Domain 3 of 6: Be asked
Converse
Agents can ask you questions and get fast, accurate answers, with a clear path to a person when an answer needs one.
The outcome and why it matters
Between reading you and acting with you comes asking. Agents ask whether a product is in stock, whether a clinic accepts a plan, how a refund works. If you can answer in seconds and accurately, you stay in the conversation. If you cannot, the agent asks someone else.
An analogy
Led by: Engineering and product, Service and operations. See the responsibility matrix.
Capabilities
5 things an organization does in this domain. Standards named here are explained in the Standards Map.
Question and answer endpoints
A way for an agent to ask a question and get an answer directly, instead of searching pages for it.
- What good looks like
- Common questions are answered by a defined endpoint that draws only on approved sources and cites them. It returns no answer rather than a wrong one.
- Typical standards and techniques
- Model Context Protocol (MCP) servers exposing tools and resources
- NLWeb (an open project for natural-language interfaces to websites)
- Plain FAQ pages marked up with schema.org FAQPage
Live facts for live questions
Answers that depend on the moment: stock, availability, opening status, wait times, order status.
- What good looks like
- Live data is read from the system of record at the time of the question, not from a cached page, and says when it was last updated.
- Typical standards and techniques
- REST or GraphQL APIs described with OpenAPI
- MCP tools backed by live data
- Webhooks and event feeds for changes
Agent-to-agent conversation
Your own agent or assistant talks directly with agents that represent other people or organizations.
- What good looks like
- The conversation has a defined scope, a declared identity on both sides, and a log. The agent says plainly that it is an agent.
- Typical standards and techniques
- Agent2Agent (A2A) protocol with Agent Cards
- MCP for tool access
- Identity and signing from the Trust domain
Handoff to people
A clear path from an agent’s question to a person, for answers that need judgment, empathy or authority.
- What good looks like
- The path is published, tested and staffed. The agent is told when a person will answer and how long it takes, and no question is dropped.
- Typical standards and techniques
- Published contact and escalation routes
- Ticketing or case-management integration
- Service-level targets for agent-raised requests
Honest answers
Rules for what you say when you do not know, when the answer is a judgment, or when the answer is a no.
- What good looks like
- Answers state their source and date. They separate fact from opinion, and they refuse politely when a question is out of scope.
- Typical standards and techniques
- An answer policy approved by marketing, service and legal
- Source citation in every answer
- Regular accuracy sampling (see Measure)
See the Standards Map for who owns each standard and how mature it is. To learn the skills, see the Specialist certification (Engineering and Marketing).
KPIs
Defined precisely, so two organizations count the same way.
- Answer rate
- Agent questions that received a substantive answer, divided by all agent questions received. A refusal to answer an out-of-scope question counts as answered; an error or timeout does not.
- Time to answer
- The median and 95th percentile time from an agent’s question to the first complete answer, split between automated answers and answers that needed a person.
- Handoff completion
- The share of questions handed to a person that were answered within the published time.
What each maturity level looks like here
- 1
Level 0: Closed
No way to ask you anything except a contact form no agent can use.
- 2
Level 1: Visible
Agents can find answers on pages but cannot ask. Contact details are public.
- 3
Level 2: Described
FAQs are structured and current. Common questions have a documented answer.
- 4
Level 3: Conversational
A question endpoint answers from approved sources, with live data and a route to a person.
- 5
Level 4: Callable
Agents hold multi-turn conversations with you, in scope and with logs. Handoffs are timed and staffed.
- 6
Level 5: Trusted
Answers are verified as accurate by sampling, and failed questions are reviewed weekly and turned into improvements.
See all domains side by side or take the self-assessment.