What working with us looks like
Five phases. At the end of them, your group owns an intelligence infrastructure.
Not a platform you rent. A connected, dealer-owned foundation with your definitions in it, agent teams running real workflows on top of it, and your people trained to operate the whole thing. Here is how it gets built, in order — and where you decide whether there is a next phase.
What this video says
- Auto Agentic is an intelligence engineering firm, not a tool vendor — the work is systems and people advancing together.
- Engagements start inside the systems a store already runs, rather than adding another platform beside them.
- Every stage is gated: the group decides whether there is a next one.
- We engineer the systems
- The connective work between the DMS, CRM, phone, service, inventory and marketing tools you already run.
- We engineer the organization
- Roles, decision rights, training and accountability, redesigned alongside the systems.
- You keep control
- Your data stays inside your boundary, and a named person approves anything that touches money.
- Every stage is a decision
- Each stage produces something you own, then you choose whether there is a next one.
The journey
Five phases. Every one ends in a decision you make.
We engineer two things in parallel: the infrastructure — systems, data and agentic workflows across every rooftop — and the organization — its structure, people, roles and the training they need to operate it. What follows is how both get built: five phases, in order. Each one produces something your group keeps, and each one ends in a decision you make — including the decision to stop.
- Phase A · Understand
Understand the dealership
We learn how the store actually runs and find where workflows are not connecting with one another — DMS, CRM, phone, service, inventory and marketing — and where the numbers disagree with each other.
The inputs that feed Blueprint V1: how the systems can and should be integrated.
Is this worth designing against?
- Phase B · Design
Design both halves
We evaluate what the engineered infrastructure should look like: workflows, data boundaries, roles, decision rights and training drawn together in Blueprint V1 — then verify it against APIs, MCPs and actual data mapping to reach a validated Blueprint V2.
Blueprint V1, then a verified Blueprint V2 your group owns whether or not we continue.
Do we move forward?
- Phase C · Engineer
Engineer the infrastructure
We start to engineer what an AI infrastructure needs to look like for the rooftop and the group to run most effectively — agent teams, permissions, approval points and workflows built alongside the people who will run them.
Working workflows, and named owners trained to run them.
Is it ready for the floor?
- Phase D · Prove
Prove it in the store
A 90-day pilot measured against your own baseline, not a vendor benchmark.
Measured results — or a clean stop with everything you built still yours.
Did the numbers move?
- Phase E · Expand
Expand what has earned it
Scale only the workflows that proved out. The connections, definitions, patterns and training are already in place.
Each workflow after the first starts further ahead.
What earns the next deployment?
- At the end · You own it
What stays with your group
- Blueprint V2 — verified against your APIs, data and systems
- Working workflows with named owners on the floor
- Your definitions, boundaries and decision rights, documented
- People trained to run and extend it without us
The Chassis Programme is curated working sessions, an anonymous all-staff readiness survey and structured evidence gathering — systems, data and infrastructure examined alongside roles, training, decision rights and governance. It produces the Blueprint; it does not include production implementation.
The complete path, and the gates
One path, from an optional assessment through the six-week Chassis programme and Verification, where we evaluate APIs, MCPs and actual data mapping against the Blueprint V1 design so the group can reach a validated Blueprint V2. Nothing rolls forward automatically: each stage produces something your group owns, and then you decide whether there is a next one.
The Chassis journey
From operational truth to an AI capability your dealership can operate and expand.
Each stage has a distinct purpose, deliverable and decision. Nothing rolls forward automatically.
Work stageDealer-owned deliverableYour decision
- Optional orientation01
Assessment
Optional
A quick self-reported orientation to current systems, data and readiness.
Optional, and not evidence enough to assign a final level.
- Work stage 102
Chassis
Six weeks
Both halves examined together — the systems and data, and the roles, training and decision rights around them.
- Curated working sessions with leaders and the people who run the work
- Anonymous all-staff readiness survey
- Structured evidence gathering
Chassis produces the Blueprint. It does not include production implementation.
- Dealer-owned deliverable03
Blueprint V1
Yours to keep
What current evidence supports: candidate workflows, organizational requirements, evidence gaps and the verification work order.
Evidence-backed, but not yet technically verified.
- Your decision04
Verify, revise, or stop.
Gate
You can stop here. The Blueprint belongs to you either way.
- Work stage 205
Verification
Approximately 30 days
Verify the design against the real environment: APIs, MCPs, actual data and how it maps, so Blueprint V1 becomes a validated Blueprint V2.
Findings correct Blueprint V1 in both directions.
