AI for automotive
AI for automotive and dealerships
Fixed ops pays the bills, and unanswered phones leak it away. Every call answered, every lead worked. 5x ROI in 30 days, or we work free.
- Dealership groups
- Service departments
- Parts counters
- Collision centres
Proof, honestly
No published dealership case study yet, and we say so. Founding clients get priority scheduling, direct founder involvement, and a co-published case study once the numbers are real. The 5x ROI guarantee applies from day one.
We guarantee 5x ROI inside 30 days of deployment, in writing, measured against a baseline you sign before we build. If the system misses the bar, we keep working for free until it clears.
Automotive, before and after
The manual path is dashed: Service calls ring out, Leads cold by morning, CSI issues found weeks late. The system path replaces it, and a person approves before anything ships: Calls answered and booked, Follow-up in minutes, any hour, Problem calls flagged same-day.
Before: by hand
- 01Service calls ring outhuman
- 02Leads cold by morninghuman
- 03CSI issues found weeks latehuman
After: the system
- 01Calls answered and booked
- 02Follow-up in minutes, any hour
- 03Your approvalhuman
- 04Problem calls flagged same-day
The research
Dealers hold 29% of service visits, down from 33% in 2018. Among owners of vehicles two years old or newer, 54% went back to the selling dealer for service in 2025, down from 72% in 2023 (Cox Automotive, 2025, US owner survey).
Cox Automotive · 2025 · 1,974 US vehicle owners aged 18 to 75 who had a vehicle serviced in the prior 12 months, fielded April to May 2025, published November 11, 2025
The service line rings out.
Advisors are with customers and the BDC is underwater. The average US dealership wrote 16,252 repair orders in 2025 at $494 per customer RO (NADA Data 2025, US averages). The caller who rings out takes their next RO to the chain store down the road.
After-hours leads go stale.
The internet lead that arrived at 9pm gets a reply at 10am. The buyer moved on before breakfast.
You hear about bad visits in the survey.
CSI problems surface weeks later, after the damage is done and the OEM has counted it.
Declined work dies with the RO.
The June MPI measured the rear brakes at 4 millimetres and the customer said not today. Nobody called in October. The chain store beside their office got the job.
The special order arrived. Nobody called.
The part sits in the bin aging toward the obsolescence write-off while the RO stays open and the customer assumes it never came in.
Every caller starts from zero.
The store holds the VIN, four years of ROs and the last advisor's notes, and the caller still gets asked what car they drive.
Run your numbers.
Your operations
Savings use the low end of our hours-reclaimed range. The math is conservative on purpose.
The math
Calculated at the low end of every range.
Every first build is covered in writing: 5x ROI in 30 days. Or we work for free.
Teams we build for
- Hoyes Michalos
- Nurse Next Door
- Fedi
- UBC Sauder
- Merchant House Capital
- Picton Investments
- Campbell Froh May & Rice LLP
- Barnakl
- Hungerford
- Breez
Teams we build for
- Hoyes Michalos
- Nurse Next Door
- Fedi
- UBC Sauder
- Merchant House Capital
- Picton Investments
- Campbell Froh May & Rice LLP
- Barnakl
- Hungerford
- Breez
Teams we build for
- Hoyes Michalos
- Nurse Next Door
- Fedi
- UBC Sauder
- Merchant House Capital
- Picton Investments
- Campbell Froh May & Rice LLP
- Barnakl
- Hungerford
- Breez
The numbers in automotive and dealerships
Dealers hold 29% of service visits, down from 33% in 2018. Among owners of vehicles two years old or newer, 54% went back to the selling dealer for service in 2025, down from 72% in 2023 (Cox Automotive, 2025, US owner survey).
Cox Automotive · 2025 · 1,974 US vehicle owners aged 18 to 75 who had a vehicle serviced in the prior 12 months, fielded April to May 2025, published November 11, 2025
One multi-location GM group recovered over $853,000 in missed service revenue with AI call handling (Numa, 2025, vendor-reported).
