Client Memory

FrontdeQ extracts and stores client-specific details from every call. On the next call from the same client, those memories are available to the AI — so returning callers are greeted as people, not strangers.

What Gets Captured

After every call, FrontdeQ analyzes the transcript and extracts memories into seven categories:

Each memory has an importance rating (high / medium / low) and a source — which call it came from — so you can verify or delete it later.

Where the Memories Show Up

When a known customer calls again, FrontdeQ loads their memories into that call's context. The AI uses them naturally:

Agent:   Hi Maria! Great to hear from you. How's the wedding
         planning going? Would you like to book another session
         with Jessica?

That's not a script. Memory — the wedding, the stylist — woven into a natural greeting. The AI won't recite the memories verbatim ("I see you're getting married in October"); it works them in where appropriate.

Memories are also visible in the customer's record in your dashboard, with source call references.

Browsing Memories

From your dashboard:

1. Memories — top-level nav
2. Search by customer name or phone number
3. Each customer's memories are grouped by category
4. Click a memory to see the source call and transcript excerpt

You can edit a memory (correct a typo, refine the wording) or delete individual memories.

Deleting Memories

Delete on-demand from the Memories page. Two options:

• Delete a single memory — removes it from future calls. Source call record is preserved.
• Delete all memories for a customer — clears the entire memory set. The AI will treat this customer as if they've never called before on the next inbound call.

For CCPA / GDPR compliance: deleting memories is permanent. There's no undo. If you delete a memory and the customer calls again, the AI will not have that memory available unless it's re-captured from a new call.

What FrontdeQ Does Not Capture

We deliberately exclude:

• Medical information. If a caller mentions a health condition, it's not persisted as a memory. Exceptions: allergies specifically related to salon services (e.g., "allergic to PPD in hair dye") are captured because they affect service delivery. Other health details are noted in the transcript only.
• Credit card details or payment info. Never captured, never stored.
• Personal data about third parties. If the caller mentions a friend or family member's details, we don't create a memory about that third party.
• Negative subjective judgments. "This client is rude" or "always late" would not be captured as a memory. The AI reads tone from the live call, not from persisted judgments.

How Memory Accuracy Is Maintained

The extraction model is Claude Haiku, prompted to capture only explicit statements the caller made. It won't invent memories. If a memory turns out to be wrong (the caller misspoke, the extraction missed context), the memory can be edited or deleted manually.

The QA system periodically reviews memories against recent calls. If a memory contradicts what the caller says on a new call (e.g., "I prefer Maria" from 2 months ago, but the caller says "let me try someone different this time"), the AI prioritizes the live statement over the memory.

Privacy & Compliance

• Encryption at rest. Memories are stored encrypted in the same database as your other customer data.
• No cross-salon sharing. Memories are scoped per-organization. Another salon using FrontdeQ cannot see your memories.
• Right to deletion. If a client asks you to forget them, use the "delete all memories" button on their record. We log the deletion for audit.
• Right to access. The dashboard shows everything captured. A client can request a copy of their memories; export from their customer record.

What Happens on Memory Extraction Failure

If the post-call memory extraction fails (the extraction is a post-call background job; failures don't affect live calls), the call still gets logged. No memory is captured for that call. The next time the same customer calls, the AI uses whatever memories exist from prior successful extractions.

Failures are logged in the dashboard's operational views. They're rare — typically caused by transient API errors that resolve on retry.

Building Memory Quickly

The memory system gets more useful as customers call more often. A brand-new client has zero memories. By their third or fourth call, the AI knows their preferred stylist, their scheduling patterns, any specific service notes they've mentioned.

For high-volume salons, expect meaningful memory coverage on returning clients within 2-3 weeks of launch.