- Dealer-owned deliverable, verified06
Blueprint V2
Yours to keep
Corrected architecture, workflow sequence, training, governance, measures, owners and unresolved risks.
A decision packet, not automatic authorization to build.
- Your decision07
Pilot, revise, or stop.
Gate
Nothing is engineered until you decide on Blueprint V2.
- Work stage 308
Pilot
90 days
Engineer and train one bounded workflow, with human approval gates and one or more agreed measures against a baseline.
- One workflow, scoped and bounded
- Training before go-live
- Human approval on anything with consequence
- Your decision09
Expand, revise or stop.
Gate
Expansion is earned by the measured result.
- Work stage 410
Expansion
Earned
Reuse proven connections, definitions, permissions, workflows, trained roles and measurement patterns across additional workflows and rooftops.
- Work stage 511
Annual Review
Annual
Reassess readiness, remeasure results, update the Blueprint and resequence the next intelligence cycle.
The next intelligence cycle. Annual Review feeds the next Chassis and Blueprint planning cycle.
What each step is worth
- 01Operational hypothesis
- 02Evidence
- 03Dealer-owned engineering plan
- 04Verified facts
- 05Measured proof
- 06Compounding capability
Every stage ends in your decision—never an automatic continuation.
The Blueprint remains dealer-owned throughout.
Phase AUnderstand
Understand the dealership
Before anything is designed, we learn how the store actually runs.
Phase A is people and evidence, not software. Curated working sessions, an anonymous all-staff survey, and a structured look at how work moves between people and systems — held together long enough to find where workflows are not connecting and how the systems can be integrated.
- Who we talk to
- Ownership and leadership, department managers across sales, BDC, service, parts, F&I and marketing, plus IT and whoever actually maintains the integrations.
- What we look at
- How work actually moves between people and systems — where the same number arrives at two different answers, where a handoff is missing, and where a workflow that looks complete on paper is not complete on the floor.
- What everyone says
- An anonymous all-staff readiness survey. Leadership's read of the organization and the floor's read of it are rarely the same document.
- What it costs you
- Curated working sessions over six weeks, plus structured evidence gathering. No system changes, no installs, nothing switched off.
Phase A ends with the inputs that feed Blueprint V1: what we currently understand about how the store runs, where workflows are not connecting, and how the systems can and should be integrated. Nothing is verified yet; everything is documented so Phase B can design against it.
We also place the group on the seven levels of the Intelligence Pyramid — a diagnostic, not a score to defend. Most groups arrive somewhere in levels 2 to 3, which is exactly where additive AI stops paying. The seven levels in full.
You end Phase A with: the inputs that feed Blueprint V1 — what we understand, where workflows disconnect, and how systems can and should be integrated.
Phase BDesign
Design both halves
The intelligence infrastructure, layer by layer.
This is the thing your group ends up owning: connected data at the base, organizational memory and approved language above it, the places your people work, the agent teams coordinating underneath, and the workflows that reach the floor.
Read this as: bottom to top, one layer at a time
Phase B · what gets built
One stack, built once, in your name.
Every layer sits inside your control boundary and is read from the bottom up. Nothing above works without the layer beneath it — which is why adding another tool at the top has never held.
Dealer-owned control boundary
Read bottom to top. Each layer is only possible because the one below it exists.
Workflows
Layer 05Sales, BDC, service and leadership workflows, each ending at a named person who approves, edits or rejects.
On the floorWork arrives already reconciled, with the evidence attached and one decision to make.
Governed context moves up
Agent orchestration
Layer 04Small teams of narrow specialist agents, each routed to the model that suits its job, coordinating across departments.
On the floorNo single model guessing at everything — three to five specialists, each doing one job well.
Governed context moves up
Chat and workspace
Layer 03Where your people ask questions, draft work and review what the agents produced, grounded in your own operation.
On the floorA manager can ask about their own store and get an answer from the store's numbers, not the internet's.
Governed context moves up
Knowledge Hub and Brand Central
Layer 02Organizational memory — agreed definitions, operating context, workflow evidence — alongside shared identity, policy and approved language.
On the floorEveryone means the same thing by the same word, and nothing goes out off-brand or off-policy.
Governed context moves up
Intelligence Foundation
Layer 01A normalized, mapped, dealer-owned data environment connecting the DMS, CRM, phone, service, inventory, marketing and F&I systems you already run.
On the floorThe numbers stop disagreeing, because there is finally one place they are reconciled.
