Main site

Automotive AI agent · Sales & service

Dealership Sales & Service Copilot AI Agent

Draft the reply, pull the vehicle, deal and repair-order history, and set out what the advisor should ask next — with prices, coverage and repair authorisation left to the person who sends it.

Typical delivery
4–6 weeks
Deployment
Your DMS & CRM
Who sends
Your staff
After launch
Agent Care

What this agent does

Drafts for the advisor, sends nothing

In

  1. 01

    Pick up the lead, the inbound text or email, and the repair order the advisor has open in the CRM or DMS.

  2. 02

    Pull the deal file, service history, declined work and coverage as recorded against that VIN.

Reason

  1. 03

    Read the thread and set out what the customer has already been told, by whom, on what channel and when.

  2. 04

    Assemble what the advisor is missing — the last promise made, prior visits and the recall status on file.

  3. 05

    Draft the reply in the store's own approved wording, and name the record behind each line it puts up.

Decide

  1. 06

    Hold back any price, payment, trade figure, incentive, coverage call or repair authorisation.

  2. 07

    Flag thin, stale or conflicting history, and say what the advisor should confirm before sending.

Out

  1. 08

    Hand the advisor a draft, a summary of the lead or repair order, and the questions worth asking.

  2. 09

    Retain the draft, what the advisor changed, what was sent, by whom, on which channel and when.

Product statement

The agent drafts, retrieves and summarises. Prices, payments, coverage calls, repair authorisation and the send itself stay with a named person at the dealership.

Example workflow

One customer message, end to end

  1. 1Message or repair order picked upInternet lead, inbound text, email, or a repair order opened at the service drive
  2. 2Customer and vehicle matchedResolved to one customer record and one VIN before any history is read
  3. 3History gatheredDeal file, prior visits, declined work, coverage and recall status as they are recorded today
  4. 4Reply draftedThe store's approved wording, the questions to ask, and every figure left blank

No human action required

Stages 1 to 4 run without a person in the loop — matching, retrieval and drafting all finish before an advisor opens anything, and nothing has reached the customer yet.

5DecisionSplits on match confidence and on what the reply would have to carry
Clean match, nothing held back

Reaches the advisor ready to read and send.

Weak match or a figure needed

Goes to the advisor with the gap named.

Sales or service advisor

Reads the draft, the history behind it and what the agent left blank, supplies any figure themselves, then sends it under their own name.

Send · Edit · Ask for more

Sent — logged back

  1. 6Draft filed against the recordSaved unsent on the customer's record; nothing leaves the dealership at this stage
  2. 7Outcome evaluatedWhat the advisor rewrote, what was sent, whether the match held, and corrections by cohort

Edits

What the advisor rewrites before sending is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Quoting a price, payment, rate or fee.
    • Putting a trade or appraisal figure to a customer.
    • Confirming incentive, rebate or programme eligibility.
    • Stating that a repair is covered under warranty.

  • Automation boundary

    Agent acts unaided

    • ✓Match the customer and the VIN before any history is read.
    • ✓Pull the deal file, prior visits, declined work and the earlier thread.
    • ✓Draft the reply in the store's approved wording, with figures left blank.
    • ✓Summarise the lead or repair order and list what to confirm.

    Drafts are written to the customer's record, never to the customer. A named person supplies any figure and sends it.

    • Authorising work or adding lines to a repair order.
    • Telling a customer a vehicle is safe to drive.
    • Clearing a vehicle of an open recall.
  • Sending anything to a customer, on any channel.

Example output

One drafted reply, annotated

Everything the agent drafts is attached to the record and the history it was built from.

Copilot output · one drafted reply

Illustrative example
Customer asked
Is the recall done, and what will the brakes cost?
Channel
Inbound text, service
Matched to
One record, one VIN
Draft status
Unsent — advisor to send
Confidence
92%
Brake price
Left blank for the advisor

As received

The customer's own words and the channel they arrived on, kept beside the draft — nothing on this side is rewritten.

Evidence used

  • Customer and VIN match
  • Repair-order history
  • Recall status as recorded

Why the price is blank

A brake figure is a quote and a recall answer is a coverage call. Both are the advisor's.

What the score decides

Confidence decides how much the advisor re-checks, not what the customer is told.

Action

SendEditAsk for more

Value

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 lead and repair order

CRM thread, inbound text or the service drive

03Draft & summary

Apply this store's own history

Use the deal file, repair-order history, the earlier thread and the wording this store already approves.

01Approved path

Give the advisor the history first

Prior visits, declined work and the last thing the customer was told arrive together, instead of across four screens.

