Agents
Knowledge and documents
How the agent answers from your own documents, policies and FAQs, what happens when the answer is not there, and how your team decides which sources it uses.
On this page
- What the agent answers from
- Answers come from your sources, not from the model
- When the answer is not there
- Word-perfect answers from an approved library
- Required disclosures at defined points
- Keeping answers in step when documents change
- How your team approves sources
- The technical detail
- Where your documents can come from
- In practice
- What your team controls
- Related
Many of the questions your customers ask are already answered somewhere in your organisation: in a policy wording, a project fact sheet, a hotel's guest policies, a claims procedure or an FAQ page. The agent answers from those documents, and only from the ones your team approves. When the answer is not in them, it says so and offers a person rather than filling the gap. Your customers get the answer your organisation would give, worded consistently, every time.
What the agent answers from
The agent draws on two kinds of source. Your systems give it the facts about a particular customer: their booking, their unit, their policy. Your documents give it everything else your organisation has written down.
| Source | Examples | What it is used for |
|---|---|---|
| Your systems | Reservation system, property management system, policy administration system | Facts about this customer, looked up during the conversation. See How integrations work |
| Your documents | Policy wordings, project fact sheets, service charge schedules, fare conditions, guest policies, procedures | General answers about your products, projects and processes |
| Your FAQs | The questions and answers you already publish or use internally | Common questions, answered in your own words |
| Approved response library | Answers your legal or compliance team has written word for word, configured during implementation | Questions where the exact wording matters |
Answers come from your sources, not from the model
Answers about your business come from the sources you have approved. The language model (the AI component that interprets the conversation and composes the reply) is used to understand the question and to word the answer. It is not used as a source of facts about your products, prices, policies or processes.
When a customer asks a question, the agent finds the relevant passages in your approved sources and answers from them. The source behind each answer is logged, so anyone reviewing the conversation can see which document the agent relied on. See Audit trail and decision records.
When the answer is not there
When your approved sources do not answer a question, the agent tells the customer it does not have that information, and offers a person, or a call back from your team. It does not guess, and it does not fill the gap with something that sounds plausible.
The unanswered question is recorded with the conversation. Your team can search for these questions in the console (the Telonic web application your team uses) and see which ones come up most often, so you can decide whether to add the answer to your documents or keep routing those questions to a person. See Insights and trends.
Word-perfect answers from an approved library
Some answers must be given in exactly the words your legal or compliance team approved: how to make a complaint, a cooling-off period, a statement about what a product does not cover. For these, an approved response library is configured during implementation. The agent recognises the question and gives the approved answer word for word, rather than composing its own.
Your compliance team owns the library. A change to it goes through the same approval and testing as any other change. See How changes go live.
Required disclosures at defined points
Some statements must be made at a particular point in a conversation, whether or not the customer asks. A notice that a call is recorded, a statement an insurer must make before giving a quote, or an explanation of who is calling on an outbound call. Required disclosures are configured during implementation, and the agent makes each one at the point you define. Each disclosure is recorded in the conversation log, so you can show it was made.
Keeping answers in step when documents change
When a document changes in a connected source, the agent's knowledge is updated with it. Documents are re-indexed automatically when they change, so the agent answers from the current version rather than the one it saw at go-live. Nobody on your team has to retrain anything.
If a document is removed from an approved location, the agent stops answering from it. The simplest way to change what the agent says about a topic is to change the document your team already maintains.
The agent answers from what your documents say. If two approved documents disagree, or a document is out of date, the agent can only be as accurate as the source. Keeping the sources current is part of owning them.
How your team approves sources
Your team decides which sources the agent can use. During implementation, the owner of each area, such as the claims manager or the head of guest relations, agrees which libraries, folders, sites or spaces hold approved material for their area. Only documents in those locations are used.
Before go-live, those owners review the agent's answers to the questions their area receives most often, and correct the source where an answer is wrong. Adding a new source afterwards is a change like any other, tested before it goes live.
The technical detail
Documents are indexed: split into passages and stored so the agent can find the passages relevant to each question by keyword and by meaning. The index is held in your deployment, in the hosting option and region you chose. Documents and images are read by a model running in your chosen region. See Hosting and data residency.
Each source is connected with credentials your IT team grants, limited to the locations you have approved. Answering from documents needs read access only. See Permissions.
Where your documents can come from
The agent connects to the document and knowledge systems enterprises already use, including Microsoft SharePoint, Google Drive, Notion, Confluence, Dropbox and Box. See Knowledge and documents integrations.
In practice
Qamar Stays, a hotel group in Saudi Arabia, connects its guest policies and FAQs, held in SharePoint, to its Telonic agent.
- The head of guest relations approves two SharePoint libraries: guest policies and the hotel FAQ.
- Hassan messages the Jeddah hotel on WhatsApp to ask whether he can bring his cat. The agent answers from the guest policy for that hotel, and the source is logged.
- He then asks whether he needs a visa to visit from Egypt. Visa requirements are not in the approved sources. The agent says it does not have reliable information on that, and offers to have guest relations contact him.
- At the weekly review, the guest relations team sees that visa questions come up often. They decide to keep sending them to a person, and add a short approved answer pointing guests to the government's visa website.
- The FAQ is updated in SharePoint. The agent answers from the new version from then on.
What your team controls
- Which libraries, folders, sites and spaces the agent answers from.
- The approved response library, and which questions use it.
- The disclosures the agent must make, and where in the conversation.
- What the agent does when the answer is not in your sources: offer a person, a call back, or both.
Related
- Knowledge and documents integrationsIntegrations
- How an agent worksAgents
- Policies as limitsAgents
- Insights and trendsQuality and analytics
- How changes go liveGovernance and control
Product names and logos are trademarks of their owners. Their mention shows systems Telonic connects to and does not imply partnership or endorsement.