Numa · 2025
Who this is built for
Absorption moves without adding headcount
Fixed ops pays the light bill, and absorption is the number you actually watch. Every recovered call is service revenue that was already dialing your store when it leaked to the shop down the road. The founding tier is the honest version of this pitch: no dealership case numbers yet, we say so on this page, and the 5x ROI guarantee applies from day one against a baseline you sign before we build.
Advisors work the lane, not the hold queue
The morning rush forces a choice between the customer at the drive and the customer on hold, and your advisors lose either way. When the overflow answers itself and books itself, the write-up in front of them gets their whole attention, and the caller still lands on the schedule. The hours booked route to your advisors' numbers, not away from them.
The caller is a known VIN
Four years of ROs, the last MPI, the declined rear brakes and the technician's cause-and-correction notes are read before the second ring. The caller gets asked what changed since October, not what car they drive. When the call needs you, it arrives with the history attached instead of a cold transfer.
Special orders get picked up, not written off
The part lands, the customer gets called the same day, and the appointment books while the job is still warm. The bin stops collecting the cost of phone calls nobody made, and the obsolescence line at year end gets smaller for reasons you can name.
What stays human
- The 7:10 rush is faces, not phones
- She prices the timing chain job herself
- The sour call gets a same-day callback
The steps the day below leaves to a person, by design.
A service advisor's Thursday on the drive
The 7:10 rush is faces, not phones
humanThe drive is four deep. She does the walkarounds, writes up the waiters, gets the loaner keys sorted. The phone rings behind her the whole time and for the first time in her career that is not her problem.
The overflow books itself into real slots
Callers the advisors cannot take are answered, matched to their customer record, and booked into the slots the shop actually set in Xtime or myKaarma, drop-off or waiter, with the op code the concern maps to. The agent says what it is and that the call is recorded, up front, every time.
The grinding noise is already on file
A customer calls about a grinding noise. The history shows rear brakes measured at 4 millimetres on the June MPI and declined. The agent books the inspection and puts the June measurement on the RO notes. It reports what the shop wrote down. It diagnoses nothing.
She prices the timing chain job herself
humanA caller asked what a timing chain replacement runs. The agent took the question and parked a callback proposal with the vehicle's history attached. She builds the estimate from the menu and the labour guide, and the number goes out under her name. Nothing was quoted until she approved it. That gate is the build, not a courtesy.
Status calls never reach the bays
Is-my-car-ready answered from the RO status without a tech putting down a wrench or an advisor leaving a write-up. Promised times that slip trigger a call out before the customer calls in.
June's declined work comes back due
The declined items on closed ROs are watched against mileage and season. The rear-brake reminders queue for customers whose consent is on the ledger, and only those. The database refuses the rest.
The sour call gets a same-day callback
humanA pickup call flagged for CSI risk, a customer back twice for the same leak. She and the service manager hear the recording and call before five, weeks before any OEM survey would have told them. The save is human. The system only made it possible to attempt.
Tomorrow's no-shows shrink tonight
Confirmations go out for tomorrow's appointments. The 8am that cannot make it reschedules by text, and the slot reopens for the next overflow caller instead of dying on the schedule.
What we build for automotive and dealerships
The phone
Service-line voice agent.
Calls answered, appointments booked straight into your scheduler, advisors freed for the customers in front of them.
Parts and status inquiries.
The is-my-car-ready and is-it-in-stock calls answered from the RO and the parts bin without an advisor leaving a write-up.
The service history
Recall and maintenance-due outreach.
Owners contacted on tracked consent when service is actually due, with recall status read per VIN from the manufacturer campaign and the Transport Canada lookup, never inferred.
Declined-work revival.
The MPI's declined items watched against mileage and season, drafted into reminders an advisor approves before anything sends.
CSI issue detection on calls.
Recorded calls flagged same-day when a visit went sideways, so a manager fixes it before the survey lands.
The chase
After-hours lead response.
Internet leads answered in minutes at any hour, with follow-up sequencing that does not give up after one email.
Sales follow-up engine.
Researched, personalized and persistent follow-up on every lead your VinSolutions or Activix CRM already tracks, not just this week's hottest.
Special-order parts chase.
The customer called the day the part lands, the appointment booked while the job is warm, the bin cleared before the write-off.