One stack, built once, in your name. Read bottom to top. Each layer is only possible because the one below it exists. 01. Intelligence Foundation — A normalized, mapped, dealer-owned data environment connecting the DMS, CRM, phone, service, inventory, marketing and F&I systems you already run. On the floor: The numbers stop disagreeing, because there is finally one place they are reconciled. 02. Knowledge Hub and Brand Central — Organizational memory — agreed definitions, operating context, workflow evidence — alongside shared identity, policy and approved language. On the floor: Everyone means the same thing by the same word, and nothing goes out off-brand or off-policy. 03. Chat and workspace — Where your people ask questions, draft work and review what the agents produced, grounded in your own operation. On the floor: A manager can ask about their own store and get an answer from the store's numbers, not the internet's. 04. Agent orchestration — Small teams of narrow specialist agents, each routed to the model that suits its job, coordinating across departments. On the floor: No single model guessing at everything — three to five specialists, each doing one job well. 05. Workflows — Sales, BDC, service and leadership workflows, each ending at a named person who approves, edits or rejects. On the floor: Work arrives already reconciled, with the evidence attached and one decision to make. We connect what you already run. The stack is built across your existing systems rather than replacing them, and it is not tied to one AI provider. If you leave, you leave with the data, the definitions and the learning in standard formats. Each rooftop keeps its own records. Every rooftop has its own boundary and its own Knowledge Hub. Only agreed definitions, approved measures and anonymized patterns pass the authorization gate into the group. Raw customer records never move up on their own.
Open, not enclosed
We connect what you already run.
The stack is built across your existing systems rather than replacing them, and it is not tied to one AI provider. If you leave, you leave with the data, the definitions and the learning in standard formats.
Across the group
Each rooftop keeps its own records.
Every rooftop has its own boundary and its own Knowledge Hub. Only agreed definitions, approved measures and anonymized patterns pass the authorization gate into the group. Raw customer records never move up on their own.
What this video says
- The day-to-day view: agent teams working underneath the workflows, with people reviewing and approving.
- Each rooftop's records stay inside its own boundary while the group still gets a coherent view.
- Nothing touching money moves without human approval.
The reviewer’s four questions
Where records live
Inside your rooftop's own boundary. Raw records are never pooled, copied, or sent outside it.
Who can reach what
By role and reason, not by reach. Access is granted for a specific job, then reviewed.
What a person signs
By consequence. AI drafts the recommendation; a named person approves the action.
How you leave
A standard-format export of your data and your trained outputs, whenever you decide.
What Phase B hands you
Blueprint V1
What the evidence currently supports, before technical verification. Every material finding carries an evidence label; the unknowns become the verification work order.
Evidence-backed, but not yet technically verified.
Blueprint V2
The same document in a verified state: corrections from V1 shown in both directions, plus the recommended architecture, workflow sequence, training plan, governance, measures, owners and unresolved risks.
A decision packet, not automatic authorization to build.
You end Phase B with: Blueprint V1 — the engineered infrastructure design your group owns, whether or not we continue.
How the work runs
Small teams of narrow specialists — not one big model.
Each agent does one job well and hands to the next. A person holds the decision at the end of the chain.
Read this as: left to right, ending at a person
How one workflow runs
From an operational signal to a measured result—with a person accountable at the decision.
A live workflow draws from dealer-owned context, coordinates specialist agents, respects organizational rules and returns measured learning to the foundation.
The illustrative signalA sales opportunity has passed its follow-up window with no appointment and no recorded outcome.
AI-ready organization
- Workflow owner
- Decision rights
- Role-specific training
- Approval rules
- Success measures
Organizational accountability applies at every step—not only at the approval.
- Machine work
Detect
Find the signal or exception.
- Machine work
Ground
Assemble the approved context required for the job.
- Machine work
Coordinate
Route defined jobs to the appropriate specialist agents.
- Machine work
Verify
Check evidence, policy, permissions and action rules.
- Human decision
Decide
A named person reviews, approves, changes or rejects.
- Approved return
Act
Return approved intelligence to the person or system responsible for the work.
- Measurement
Measure
Compare the result with the agreed baseline.
Dealer-owned intelligence foundation
- Per-rooftop data boundaries
- Governed operational data
- Knowledge Hub
- Agent orchestration
- Permissions
- Workflow memory and evidence trail
Each rooftop keeps its own data. Authorized intelligence can move; ownership does not.
Measured evidence returns to operating memory. This is governed operational learning—not automatic model training. Governed evidence, agreed definitions, workflow improvements and measured outcomes are captured for authorized reuse. The model does not train itself on your data.
AI drafts and coordinates. People decide. Results strengthen the foundation.
