
- Typical delivery
- 4–6 weeks
- Deployment
- Your catalogues
- Fitment call
- Parts manager
- After launch
- Agent Care
What this agent does
Shortlists the numbers, fits none
In
- 01
Take the request as it arrives — a VIN and a symptom, a technician's description, or half a number at the counter.
- 02
Read what the vehicle was built as — model year, build date, plant, option codes and the market it was built for.
Reason
- 03
Search the catalogue the franchise publishes, and follow each number through its supersession chain to the one in force.
- 04
Check the build date against mid-year running changes, and the option codes against fitment that depends on a package.
- 05
Say where the catalogue, the supersession record and the fitment qualifiers do not agree on one number.
Decide
- 06
Hold any number under a stop-sale, a recall or a do-not-install notice, and name the bulletin that stopped it.
- 07
Route safety-critical, option-dependent and unresolved fitment to the parts manager or the technician who verifies it.
Out
- 08
Return candidates ranked, each with the build evidence, the supersession chain and what could not be resolved.
- 09
Retain what was asked, what was searched, what was returned, what was ordered and what came back off the vehicle.
Product statement
The agent searches and evidences. Ordering, pricing, stock commitment and the decision that a part fits this vehicle belong to the parts department, and a technician confirms it at the bench.
Example workflow
One parts request, end to end
- 1Request receivedA VIN and a symptom from the drive, a technician's request at the bench, or a partial number at the counter
- 2Vehicle resolved to its buildVIN decoded to model year, build date, plant, option codes and the market the vehicle was built for
- 3Catalogue searchedThe illustrated group, the fitment qualifiers, the option variants and the links in the supersession chain
- 4Fitment evidence assembledBuild date against running changes, option codes against variants, stop-sale status, and where sources disagree
No human action required
Stages 1 to 4 run without a person in the loop — the decode, the catalogue search and the supersession walk all finish before anyone opens the ticket, and nothing has been ordered at the end of them.
Reaches the counter ready to check and order.
Held with the doubt named, for a person.
Parts manager
Reads the candidates, the build evidence behind each and what the agent could not resolve, checks the number against the vehicle in front of them, then orders under their own name.
Order · Correct · Send to the technician
Ordered — logged back
- 6Candidates filed against the ticketWritten to the counter ticket or the repair order; no order is placed and no stock is committed here
- 7Outcome evaluatedWhat the counter changed, what was ordered, what was returned and what came back off the vehicle later
Returns
A part returned, or swapped at the bench, is scored against what the agent proposed.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Placing an order or committing stock.
- Quoting a price, a core charge or availability.
- Certifying that a part fits a vehicle.
- Releasing a number under an open stop-sale.
Automation boundary
Agent acts unaided
- ✓Decode the VIN to the build recorded against that vehicle.
- ✓Walk each supersession chain to the number in force today.
- ✓Check build date, option codes and market against the fitment record.
- ✓Name what is unresolved and where two catalogue sources disagree.
The agent writes candidates to the ticket, never to the order. A named person orders, and a technician confirms the fit.
- Substituting a non-genuine or will-fit part.
- Deciding that a recall or campaign applies.
- Clearing a safety-critical part without a technician.
- Changing catalogue precedence or supersession overrides.
Example output
One parts request, annotated
Everything the agent proposes is attached to the build it was read from and the catalogue entry behind it.
Lookup output · one counter request
Illustrative example- Technician asked
- Front pads, judder under braking
- Came from
- A repair order at the bench
- Decoded to
- One VIN, one build date
- Returned
- Two candidates, one held
- Confidence
- 87%
- Fitment
- Confirmed at the bench
As received
The request as the technician wrote it and the VIN it arrived against — nothing on this side is inferred.
Evidence used
- Build date and option code
- Supersession chain walked
- Open stop-sale check
Why one was held
A stop-sale is open on the second number. The hold is on every tier, not an upsell.
What the score decides
Confidence ranks what the counter checks first. It never decides that a part fits.
Action
OrderCorrectSend to the technicianValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every parts request
Counter, bench or the service drive
Search what this VIN was built as
Use the build record, the option codes, the market, the supersession chain and the fitment qualifiers the catalogue carries against that vehicle.