The economics
Before and after economics
Line
- The phone during the morning rush
Before
Advisors choose between the customer at the drive and the caller on hold, and the overflow rings out.
After
The overflow is answered and booked. Advisors keep the lane. Qualitative on purpose, until the founding numbers are real.
- Phone and follow-up hours in scope
Before
At the page defaults, 8 people across the BDC, the advisors and the parts counter spending 10 hours a week on phone tag, status calls and follow-up is 80 hours, $3,600 a week at $45 loaded cost.
After
The calculator above prices your own baseline, and the 5x ROI guarantee is measured against a baseline you sign before we build.
- Declined work on closed ROs
Before
The MPI's declined items close with the RO and are never seen again.
After
Watched against mileage and season, worked on tracked consent, approved by an advisor before anything sends.
- Special-order parts on the shelf
Before
Parts age toward the obsolescence write-off waiting for an arrival call nobody owns.
After
The customer is called the day the part lands and the appointment books while the job is warm.
No dealership client outcome numbers exist yet, and this table does not pretend otherwise. The only dollar figure is arithmetic on this page's calculator defaults, 8 people at 10 manual hours a week each at a $45 loaded hourly cost, which is 80 hours or $3,600 a week in scope. Every other row is a before-and-after state, not a measured result. The founding engagement's numbers get co-published when they are real.
Where the data comes from
DMS
CDK, Reynolds and Reynolds, PBS Systems or Tekion holds the operating truth of the store: customer records, VINs, open and closed ROs, appointments, parts inventory and the special-order queue. The build reads it through the vendor's integration program or scheduled exports, so a caller matches to a record and a booking lands where the shop already works.
Where it stops. Customer and vehicle records stay in the dealership's accounts. Nothing writes to the accounting side of the DMS, and booking writes go only where the store's scheduler expects them. One rooftop's customers are scoped from another rooftop's at the database layer.
Service history
The corpus. Years of ROs, op codes, technician cause-and-correction lines, MPI results with their measurements, declined items and warranty history, indexed per VIN so the caller is a known car before the second ring. This is what makes the answer grounded in what the shop actually did, not in what a model guesses a shop might do.
Where it stops. The history and its vectors live in your own database instance under row-level security, and API terms are no-training. Your technicians' notes never improve anyone else's product, and nothing personal leaves your accounts to make a model smarter.
Phone system
The lines themselves, the ring groups, the miss log and the recordings. The agent takes the overflow and the after-hours ring that today ends in voicemail, and the store's own recorded calls become the graded eval set that booking accuracy and misroute rate are measured on before any customer reaches the system.
Where it stops. Every call opens with what the agent is and that the call is recorded. Recordings stay in your accounts. Outbound is gated on the consent ledger in the data layer, so a customer who opted out cannot be dialed, because the database refuses the call.
How an answer is found in your service history
A question runs against your service history two ways at once: vector search for meaning and full-text search for exact wording. Reciprocal rank fusion merges both result sets, and the answer carries the source it came from.
- 01A question
- 02Vector search (meaning)
- 03Full-text search (exact wording)
- 04Rank fusion
- 05Answer, with its source
How the system is built for automotive and dealerships
See the full capability mapRetrieval
Service history, vehicle records and past conversations indexed so a caller is recognized and the answer is grounded in what the shop actually did to that car. Nothing annoys a customer faster than being asked twice.
- pgvector
- Full-text BM25
- Reciprocal rank fusion
Agents and orchestration
Agents answer the phone that currently goes to voicemail, book service, and run follow-up on a schedule. Quotes and anything that commits the shop are proposals. A service advisor stays in the loop.
- agent-worker
- Autonomy guard
- Escalation handoff
Evaluation
Graded on real recorded calls your team already handled, end to end including transcription error. Booking accuracy and misroute rate are the numbers that matter, and both are measured before the phones switch over.
- Eval graders
- End-to-end voice grading
- quality-worker
Models
Speech and frontier language models for the conversational core. Ordinary classifiers for call intent and CSI risk flagging, which are faster, cheaper and easier to audit than a large model on the same job.
