Carry Useful SMS Context into the Next Voice Call

Voice and SMS context continuity can save a caller from repeating a recent request. It is most useful when the assistant remembers a small, relevant piece of history and still lets the caller correct it.

Burki's current continuity is bounded. When enabled, later calls and SMS can restore up to twenty committed turns from the previous twenty-four hours for the same tenant, assistant and customer phone. It is not an unlimited customer memory or a guarantee that every channel shares the same conversation.

Decide what recent context should do

Use restored context to orient the conversation: the caller recently asked about a service, received a follow-up or corrected a detail. Do not treat it as permission to repeat an external action automatically.

A previous request may no longer be current. The caller could have solved the problem, changed their availability or be calling about something else. A short acknowledgement followed by an open question is often more useful than a long recap.

The existing voice and SMS guide provides the broader idea. The actual enabled settings, retention rules and current implementation limits determine what information is available in a particular call.

Know which message history qualifies

Call-origin SMS needs a delivered receipt before it enters the shared context. Ordinary AI SMS replies can be retained after provider acceptance, which is not the same as delivery proof. Keep that distinction when the assistant refers to a previous message.

Messaging providers expose different status stages; Twilio's official outbound message-status guide is a useful reference for why acceptance and delivery should not be treated as identical.

Even a delivered message does not prove the customer read or understood it. The assistant should not say “as you saw in the text” unless the caller has established that they saw it.

A hypothetical return call

Imagine a caller who recently asked for a business's opening hours and received a delivered text from a call-related action. They call again with a question about visiting on a particular day.

The assistant might use the recent context to avoid asking why the hours were sent, but it should still answer the new question from current approved information. If the business changed its hours, the old message should not override the current source.

This is a hypothetical example. It illustrates how continuity can support a conversation without treating old content as permanent truth or proof that a visit was arranged.

Do not expect live cross-channel updates

Incoming texts do not update an already active call in the current implementation. A caller who sends a correction while speaking should be asked to state the correction in the conversation or follow the business's supported process.

WhatsApp text history remains separate. Do not describe the feature as a fully synchronized omnichannel memory. A shared Messages interface is not evidence that every channel feeds the same active assistant context.

The timing boundary matters for support staff too. A later call may have access to eligible committed history that was not available during an earlier live session.

Review continuity with identity and privacy in mind

Phone-based context can be helpful, but a phone number is not a universal identity guarantee. Shared family or office phones require care. Avoid volunteering sensitive history merely because a number matches.

Current storage and redaction settings apply. Turning continuity on does not mean every old missing projection is reconstructed, nor does it erase the need to review retention and access expectations.

The AI receptionist guide can help decide where a short context-aware opening belongs in the caller experience. Keep the greeting proportionate to the task and let the caller redirect it.

Before adopting continuity, review one recent delivered call-origin text, one message whose delivery is unknown and one caller correction. Confirm what the next conversation can actually use. The useful outcome is less repetition with accurate boundaries, not an assistant that claims to remember everything.