An installed base of a few hundred thousand appliances produces signals no service organisation can read by hand: warranties expiring, repeat calls clustering on one failure code, one partner's evidence quality slipping. The operating doctrine: deterministic signals, AI delivery, human approval — engines watch every unit, agents deliver what matters, people decide.
What runs by itself
- AMC offers queue as warranties expire — generated per unit on your schedule, with the unit's service history attached for pitch and price.
- Installation jobs generate from retail sales — routed to the right partner within the promised window, no coordinator required.
- Partner claims assemble from job cards — swept on schedule with evidence attached.
- Nudges publish overnight — repeat-call clusters, ageing serials in dealer stock, cases drifting toward SLA breach, partners whose claim quality dipped — in the right queue by morning.
What agents draft — and ask before doing
Sense answers the service head in plain language: "why are repeat calls up on the new washer range?" — and can trace it to a failure code, two partners and a substitute part, because job cards, parts lines and serial history share one model. It drafts recall lists, partner coaching nudges and win-back offers; every write waits for approval. Each manager's day opens with a deterministic daily brief, phrased by the model, numbers computed before the model ever saw them.
Vision and voice where they earn their keep
Sense Vision measures dealer display compliance from photos on the beat. AI Studio lets the brand build its own agents — grounded in product manuals and service bulletins, budgeted, audited, permission-bounded — so the diagnostic assistant that helps a technician at the doorstep is your agent on your knowledge. Governance: how Sense is governed.
Common questions
What after-sales work can run unattended?
The deterministic engines: install-job generation from sales, AMC offer queues from expiring warranties, claim assembly from partner job cards, replenishment suggestions for parts, and scheduled nudge sweeps — output always landing in a human queue.
How would an agent catch a quality problem early?
By reading the same records the network writes: repeat calls clustering on a failure code and a part batch surface as a pattern with the affected serials listed — weeks before the quarterly review would have noticed.
Do agents contact customers directly?
Agents draft outreach — AMC offers, recall notices, follow-ups — and route them for approval under your messaging governance. Nothing reaches a customer that a human didn't release.
Can we build a diagnostic assistant on our own manuals?
Yes — AI Studio agents ground in your uploaded knowledge (manuals, bulletins, fault trees) and answer within it, assistively — supporting the technician, not replacing the qualified procedure.