Agents
Memory and context
What the agent remembers within a conversation, across conversations and across channels, how it uses only what matters, and how your retention settings shape what is kept.
On this page
The most common frustration in customer service is having to explain everything again. A customer calls about a claim, sends photographs on WhatsApp two days later, and is treated as a stranger. The agent remembers what each customer has told you, on every channel, and brings the parts that matter into each new conversation. How much it remembers, and for how long, is decided by your retention settings.
What the agent remembers
Memory works at several levels, from the conversation in progress to a customer's history over years.
| Level | What the agent remembers | What it means for the customer |
|---|---|---|
| Within a conversation | Everything said so far, including corrections | They do not repeat themselves mid-call |
| Across conversations | Previous conversations with the same customer | A returning customer is recognised, and their history is known |
| Across channels | Calls, WhatsApp messages, SMS, emails and web chats as one history | A conversation started on the phone continues on WhatsApp |
| Conversations your team handles | What a colleague discussed with the customer, configured during implementation | The history is complete, not just the part the agent handled |
| Long-term profile | Preferences, history and open commitments, built up over months and years | The next conversation starts from what is already known |
Within a conversation
The agent holds everything said so far in the conversation. When a customer changes their mind or corrects a detail, such as a date or a room type, the correction overrides the earlier detail. When they add a condition late in the conversation, the agent takes it into account without starting again.
Across conversations and channels
When a customer gets in touch again, on any channel, the agent recognises them and draws on their history. Every channel writes to, and reads from, the same record of each customer. There is one memory with several ways in, rather than a separate history for each channel.
Recognising a customer is not the same as trusting that they are who they say they are. Contacts are matched to customers on verified identifiers: phone number, email address, customer ID, and booking or policy reference. Before sharing account details, the agent confirms the customer's identity using the checks you set. See Matching contacts to customers.
Conversations your team handles
The history is most useful when it is complete. Conversations your own team holds, on calls, WhatsApp or email, can be captured into the same record, so when the agent picks up the next conversation it knows what your colleague agreed. This is configured during implementation for the channels and systems your team uses. See Conversations your team handles.
The long-term profile
Over months and years, the agent builds a profile of each customer from what they have said and done: their preferred language, the channel they tend to use, their history with you, and anything still open. Relevant details are extracted from each conversation and kept in the profile, for as long as your retention settings allow.
A customer's preferred language is remembered, so their next conversation starts in it without asking. Commitments made to them are held until they are delivered. See Commitments and promise tracking.
Only what matters to this conversation
A long relationship produces a long history. The agent does not re-read all of it at the start of every conversation. It finds the parts that matter to the conversation in front of it: an open claim, a recent change, a promise not yet kept. That keeps the agent informed without slowing the conversation, and keeps its attention on what the customer is asking now.
The technical detail
Conversation state and each customer's long-term memory are held independently of any channel, so a voice call and a message written two days later resolve to the same context. They are stored in your deployment, in the hosting option and region you chose. See Hosting and data residency.
Relevant history is found by searching the customer's record by keyword and by meaning, rather than by loading the whole history into each conversation.
Your CRM (the system that holds your customer and account records) stays the system of record (the authoritative source) for customer and account data. The Telonic record is the system of record for conversation history. Access to the record is controlled by role, and every access is logged. See Access control and single sign-on.
How long memory lasts
Nothing is stored without your permission, and you set how long anything is kept. You can set different retention periods for different kinds of content, in line with your data policy and the regulations that apply to you. When a period ends, that content is deleted, and the agent no longer draws on it.
If your policy does not allow conversation content to be kept, the deployment can run without storing it, or store it only in storage you control. This is configured during implementation. With content storage off, the agent works from what is said during each conversation and from what your own systems hold. A content-free audit record of what the agent did (its actions, the sources it used, the triggers that applied and the outcomes) is still kept, for the period you set.
| Your setting | What the agent can draw on | What is kept |
|---|---|---|
| Content stored in your deployment | Everything on this page, within your retention periods | Conversations, summaries and the profile, plus the audit record, each for the period you set |
| Content held only in storage you control | The same, read from your storage | Content in your storage. The audit record in your deployment, for the period you set |
| Content storage off | The conversation in progress, and what your own systems hold | A content-free audit record only, for the period you set |
See Storage, retention and deletion.
Memory stays inside your deployment. What Telonic carries from one deployment to the next is industry knowledge, such as workflows and edge cases. It never carries your customers' data.
In practice
Wadi Assurance uses a Telonic agent for motor claims on voice, WhatsApp and email.
- On Monday, Hamza calls to report a rear collision. The agent confirms his identity, takes the first notice of loss (the first report of a claim), and gives him a claim reference.
- On Wednesday, he sends four photographs on WhatsApp from the same number. The agent recognises him, attaches the photographs to the claim, and confirms they have arrived.
- On Friday, Layla in the claims team calls him and agrees a hire car until the repair is done. Her call is captured into his record.
- The following Tuesday, Hamza messages: "Any news on my car?" The agent draws in the open claim, the estimate and Layla's agreement, and replies in Arabic, his preferred language. It tells him the claim is with the assessor and that his hire car continues until the repair.
- Nine months later, at renewal, the agent knows the claim history and answers his question about the effect on his no-claims discount from his renewal terms.
What your team controls
- How long each kind of content is kept, and whether content is stored at all.
- Which conversations your own team holds are captured into the record.
- The identity checks a customer passes before the agent uses their history.
- Who in your organisation can see each part of a customer's history.
Related
- The customer recordCustomer record
- Matching contacts to customersCustomer record
- Conversations your team handlesCustomer record
- Storage, retention and deletionSecurity and data protection
- Handover to your teamAgents
Product names and logos are trademarks of their owners. Their mention shows systems Telonic connects to and does not imply partnership or endorsement.