- Speech models
- Frontier models, one gateway, routed per task
- Intent classifiers
Data boundary
Customer and vehicle records stay in your accounts. Consent state is enforced in the data layer, so an agent cannot call or message someone who asked not to be contacted.
- Supabase row-level security
- Consent ledger
- No-training API terms
DMS, Service history, Phone system feed a hybrid index. The agent runtime works from that index, and every consequential action passes a human approval before it reaches Ask your team, Your AI team, Workflow builder.
Your systems
- 01DMS
- 02Service history
- 03Phone system
The system
- 01Hybrid index
- 02Agent runtime
- 03Your approvalhuman
Where your team works
- 01Ask your team
- 02Your AI team
- 03Workflow builder
What that means in practice
Agents and orchestration
AI that does the work instead of just answering: looks things up, calls your systems, completes multi-step tasks, and knows when to hand off to a human.
Where we stop. Multi-agent swarms are oversold; most jobs need one well-guarded loop. If a cron job and a script solve it, that is what we build, because 90 percent per-step accuracy compounds to 59 percent over five chained steps and no framework changes that arithmetic.
Vision, documents and speech
AI for eyes and ears: reading documents, watching camera feeds, transcribing calls.
Where we stop. Bespoke computer vision is justified by volume and latency, not by novelty. Below that bar, a vision-language model on demand is cheaper to run and easier to maintain, and we will tell you which side of the bar you are on before anything is built.
Classic and predictive ML
Not every problem needs a language model. Predicting numbers, churn, demand, fraud risk, is usually solved better, cheaper and more explainably with proven statistical ML.
Where we stop. When the input is language, judgment or unstructured documents, classic ML underperforms and we say so. The discipline runs both ways: if your problem is a prediction problem, you will hear that it does not need an LLM from us before you pay for one.
The objections
We already pay for Xtime and an after-hours answering service.
The scheduler books the callers who reach it, and the answering service takes a message that gets returned tomorrow. Neither one answers the question, reads the service history, or books the appointment at 9pm. This build does all three and hands your scheduler a booked slot instead of a voicemail.
Our ROs are one-line stories. Customer states noise, could not duplicate.
That is the expected corpus, not a disqualifier. The first phase indexes the ROs, op codes and cause-and-correction lines exactly as your techs wrote them, then grades whether they can answer the questions your own recorded calls actually asked. If four years of one-liners cannot carry the answer, we tell you what the history can support and scope the build to that instead of on top of it.
Advisors are paid on what they write. They will treat this as competition.
The agent answers the calls that currently ring out while an advisor has a customer at the drive. Every hour it books lands on an advisor's RO, under the shop's own dispatch rules. In a commissioned lane an answered phone is money, and work an advisor did not have to abandon a write-up to take is the one intervention the lane never fights.
A wrong price on the phone costs us the job and the customer.
The agent does not price work. A cost question becomes a callback proposal with the vehicle's history attached, and the estimate leaves under an advisor's name after an advisor builds it. That line is enforced in the build, not in a training deck, and it is the same line the guarantee is measured behind.
We are a group with three DMSes across nine rooftops.
One rooftop goes first. The build reads that store's DMS as it is, booking accuracy and misroute rate get measured on that store's own recorded calls, and the next store's phones move only after the numbers hold. A group standard comes out of stores that work, not out of a memo from head office.
