The next conversation should start with context

The customer has already explained

An appliance repair has been arranged. During the booking call, the customer explained that they work mornings, that the building entrance is on a side street, and that the technician should call when they arrive. Later, the customer opens WhatsApp with a short question: “Can we make it later?”

The question is easy to understand when the earlier conversation is available. It refers to the repair appointment, the customer needs a later time, and a morning visit is unlikely to work. Without that context, the next agent has to reconstruct a story the business has already heard.

This fictional service booking is the example throughout this article. It illustrates a shared-memory workflow; it is not a customer deployment or a claim that a particular scheduling system is already connected.

The opportunity is practical. Preserve the information that helps the next agent ask a useful question, check the right record, and continue toward the customer's requested outcome. The customer should be able to change channels without becoming the person responsible for joining the business's conversations together.

Give each interaction a place in the same relationship

Unified Memory is the business's customer context shared across ConnectX agents. Calls and messages contribute to that context, and another agent can build on it when the customer returns. In healthcare, the same principle applies to the patient profile used across appointment agents, Medical Copilot, and Medical Chat.

That creates continuity at the level of the relationship. A booking call contributes the request and the interaction. A later WhatsApp exchange contributes a new preference or a change. The next agent has a starting point that reflects the work already done.

The distinction between stored information and useful context is also present in Sierra's Context Engine perspective, which describes bringing business records and observations from interactions together at the point of a decision. Our focus here is the customer experience that a shared record should support: the next agent understands what has happened and what still needs to happen.

For the repair booking, the next useful question is about the later time the customer wants. It should not require another explanation of the appliance problem, the building entrance, and the morning work schedule just to reach that question.

Remember the preference. Check the current situation.

A remembered preference and a confirmed appointment serve different purposes. “Afternoons work better” helps an agent choose relevant options. “Thursday at 16:30” describes a specific booking, with a current status that can change.

In the example, WhatsApp picks up the repair request and the afternoon preference. Before promising a later visit, the agent needs to check the appointment and the available options in the scheduling system. The earlier conversation can explain the request; it cannot make a time slot available.

This is an important evaluation boundary. A convincing recollection is only part of a good response. The agent also has to distinguish what was requested, what was proposed, and what was actually completed.

Suppose the customer accepts a later option, but the scheduling update fails. The conversation now contains an agreement, while the appointment system still contains the earlier time. A useful response explains that the change has not been confirmed and follows the agreed exception process. Saying “all done” would leave the customer with a promise the operational record does not support.

The same test should be repeated when a customer changes an earlier preference. “Afternoons are usually better” should help the agent. It should not prevent the customer from choosing a morning this time. Useful memory makes room for new information.

Carry the unresolved work across the channel

An interaction can end while the customer's task is still open. The customer may need to check with someone else. A supervisor may need to authorize a change. A connected system may not complete the requested action immediately.

For a cross-channel evaluation, include one of those interruptions. End the first call with the requested change still unresolved, then continue on WhatsApp. Inspect whether the next agent can identify the pending decision and explain the next step without presenting it as completed work.

If a human decision is needed, Soft Forwarding brings the supervisor into a separate conversation with the relevant brief. The supervisor can guide the agent or ask to take over. Shared customer context helps the brief stay focused on the actual problem and what has already been attempted.

After a resolution, the next interaction should reflect the outcome. In our repair example, that means the confirmed time and any still-relevant arrival instructions. The person receiving the technician should not have to discover whether the phone agent and the WhatsApp agent were working from different versions of the appointment.

Shared context also matters inside the clinic

A patient's relationship with a clinic spans several conversations. An appointment call establishes the booking and practical preferences. WhatsApp can carry scheduling updates. In the doctor's account, Medical Copilot brings the patient profile and relevant context into the live consultation, alongside the speaker-labelled transcript and developing clinical sections.

Medical Chat provides a different part of that workflow. The doctor can discuss the patient's context and work with MRI images, radiology reports, and lab results there. Those clinical capabilities belong to Medical Chat in the doctor's account; they are not capabilities we attribute to the calling agent.

The shared profile connects the journey, while the work of each agent remains specific. A scheduling conversation needs the context to arrange the next appointment. A clinical conversation needs the information the doctor can inspect and use. Discuss the information available to each role and the data flow as part of the workflow design.

The value to the doctor is continuity: the consultation begins with existing context in view, and today's conversation can contribute to the record the care team uses next. Follow that part of the journey in the consultation-to-reviewed-note example.

Customer memory and learned solutions have different jobs

Unified Memory answers a question about this customer: what has the business learned from their interactions, and what context matters now?

The Knowledge Loop answers a different question: what useful approach can an agent learn from the way a human resolved a problem? ConnectX scores learned solutions, and high-scoring solutions become usable automatically in future calls. Call Evaluation examines the conversation and its resolution, including human assistance.

In our service example, the customer's side-street entrance belongs with their context. A supervisor's method for resolving a scheduling exception is a candidate for scored learning. Keeping those roles clear helps teams evaluate whether an agent is remembering an individual circumstance or applying an approach to a new problem.

Read the handoff and learning story for a complete example of the supervisor's contribution and what can become useful afterward.

Test continuity through a change

A useful memory demonstration should include a change, not just a successful recollection. Start with an ordinary booking call, move to WhatsApp, change one preference, and inspect the next action. Then introduce an unresolved update and see what the following agent says.

This gives a buyer something concrete to assess. The agent remembered the relevant context, accepted a correction, checked the current situation, and carried the task forward. Each observation is more useful than a generic claim that a system “has memory.”

Explore Unified Memory in ConnectX AI Voice Agents with a familiar customer journey. Bring the points where people repeat themselves today and the moments where an earlier conversation would help the next person act. Those are the places where shared context can make the experience feel like one continuous relationship.