Put the evidence beside the number
The build date, the option code and the chain that led to the number arrive with the candidate, instead of being reconstructed across three catalogue screens.
Say plainly what is unresolved
Where the sources contradict each other, where the option code is missing and where the part is safety-critical, the request waits for the person who verifies it.
Build an evidence trail
Retain the request, the build decoded, the sources searched, the candidates returned, the stop-sale check, what was ordered and what came back — on both paths.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Parts catalogues
- OEM catalogue
- illustrated groups
- Interchange
- aftermarket fitment data
Vehicle build data
- VIN decode
- option and build codes
- Factory build records
- in-service data
DMS & parts inventory
- CDK Global
- Reynolds
- Tekion
- Parts stock, orders and returns
Agent
Parts lookup & fitment
- Decodes the build
- Walks supersessions
- Holds stopped numbers
Recall & stop-sale sources
- Campaign bulletins
- stop-sale notices
- Recall lookups
- equipment recalls
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the part you fit
Each control wraps the one inside it. A candidate clears every layer before the counter reads it, and the order itself sits outside all six.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Decide which numbers may be offered at all
- L3
- Holds a number nobody may fit today
- L4 – L5
- Leave the fitment call with a person, trail kept
- L6
- Pulls automation back when signals degrade
How Nestack evaluates it
Evaluate the whole lookup — not only the number at the top.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the candidate list the counter opens
- E1Final-output evaluationWas the number the counter ordered the one that fitted?
- E2Step-level evaluationDid it read the right build date, option codes and market?
- E3Tool evaluationDid it search the right catalogue and follow the chain to its end?
- E4Stopped-number recallDid a stopped, recalled or superseded number reach the counter?
- E5Slice evaluationHow does accuracy change across build, system and franchise?
- E6Business outcomeWhat was returned, refitted, or came back off the vehicle?
Floor — the part that stayed on the vehicle
Failure modes
Where each failure originates in the agent
Seven failure modes plotted against the five stages of the agent lifecycle. None of them orders a part — the stop-sale hold, the counter's own check and the technician's verification at the bench are the controls that stop them.
Agent lifecycle
Direction of processing →
01 · Build decode
2 modes
- PF-01
Built after the running change
The build date is ignored and the model year answers instead.
- PF-02
Option code not in the build
A package-dependent variant is inferred from the trim name.
Stage resolves
The VIN, the build date, the option codes and the market
02 · Catalogue search
2 modes
- PF-03
Two catalogue sources disagree
Two sources give two numbers and the conflict is dropped.
- PF-04
No fitment published yet
A new build is not covered by the illustrated group.
Stage searches
The illustrated group, the qualifiers and the interchange
03 · Supersession
1 mode
- PF-05
Chain stops one link early
A number replaced twice is offered as the current one.
Stage follows
Each number through to the one in force today
04 · Output
1 mode
- PF-06
Stopped number reaches the counter
The stop-sale was posted after the last catalogue sync.
Stage returns
The candidates the counter reads and orders from
05 · Change / Version
1 mode
- PF-07
One number splits into two
A bulletin re-splits a part and open tickets are not re-checked.
Stage tracks
Model, catalogue, bulletin and stop-sale changes
- Sev-1 · the wrong part goes on a vehicle
- Sev-2 · the counter is pointed at the wrong variant
- Sev-3 · the request stalls and is worked by hand
Affected slices
Two vehicles, one model year, different parts
March and August builds can share a model year and still take different parts. The builds that moved mid-year, and the variants an option package decides, carry most of the wrong numbers. Nestack reports accuracy by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Mid-year running-change builds | 5.2% | 2.6× | Review | |
| Option-dependent brake variants | 4.3% | 2.2× | Review | |
| Long supersession chains | 3.5% | 1.8× | Watch | |
| Routine service consumables | 1.2% | 0.6× | Normal |
Bar: wrong-number rate lift vs. routine-consumable baseline · scale 0–4.0× · tick at 2.0×
2 of 4 slices over threshold
Evidence-linked improvement
A part that came back off the car is a test case
A return, or a number swapped at the bench, is not a counter mistake to absorb. It changes the build rules, the supersession walk or the source we trust.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Returns, refits or a wrong-number report move in a cohort.