Built around your rules
| Regime | What it demands here | How the system complies |
|---|---|---|
| PIPEDA | Customer and vehicle personal information, the VIN-linked service history included, collected and used under Canadian privacy law. | Records stay in dealership-controlled systems, scoped at the database layer, under no-training API terms. Nothing personal leaves your accounts to make a model smarter. |
| CASL | Service reminders and marketing messages need consent, sender identification and a working unsubscribe. | Outreach runs on a tracked consent ledger enforced in the data layer. An agent cannot message an opted-out customer, because the database refuses. |
| BC Motor Dealer Act and the VSA | Licensed dealers carry advertising and disclosure duties under BC's motor dealer regime, administered by the Vehicle Sales Authority. | AI-drafted marketing passes human review against your obligations before it ships. The system drafts. A licensee approves. |
| Call recording disclosure | A commercial organization recording calls says so at the start of the call, under PIPEDA guidance. | Voice agents identify themselves as automated and disclose recording up front, on every call, inbound and outbound. No exceptions are configurable. |
| OMVIC and AMVIC | Dealer conduct and advertising rules under the Ontario and Alberta regulators, all-in price advertising included. | Built to respect them where the group operates. Drafts carry the required disclosures and a registrant approves before anything publishes. |
| Motor Vehicle Safety Act recall campaigns | Recall outreach must state a vehicle's actual recall status. A wrong recall claim is a safety, trust and liability problem. | Recall status is read per VIN from the manufacturer campaign and the Transport Canada lookup, never inferred by a model, and the outreach names the campaign it cites. |
The regimes that govern automotive, what each demands, and how the system complies
Regime
- PIPEDA
What it demands here
Customer and vehicle personal information, the VIN-linked service history included, collected and used under Canadian privacy law.
How the system complies
Records stay in dealership-controlled systems, scoped at the database layer, under no-training API terms. Nothing personal leaves your accounts to make a model smarter.
- CASL
What it demands here
Service reminders and marketing messages need consent, sender identification and a working unsubscribe.
How the system complies
Outreach runs on a tracked consent ledger enforced in the data layer. An agent cannot message an opted-out customer, because the database refuses.
- BC Motor Dealer Act and the VSA
What it demands here
Licensed dealers carry advertising and disclosure duties under BC's motor dealer regime, administered by the Vehicle Sales Authority.
How the system complies
AI-drafted marketing passes human review against your obligations before it ships. The system drafts. A licensee approves.
- Call recording disclosure
What it demands here
A commercial organization recording calls says so at the start of the call, under PIPEDA guidance.
How the system complies
Voice agents identify themselves as automated and disclose recording up front, on every call, inbound and outbound. No exceptions are configurable.
- OMVIC and AMVIC
What it demands here
Dealer conduct and advertising rules under the Ontario and Alberta regulators, all-in price advertising included.
How the system complies
Built to respect them where the group operates. Drafts carry the required disclosures and a registrant approves before anything publishes.
- Motor Vehicle Safety Act recall campaigns
What it demands here
Recall outreach must state a vehicle's actual recall status. A wrong recall claim is a safety, trust and liability problem.
How the system complies
Recall status is read per VIN from the manufacturer campaign and the Transport Canada lookup, never inferred by a model, and the outreach names the campaign it cites.
Customer and vehicle data stays inside dealership-controlled systems, recordings are disclosed, and outreach runs on tracked CASL consent.
How we build it for automotive and dealerships
Answer the phone that currently goes to voicemail.
Missed calls are measurable before the build starts, which makes the baseline honest. This is the fastest payback in the shop.
Grade on your own recorded calls.
Booking accuracy and misroute rate are measured end to end, transcription included, before any customer reaches the system.
Add retention work once booking is reliable.
Service follow-up and CSI risk flagging reuse the same records. They come after the shop trusts the phone.
What we will not automate
Quoting work and anything that commits the shop. Consent state is enforced in the data layer so an agent cannot contact someone who asked not to be.
Where your team works
Tour the platformAsk your team
Chat with a team that already knows your business.
Every agent is briefed on your documents, your data, and your preferences. Ask for the number, the draft, or the plan and cite where it came from.
Your AI team
A roster, not a black box.
Every agent on your account is named, scoped, and inspectable: what it does, what it may touch, and what it has done lately.
Workflow builder
Describe the workflow. Watch it assemble.
Say what should happen in plain language and the builder assembles the automation on a canvas you can read, run, and change.
Cost per outcome
Every dollar of spend traced to the work behind it.
Outcomes delivered, cost per outcome, value attributed, return on spend. The same measurement the guarantee is settled against, live on one page.
The hard questions
Free · 3-5 days
Know your number in five days.
We map your operations, find the highest-ROI automations, and hand you a ranked plan with the payback math attached. Yours to keep, whoever builds it.
Prefer to talk first? Book 15 minutes with James. No pitch deck.