AI receptionist memory: how cross-channel context works
Agentric Team ·
The most underrated feature of a unified assistant: it doesn't start every conversation from zero. Here's how memory across phone and chat actually works — and where its edges are.
The identity layer
Memory requires knowing who you're talking to. Two identifiers do the work:
- Phone calls identify themselves — the caller's number arrives with the call. Caller 07700 900 123 is the same contact every time they ring.
- Web chat starts anonymous — visitors are just a browser session until they share a phone number or email, at which point the assistant links the conversation to their contact record.
Once a chat visitor shares a number that matches a past caller, the threads join: the person who rang about a boiler quote on Tuesday is recognised when they message on Thursday.
What gets remembered
Each conversation ends with a summary: what they wanted, what was booked, what's unresolved. On the next interaction — either channel — recent summaries are injected into the assistant's context, alongside the contact's stored details.
So the Thursday chat opens with the assistant already knowing: Tuesday's call was about a combi boiler quote, a survey got booked for Friday, the caller's name is Sarah.
What this means in practice
- No re-explaining. "Following up on the quote" just works — context is there.
- Smarter conversations. The assistant can reference the booked survey naturally.
- Continuous intake. Phone call → chat → phone call is one relationship, not three strangers.
The honest limits
- Anonymous chatters stay anonymous until they identify themselves — memory needs a key.
- Memory is summarised, not a total transcript replay — the last handful of conversation summaries, plus core contact details.
- Two people sharing a phone number share an identity — rare, but it happens.
Why it matters commercially
Continuity is what makes it feel like a receptionist rather than a robot. "Oh — hi again, the survey's still booked for Friday" is the moment a customer stops evaluating the AI and starts valuing the service. For repeat-business trades, that recognition is loyalty infrastructure.