- 02
Diagnose
The build decode, source precedence, the supersession walk, the option qualifier and the stop-sale sync are each ruled in or out.
- 03
Improve
Precedence and qualifier rules change under your parts department's change control, with a named approver.
- 04
Verify
Re-run on stored requests from that build, system and franchise, including the ones that came back.
- 05
Learn
The returned number becomes a standing check, and the reason enters your parts department's own notes.
Learn → Detect
The return edge. The next request is searched against one more conflict your counter has already resolved the hard way.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, build data and catalogues, matching and holds, evaluation, integration, then supervised lookup and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Parts-lookup discovery and boundaries.Week 1
- 02VIN decode and build-record access.Week 1, Week 2
- 03Catalogue, interchange and source precedence.Week 1, Week 2
- 04Supersession chain and obsolescence rules.Week 2
- 05Option-code and running-change qualifiers.Week 2, Week 3
- 06Symptom and partial-number matching.Week 3
- 07Stop-sale, recall and hold checks.Week 3, Week 4
- 08Safety-critical routing to the technician.Week 4
- 09Evaluation suite, slices and return replay.Week 4, Week 5
- 10Candidate write-back to the counter ticket.Week 5
- 11Catalogue-conflict reporting and escalation.Week 5, Week 6
- 12Observability, deployment and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne franchise, one catalogue | ProductionProduction DMS and catalogue integration | AdvancedMulti-rooftop / multi-franchise |
|---|---|---|---|
| Introduced at Pilot | |||
| Candidates with VIN build-record evidence | ✓ | ✓ | ✓ |
| Supersession chains walked and shown | ✓ | ✓ | ✓ |
| Option-code and running-change qualifiers | ✓ | ✓ | ✓ |
| Stop-sale, recall and hold checks | ✓ | ✓ | ✓ |
| Ordering and fitting stay with your people | ✓ | ✓ | ✓ |
| Catalogue-conflict reporting | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Multiple catalogues in one lookup | — | ✓ | ✓ |
| Candidate write-back to the counter ticket | — | ✓ | ✓ |
| Observability and evaluation | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Aftermarket and interchange sources | — | — | ✓ |
| Multi-rooftop and multi-franchise controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on franchises and catalogues in scope, DMS and build-data access, request volume, safety-critical routing and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01The catalogues you hold and the interchange sources you trust
→Catalogue, interchange and source precedence
Week 102How you reach build data and VIN decode for the makes you carry
→VIN decode and build-record access
Week 203Your supersession, obsolescence and substitution rules
→Supersession chain and obsolescence rules
Week 204Where stop-sale, recall and hold notices reach your parts department
→Stop-sale, recall and hold checks
Week 305Which parts a technician must verify before anything is ordered
→Safety-critical routing to the technician
Week 406A year of returns and the counter tickets that produced them
→Evaluation suite, return replay and failure-mode testing
Week 407Named parts staff and a technician to check what is proposed
→Counter write-back, then supervised lookup
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Phases are drawn over the weeks they actually occupy. Week 5 carries both the conflict reporting and the first lookups your counter orders from.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Parts-lookup discovery, catalogue sources and source precedence
W2Build-record access, VIN decode and supersession rules
W3Symptom matching, option qualifiers and stop-sale checks
W4Evaluation suite, return replay and safety-critical routing
W5Counter write-back, conflict reporting and the first supervised lookups
W6Your counter orders from the candidate list, then Agent Care starts
- Reading the band
- Stop-sale and hold checks are built in week 3, before any candidate reaches a counter in week 5. A number nobody may fit should never be offered in the first place.
- At the end of W6
- Your counter has been ordering from candidate lists beside its own catalogue searches, with every part fitted logged against the number that was proposed for it.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Automotive AI agent
Build a parts-lookup agent around your own catalogues.
Show us a month of counter tickets and the returns behind them — the VINs, the numbers ordered and the ones that came back. We'll search the same requests against your catalogues and mark every wrong number the build data would have caught first.
- Typical delivery
- 4–6 weeks
- Deployment
- Your catalogues and DMS
- Fitment call
- Your parts department
- After launch
- Agent Care available
Nestack Agents · Parts lookup & fitment
AGT-AUT-05 · Agent Care available after launch