02Human review

Stop the draft where a person is needed

A reply that would have to carry a price, a coverage call or a recall answer waits for the advisor who owns it.

04

Build an evidence trail

Retain the thread, the sources read, the draft, the advisor's edit, what was sent and on which channel — on both paths.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

  • DMS

    • CDK Global
    • Reynolds & Reynolds
    • Tekion
    • Dealertrack
    • OEM DMS feeds

  • CRM & lead sources

    • VinSolutions
    • Elead
    • DriveCentric
    • Marketplace and OEM leads

  • Service & warranty

    • Repair orders
    • declined work
    • OEM warranty portal
    • recall lookup

Agent

Dealership sales & service copilot

  • Matches the record
  • Drafts the reply
  • Leaves figures blank

  • Customer channels

    • Texting platform
    • email
    • Telephony
    • consent and opt-out lists

  • 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 your customer

Each control wraps the one inside it. A draft clears every layer before an advisor reads it, and the send itself sits outside all six.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeReturn drafting to staff if evaluations or production signals degrade.Roll back
L5TraceabilityRecord the sources read, the draft, the edit, what was sent and by whom.Record
L4Human sendNothing reaches a customer except under a named person who presses send.Gate
L3Consent and channelThe opt-out records we connect are read before a channel is drafted for.Check
L2Record match gateNo history is read until the customer and the VIN resolve to one record.Verify
L1Figure and claim blockA price, a payment or a coverage answer comes from the advisor, not from the draft.Withhold
Model coreReply drafted — the summary, the questions to ask and the figures left blank
L1 – L2
Decide what a draft is allowed to contain
L3
Decides whether the customer may be written to
L4 – L5
Keep the send with a person and the trail intact
L6
Pulls automation back when signals degrade

How Nestack evaluates it

Evaluate the whole draft — not only the sentence the customer reads.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the draft the advisor opens

  1. E1Final-output evaluationWas the drafted reply true to the record it was built from?
  2. E2Step-level evaluationDid it read the right thread, repair order and coverage entry?
  3. E3Tool evaluationDid it resolve the correct customer, VIN and repair order?
  4. E4Withheld-figure recallDid any price, coverage call or recall answer survive into a draft?
  5. E5Slice evaluationHow does draft quality change across customer and vehicle cohorts?
  6. E6Business outcomeHow much did advisors rewrite, and what went out wrong anyway?

Floor — the message the customer is finally sent

Failure modes

Where each failure originates in the agent

Seven failure modes plotted against the five stages of the agent lifecycle.

Agent lifecycle

Direction of processing →

  1. 01 · Match / retrieval

    2 modes

    • DL-01

      Wrong record, shared number

      A household's shared mobile pulls the other buyer's deal file.

    • DL-02

      Recall status read as current

      The portal was last synced before the campaign was posted.

    Stage resolves

    The customer, the VIN, the deal file and the repair order

  2. 02 · Summarising

    1 mode

    • DL-03

      Goodwill read as coverage

      A past goodwill repair is summarised as warranty cover.

    Stage reads

    The thread, the history and what the customer was told

  3. 03 · Drafting

    2 modes

    • DL-04

      Old payment figure repeats

      A number from an earlier thread returns and is sent unread.

    • DL-05

      Spanish drafts thin out

      Detail and the store's disclosure line are lost in translation.

    Stage drafts

    The reply, the questions to ask and the blanks left

  4. 04 · Handoff / send

    1 mode

    • DL-06

      Draft made for a stopped number

      The opt-out landed in the texting platform, not the CRM.

    Stage presents

    The draft the advisor edits and sends in their name

  5. 05 · Change / Version

    1 mode

    • DL-07

      Silent template regression

      A wording change quietly weakens what drafts withhold.

    Stage tracks

    Model, prompt, wording and programme changes

  • Sev-1 · a figure or claim reaches the customer
  • Sev-2 · the advisor is briefed wrongly
  • Sev-3 · drafts thin out, advisors rewrite
See the complete Automotive failure-mode catalogue

Affected slices

Some customers get a worse draft than others

Rewriting is not spread evenly. A language the store rarely writes in, a model year with thin history, and a customer whose file sits at another rooftop carry most of the advisor corrections. Nestack reports draft quality by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Spanish-language customer threads6.4%2.9×Review
First-model-year vehicles5.0%2.3×Review
Cross-rooftop customer records4.0%1.8×Watch
Routine repair-order updates1.5%0.7×Normal

Bar: advisor-rewrite-rate lift vs. routine-update baseline · scale 0–4.0× · tick marks 2.0×

2 of 4 slices over threshold

Evidence-linked improvement

Every rewrite becomes a rule about what a draft may say

An advisor deleting a line from a draft is not a one-off correction. It changes the wording template, the withheld-figure list or the record the copilot reads.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Advisors are rewriting more of the draft, or a blocked figure got through.