Illustrative process. Integrations, agents, permissions and measures are configured and verified per engagement, and anything touching money or a customer commitment stops at the configured human gate.
Legend
- Machine work — agents draft, coordinate and evidence
- Human decision gate — a named person approves, changes or rejects
- Approved return — governed action and measurement
- Return loop — measured evidence back to operating memory
HOW ONE WORKFLOW RUNS — From an operational signal to a measured result—with a person accountable at the decision. A live workflow draws from dealer-owned context, coordinates specialist agents, respects organizational rules and returns measured learning to the foundation. The illustrative signal: A sales opportunity has passed its follow-up window with no appointment and no recorded outcome. AI-ready organization (applies across every step): Workflow owner; Decision rights; Role-specific training; Approval rules; Success measures. Organizational accountability applies at every step—not only at the approval. 01 Detect — Machine work. Find the signal or exception. Example: The lead is past its follow-up window. 02 Ground — Machine work. Assemble the approved context required for the job. Example: CRM record, contact history, store definitions and permitted sources. 03 Coordinate — Machine work. Route defined jobs to the appropriate specialist agents. Example: Profile the opportunity, set priority and draft the next approach. 04 Verify — Machine work. Check evidence, policy, permissions and action rules. Example: Confirm what is known, what is inferred and what the store permits. 05 Decide — Human decision. A named person reviews, approves, changes or rejects. Example: The BDC manager remains accountable. 06 Act — Approved return. Return approved intelligence to the person or system responsible for the work. Example: The prioritized callback and approved approach return to the BDC workflow. 07 Measure — Measurement. Compare the result with the agreed baseline. Example: Response time, contact outcome and appointment result. Dealer-owned intelligence foundation: Per-rooftop data boundaries; Governed operational data; Knowledge Hub; Agent orchestration; Permissions; Workflow memory and evidence trail. Each rooftop keeps its own data. Authorized intelligence can move; ownership does not. Measured evidence returns to operating memory. This is governed operational learning—not automatic model training. Governed evidence, agreed definitions, workflow improvements and measured outcomes are captured for authorized reuse. The model does not train itself on your data. AI drafts and coordinates. People decide. Results strengthen the foundation. Illustrative process. Integrations, agents, permissions and measures are configured and verified per engagement, and anything touching money or a customer commitment stops at the configured human gate.
The specialists inside steps 03 and 04
A workflow brings a small team of agentic AI specialists together around one dealership job. A sales workflow should not behave like a service workflow, and a manager should not receive the same output as a frontline employee. Auto Agentic designs each workflow around the job to be done: the systems it must read, the specialist agents it needs, the permissions they carry, the people they support and the decision that remains human.
One example, run by agentic AI specialists published in the library. Reading the systems, interpreting what happened, drafting the action and checking it against your policy are different jobs, so they are different specialists sharing one governed context.
The CRM shows a lead past its follow-up window with no appointment and no logged outcome.
- 01
Dana
Leads Profiler
Reads the opportunity
Profiles the opportunity from the record it is given — how this buyer has behaved so far, and what that suggests about the way to re-open the conversation.
Hands the profile forward with the record attached.
- 02
Susan
Sales Manager
Sets the priority
Ranks the opportunity against the rest of the board and says whether it is worth working today, with the reasoning attached.
Hands the prioritized opportunity to the coach.
- 03
Declan
Sales Coach
Drafts the approach
Writes the specific approach for this buyer at this point in the deal, in the store's own voice — not a generic follow-up template.
Hands the draft to the policy check.
- 04
Magnus
Business Systems Analyst / SOP Expert
Checks the rules
Tests the draft against your store's follow-up SOP, contact permissions and what your group has agreed can be offered.
Hands it to the person who owns the call.
The BDC manager approves, rejects or redirects. Nothing touching money or a customer commitment leaves the store without that step.
Every step works from the same governed context and adds to the same evidence trail, so the manager can see what was read, what was assumed and what was measured before approving.
Each completed workflow adds more than one new capability. It establishes trusted connections, shared definitions, governed context, approval patterns and trained users that the next workflow can reuse. As those workflows connect, the dealership begins to see relationships across departments that no isolated tool could reveal. The operation gets smarter because its intelligence can travel; its people stay in control because every handoff and decision remains explicit.
An illustration of how a team is composed, using agents published in the library. Agent availability and system connections are confirmed per engagement.
You end this section with: a named person holding the decision at the end of every chain.
Phase CEngineer
Engineer the first workflows
Five workflows, shown in a working store.
Each one is shown in the store, not in a slide. Press play if you want it; the bullets carry the same content if you don't.
