Heep Customer Profile: The AI Memory That Brings Your Regulars Back 2.3× Faster
How Heep’s AI memorizes preferences, allergies and history for each guest — across every channel — turning one dinner into a 2.3× faster return rate.
The question we get about AI memory is almost always the wrong one. It isn't "does your AI remember guests", it's "what does it re-attach them to".
A profile isn't built from a name: a name repeats, gets spelled three ways and changes channel along the way. What holds is a stable identifier, a phone number, an email address, an Instagram account. As soon as one reappears, months later, the conversation re-attaches to the existing thread and the profile comes back with it.
What follows describes what it keeps, what it infers rather than records, and the three things it refuses to store.
2.3× return frequency for regulars recognized by Heep AI, vs. unidentified guests
Internal sample — 40 partner restaurants, Q1 2026
What the AI sees when a guest types "Evening, it's Marc"
Marc sends an Instagram DM at 7:04 pm. Before any human has glanced at it, the recognition has happened: the account matches a thread opened in February, whose history loads with the conversation.
For your front-of-house team, the gain is invisible, and that's exactly the point. One thread per guest, across every channel, and the profile is already there when the AI agent replies.
- 87% of returning guests are automatically recognized, regardless of channel (Heep · internal data Q1 2026)
- The history is up before the reply is written, not after the service
- A higher average spend when the table, the wine and the dish are proposed from preferences the system already knows
One precision: these are observations on our own base, not market averages, and we do not publish a percentage for them. The recognition rate depends above all on the channel of origin: a withheld number is attached to nothing.
Want to see a real Heep guest profile? See the Heep guest profile in a 2-min demo
The 7 attributes Heep remembers — and the 3 it deliberately forgets
A useful memory is a bounded one. Too much history kills trust, too little makes the AI useless. The exact scope fits in ten lines.
What Heep remembers
- Name & phone number — cross-channel unification key
- Preferred language — detected on the first interaction
- Allergies & dietary requirements — declared or inferred from context
- Favourite dishes — aggregated from order history
- Preferred table — deduced from past bookings
- Meaningful occasions — birthday, wedding anniversary
- Cancellation pattern — to adapt the reminder cadence
What Heep deliberately forgets
- Full payment details — never stored on the Heep side
- Private conversations outside restaurant context — purged at each session
- Data on accompanying minors — excluded from the profile
Two of those seven lines are inferences, not declarations: the preferred table and the favourite dishes are computed from history, so they are wrong while it is short. The limitation corrects itself after a few visits.
Case study
"Ms L. books for Tuesday via Instagram. She said nothing about her gluten allergy that evening. Heep retrieved the information from a WhatsApp exchange four months earlier and suggested three safe dishes — before she even asked."
Evening service — 80-cover brasserie, Lyon
How the memory personalises every outgoing message: tone, language, dishes, timing
Memorising without acting is just storage. Memory is only worth keeping if it settles something at the moment the reply gets written, in four places.
Memorised language detection
A Japanese guest who has booked twice in Japanese gets a reply in Japanese, even if they write in English this time. The preferred language on file overrides on-the-spot detection, unless the guest switches and sticks to it.
Tone calibration
Weekly regular? Warm and direct. First interaction? Formal and clear. The tone is set from the relationship history, not from a generic heuristic.
Direct preference suggestion
A guest who consistently orders the tasting menu gets it as the opening line: "The tasting menu is available tonight, shall I hold it for you?"
Adjusted reminder cadence
A guest who frequently cancels the day before gets a confirmation 2 days out rather than 12 hours. The cancellation memory becomes a table-recovery tool, not a judgement call.
On channels where the guest writes the least, such as customer recognition on WhatsApp Business, personalisation tips 1 in 3 bookings from a plain "OK" into a consented upsell.
Real case: an 80-cover brasserie in Lyon, from 18% to 41% recognised regulars in 6 months
An 80-cover brasserie, lunch and dinner, a mix of tourists and neighbourhood regulars. Before Heep: notes in a notebook, passed between shifts when remembered. After: a unified profile fed by the conversations themselves.
- 18% → 41% of regulars automatically recognized (month 1 vs month 6)
- +£9 average spend on guests identified as regulars
- 2.3× more return visits for a recognized regular vs an anonymous guest
- Zero team friction — no form to fill in, no tool to open
The turning point was the Ms L. episode. Gluten-intolerant, she had written in January, then booked through Instagram in May: the allergy came back with the thread, the chef prepared three gluten-free dishes, and she returned three times that quarter.
That doubling is not automatic: it tracks the share of requests arriving on an identifying channel. Two settings decide the rest, and both sit on your side. How long preferences are kept, set to 24 months by default and adjustable per venue. And the direction of the sync with your booking software: shared attributes (name, phone, allergies, cover count) are aligned on both sides, the conversational memory stays with Heep. The two things to check before leaning on it during service.
Frequently asked questions
Is Heep GDPR-compliant when storing customer data?
Yes. Never resold, processed under a lawful contract basis, and a Data Processing Agreement (DPA) is signed with every restaurant partner. Full payment details and conversations outside restaurant context are never stored. Where the data is hosted and which sub-processors are used is published on our security page.
What happens if a guest requests profile deletion?
On a simple request (to the restaurant or directly to Heep), the profile is purged within 30 days. Accounting data linked to an issued invoice is retained for the statutory period but dissociated from the conversational profile.
How long are preferences kept without a new visit?
24 months by default, configurable per restaurant. After that period with no interaction, the profile is anonymised: aggregate statistics are retained but the identity is removed.
Can my floor staff consult the profile before the guest arrives?
Yes. A “pre-service” view lists the evening’s bookings with key details: allergies, occasion, favourite dishes, preferred table. Accessible on tablet or computer, with access logging.
Does the Heep customer profile export to my existing CRM (Zenchef, Sevenrooms)?
On Zenchef and SevenRooms, the booking and its notes are written into your system with what the service needs (name, phone, allergies, cover count). Heep-specific attributes (conversational memory, preferred tone) stay on the Heep side. Do not assume a full record copied both ways: have the exact scope confirmed on your contract before building a process on it.