  2. 02

    Diagnose

    The store reads back which part failed — the match, the history, the wording or the block list.

  3. 03

    Improve

    The wording template and the block list are re-cut, and the version is stamped.

  4. 04

    Verify

    The same threads are drafted again, the ones that went out wrong included.

  5. 05

    Learn

    The thread is kept, and what a draft may no longer say enters the store's wording rules.

  6. Learn → Detect

    The return edge. A change to what a draft may say is signed off by whoever answers for the store's advertising, before it ships.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, matching and retrieval, drafting, evaluation, integration, then production validation and handover.

Workstream

Week 1Week 2Week 3Week 4Week 5Week 6

  1. 01Workflow discovery and automation-boundary definition.Week 1
  2. 02DMS, CRM and OEM portal access assessment.Week 1, Week 2
  3. 03Approved wording, disclosures and withheld figures.Week 1, Week 2
  4. 04Customer and VIN matching rules.Week 2
  5. 05Deal, service and repair-order retrieval.Week 2, Week 3
  6. 06Lead and repair-order summarisation.Week 3
  7. 07Reply drafting in the store's own wording.Week 3, Week 4
  8. 08Consent, channel and opt-out checks.Week 3, Week 4
  9. 09Evaluation suite and withheld-figure testing.Week 4, Week 5
  10. 10Cohort slice testing, language and rooftop.Week 5
  11. 11CRM draft write-back and advisor handoff.Week 5, Week 6
  12. 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 tierPilotOne store, one channelProductionSales and service, liveAdvancedMulti-rooftop / group
Introduced at Pilot
Drafted replies and lead summaries✓✓✓
Withheld figures, coverage and recall answers✓✓✓
A named person sends every message✓✓✓
Consent, channel and opt-out checks✓✓✓
Baseline evaluation✓✓✓
Store wording and disclosure templates✓✓✓
Introduced at Production
Deal, service and repair-order history—✓✓
CRM draft write-back and advisor handoff—✓✓
Observability and evaluation—✓✓
Introduced at Advanced
Drafting in additional languages——✓
Multi-rooftop and group controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on rooftops, DMS and CRM integrations, message volume, OEM portal access, wording approvals 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.

  1. 01Your approved wording, disclaimers and what a reply may never say

    Approved wording, disclosures and withheld figures

    Week 1

  2. 02Read access to the DMS, CRM and the OEM portals you use

    DMS, CRM and OEM portal access assessment

    Week 2

  3. 03How customers and vehicles are matched today, duplicates included

    Customer and VIN matching rules

    Week 2

  4. 04Your consent, opt-out and contact-hours rules for each channel

    Consent, channel and opt-out checks

    Week 3

  5. 05Threads that went wrong, including the ones a manager had to fix

    Evaluation suite, regression cases and failure-mode testing

    Week 4

  6. 06The languages your customers write in and the rooftops they use

    Cohort slice testing across language and rooftop

    Week 5

  7. 07Named advisors and a manager to read what the copilot drafts

    Advisor handoff workflow, then supervised drafting and validation

    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 slice testing and the first drafts an advisor sends.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Automation boundary, approved wording and what a draft may never carry

W2DMS and CRM access, customer and VIN matching

W3History retrieval, summarising and the first drafted replies

W4Evaluation suite, withheld-figure testing and consent checks

W5Draft write-back, slice testing and the first supervised sends

W6Advisors work from live drafts, then Agent Care starts

Reading the band
Withheld-figure testing runs in week 4, before an advisor sends anything drafted here.
At the end of W6
Advisors have been sending from drafts beside their own replies, with every send logged against the draft it came from.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Automotive AI agent

Build a copilot around the way your advisors already answer.

Show us a week of real threads — the internet leads, the service texts and the repair orders behind them. We'll draft against them in your own wording and show you every line the copilot refused to write.

Typical delivery
4–6 weeks
Deployment
Your DMS, CRM and OEM portals
Who sends
A named person, always
After launch
Agent Care available

Nestack Agents · Dealership sales & service copilot

AGT-AUT-01 · Agent Care available after launch

Build a dealership copilot

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your setup

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.