How ShiftPilot AI assistance works
ShiftPilot AI's assistance is a controlled layer, not an autonomous agent. It reads the dealership's own records to draft, summarize, prioritize, and surface opportunities — and anything customer-facing or otherwise sensitive is proposed to a person and executed only after approval.
Propose, review, approve
Every suggestion arrives as a proposal attached to the record it came from. A user with the right role reads it, edits it if needed, then approves or dismisses it. Approvals and dismissals are recorded, so there is a trail showing who decided what.
High-risk work — sending customer messages, submitting finance data, changing pricing, running exports — never executes on the AI's own initiative.
Grounded in dealership data
Suggestions are generated from the store's own leads, customers, vehicles, deals, service records, and communication history, scoped by the same role and tenant rules that apply to the person viewing them. If a suggestion is based on data that has since changed, the interface marks it as stale rather than letting a manager act on an old picture.
By department
- Sales and BDC
- Drafts follow-up messages against the actual customer record, ranks who to call next, and flags leads going cold. A person reviews and sends.
- Service
- Summarizes inspection findings and open estimates, and highlights declined work worth revisiting. Advisors decide what reaches the customer.
- Finance readiness
- Checks a deal's documents and structure against a readiness checklist and explains what is missing. It does not pull credit or obtain lender decisions.
- Inventory
- Surfaces aging and turn patterns and drafts merchandising language. Market valuation feeds require a provider agreement.
- Managers
- Summarizes the day, explains where a number came from, and puts proposed actions in one approval queue.
- Multi-rooftop groups
- Applies the same signals and the same approval rules per rooftop, so stores are comparable without merging their data.
Limits worth stating
Model output can be wrong. Numbers presented as estimates stay labelled as estimates until a lender or the dealership confirms them. AI does not create integrations: if a provider is not connected and approved, nothing is sent anywhere.
Model calls run server-side, and dealerships remain the owners of their data. See the trust center for the control summary, or the product overview for the surrounding workflows.