Resatisfy AI
Give every customer a concierge.
Your advisors answer the same six questions a hundred times a day. The concierge answers them instead — accurately, instantly, and grounded in what your business actually says.
- Knows the customer, the location, the service, the queue position and the live estimate
- Answers policy questions from your own knowledge base entries
- Refuses to invent a price, a policy or an availability it cannot verify
- Hands off to a person and notifies the location when it does not know
Premier Auto Group
Fairfax BMW
You're checked in
Thanks Daniel. We've got you.
Est. wait
26m
Ahead
3
Checked in
In queue
In service
Complete
Ask us anything
Can I leave and come back?
Absolutely, Daniel. Tap “Leave & notify me” and I'll text you when you're about 10 minutes from being up. Your place in line is held.
Capabilities
What you actually get.
- Visit-aware
- Every answer is computed against the live visit: who checked in, what for, how long ago, how many are ahead and which advisor owns it.
- Grounded in your knowledge base
- Loaner policy, what to bring, pricing rules, hours, amenities, warranty coverage. If it is not in there, the concierge does not assert it.
- Honest escalation
- When there is no grounded answer, it says it will get a person to confirm and raises an in-app alert. That behaviour is enforced in code, not just in a prompt.
- Staff take-over
- Any employee can step into the thread. The customer sees a name, not a bot, and the AI stops replying on that conversation.
- Suggested replies
- Staff can draft with the concierge from inside the inbox — same context, same grounding, human sends it.
- Works without a model
- With no API key configured, a deterministic engine answers from live visit state and the knowledge base so the product is fully testable offline.
How it works
End to end, in order.
4 stages · one record per visit
- 01
Retrieve
The question is matched against your published knowledge entries, scoped to the customer's location.
- 02
Ground
Live visit context and the matching entries are assembled into the only material the answer may draw on.
- 03
Answer
Two to four sentences, specific, no filler, using the customer's first name once.
- 04
Escalate
If nothing grounded the answer, the concierge says so, alerts the location, and the exchange lands in the staff inbox.
Everything the concierge says is auditable.
Every answer is stored on the conversation with the intent it matched, whether it deflected, and which knowledge entries it used. Managers can see exactly what customers are being told.
Every wait is inside Fairfax BMW's thresholds
Target 20 min · warning 25 min · critical 45 min, set on this location.
Alex Duarte has been waiting 12 minutes with no status update
Sentiment is critical. Manager intervention and an immediate service update recommended.
| Customer | Service | Wait | Staff | Status |
|---|---|---|---|---|
| John Reyes | Express oil change | 8m | Derek Hollis | Waiting |
| Sarah Whitfield | Walk-in service | 27m | — | Waiting |
| Alex Duarte | Tire service | 12m | — | Waiting |
| Maria Delgado | Service appointment | 38m | Chris Boone | Called |
| Ingrid Lindqvist | Service appointment | 24m | Monica Reyes | In service |
Alex Duarte · wait against thresholds
12m
Questions
The things buyers actually ask.
- Which model does it use?
- Anthropic's Claude when ANTHROPIC_API_KEY is configured. Without a key, a deterministic engine composes answers from live visit state and the knowledge base — every workflow still works.
- Can it hallucinate a policy?
- The retrieval and deflection logic sit outside the model. If no knowledge entry matches, the deflection path runs regardless of what a model would have said.
- Does it ever change a booking?
- No. It can pass a reschedule request to staff, but it takes no action on the customer's record. Every state change goes through a person or an automation you configured.
- What languages does it support?
- The model path follows the customer's language. The deterministic demo engine is English only.