What this video says
- One morning briefing replaces the round of Monday reports each department writes separately.
- It reads across sales, service, BDC and inventory so leadership starts from one version of the day.
- What needs a decision is surfaced to the person accountable, with the underlying numbers attached.
What this video says
- Closing-rate differences between reps are traced back to what actually happened in the conversation.
- Coaching points are specific to the person and the deal, not a generic sales course.
- The manager keeps the call on what to coach and when.
What this video says
- Intent that arrives after hours is captured and put in front of the team the next morning.
- Response quality and follow-up gaps are measured per agent instead of per campaign.
- Nothing is sent to a customer without a named human approving it.
What this video says
- Recommended services that were never presented become visible by advisor and by day.
- Advisors get coaching tied to their own repair orders, not to a department average.
- Fixed operations leadership sees the pattern while it can still be corrected.
What this video says
- Shows the shape every workflow shares: gather, reconcile, interpret, recommend, hand to a person.
- Built for a gap the four standard workflows do not cover.
- Runs on the same foundation, so it inherits the connections already built.
What changes for your people
Systems and people move in the same week, or neither moves. The workflows above only hold because these four roles change with them.
Front line
Less re-keying and chasing between systems. More time on the customer in front of them.
Managers
Coaching from what actually happened, instead of from a report written after the fact.
Leadership
Decision rights written down and named, so accountability is clearer than it was before.
IT
One reviewed architecture to govern, rather than another vendor connection to babysit.
And what that looks like on a Monday
Monday · 7:40am
Across the store
The signal
The briefing has already reconciled the weekend across sales, service and BDC — one version of the numbers instead of six departmental reports.
The GM decides
Opens the day with a short list of decisions, and picks which two get attention before 10am.
Tuesday · a deal that stalled
Sales and BDC
The signal
The workflow surfaces what actually happened in the conversation — not just the CRM status — and the after-hours intent nobody answered.
The sales manager decides
Coaches that rep on that specific step, and decides what happens to the deal today.
Thursday · declined work
Service drive
The signal
Recommended work that was never presented shows up by advisor and by day, with the evidence attached.
The fixed-ops leader decides
Coaches the presentation, and owns the customer conversation that follows.
You end Phase C with: working workflows, and trained owners inside your stores who can run them.
Phase DProve
Prove it in the store
What you are actually building.
You have seen what it does. This is what it is — measured in a 90-day pilot against your own baseline, not a vendor benchmark.
What this video says
- The connective work is done once per rooftop; every workflow after that reuses it.
- Each deployment adds connections, patterns and trained people to the same foundation.
- That is why the second workflow is cheaper and faster than the first, and the fifth more so again.
You end Phase D with: a measured result against your own baseline — or a clean stop, with everything you built still yours.
Phase EExpand
Expand what has earned it
Why the fifth deployment is nothing like the first.
Only the workflows that proved out get scaled — and each one starts further ahead, because the connections, definitions, patterns, people and governance are already in place.
Why the next workflow starts further ahead
Reusable architecture reduces how much has to be rebuilt.
Shared intelligence foundation — reuse what already exists
- Connections
- Data mappings
- Permissions
- Agent patterns
- Trained people
The foundation is built once — then extended, governed and improved.
The connections
The integration work is done once per rooftop. Every workflow after it reuses the same connections.
The definitions
Once the group agrees what a term means, every later workflow inherits that agreement.
The patterns
What one department learns becomes context the next department's work can use.
The people
Staff trained on the first workflow are already fluent for the second.
The governance
Approval points, permissions and audit are established once, then extended.
You end Phase E with: each new workflow starting further ahead than the one before it, on infrastructure you already own.
Next step
Three sensible next steps.
Whichever you pick, the first conversation is about your situation — not a demo.
The four co-founders do this work themselves — you can read who they are in About Auto Agentic.
Talk it through with the team
Bring your own situation. We will tell you plainly whether this is the right time for your group.
Ask the advisor a hard question
Pricing, data ownership, jobs, integrations. It answers from our governed material, and routes anything commercial to a person.
Read the condensed site
The two halves, the problem AI runs into, the gated stages and where your data stays, in one scroll.
Not ready to talk yet?
Barry Hillier’s book sets out the industry argument behind this work — why intelligence has to be engineered rather than bought, and what that means for a dealership group.
Ask the advisor
Ask about the problems affecting your day — or where you want more control.
You've seen the stages. Ask what Chassis, Blueprint V1 and verification would mean in your stores — the systems and data underneath, what your people do differently, and what each stage has to prove. Then keep going; the advisor carries the conversation forward.