Customer information
Help visitors navigate approved product, service or process information. Answers should stay within the available knowledge and offer an appropriate route when the requested detail is unavailable.
AI-augmented ChatbotsA useful answer.
A useful answer.
A sensible next step.
Knowledge-backed conversations for customer and internal enquiries, with defined actions and a clear handoff when a person is needed.
Our position
Design the conversation around the task.
We start with the questions and situations the chatbot is expected to support. The objective may be helping someone find accurate information, qualify an enquiry or complete a defined step. Conversation design, approved knowledge and action boundaries are developed together.
A chatbot should recognise the edge of its competence. We define what it may answer, which sources it uses and when it should ask a clarifying question or hand over. A confident-sounding response is not a substitute for a supported answer.
What we configure
A specific scope, built around your work.
Qualification and guided actions
Collect the information required for a defined next step. Where connected actions are in scope, specify the required fields, confirmation and destination rather than treating every conversation as permission to act.
Internal knowledge support
Help colleagues locate guidance and understand a process. Apply the appropriate access boundary and provide a route for correcting missing or contradictory information.
How we workPrepare the knowledge.
Prepare the knowledge.
Test the conversation.
The sequence is shaped around AI-augmented Chatbots, with scope and responsibilities agreed before implementation.
- 01User intent
- 02Approved knowledge
- 03Answer or clarify
- 04Action or handoff
Known → answerUncertain → clarifyOutside authority → hand off
Define supported intents
Map the questions and tasks that belong in the experience. Prioritise frequent, bounded needs. Identify topics requiring a person, requests outside scope and the minimum information needed for a handoff.
Prepare the knowledge
Select approved sources and assign owners. Structure information around the questions users actually ask. Decide how updates, conflicting guidance and unavailable answers should be handled.
Design the dialogue
Create clear prompts, clarification paths and useful fallback responses. Connect only the actions in scope, with validation and confirmation where required. A conversation should have a practical completion state.
Evaluate the edge cases
Test ambiguous requests, missing knowledge, incorrect assumptions and handoff scenarios. Inspect whether the conversation stays within its role and whether the receiving person gets the relevant context.
Explore the approach
Try three conversation boundaries.
These scripted examples demonstrate intended behaviour. They are not a live AI assistant and do not send information.
1Question2Find context3Answer4Clarify
Use approved knowledge
Visitor: “What information do you need to assess an automation?”
Assistant: “A process description, representative inputs, the systems involved and the intended outcome. Is there a particular workflow you want to examine?”
The implementation scopeWhat the work
What the work
leaves with you.
These are the delivery components we define together. The depth, access and handover arrangements follow the agreed engagement.
Conversation scope
Supported tasks, exclusions, tone and completion states.
Knowledge setup
Approved sources with an update and ownership process.
Chat experience
The agreed interaction, validation and connected actions.
Evaluation set
Test conversations and checks for fallback and escalation.
Capital & operation
Make repeat questions reusable.
A knowledge-backed chatbot can make a well-defined body of information more accessible without assigning a person to repeat every basic explanation. The cost case must include source maintenance, interaction usage and the work of handling escalations. We assess the proportion of enquiries that can be supported usefully within a bounded scope. Resolution quality matters more than simply counting automated replies.
Before we begin
Practical questions.
Will the chatbot make things up?
AI responses can be incorrect. We design around approved knowledge, clear limits, evaluation and fallback behaviour; we do not claim that any model is incapable of error.
Can it answer from our documents?
Where the sources and access model are suitable, yes. Source selection, permissions and updates need to be configured and tested as part of the experience.
Can it connect to the CRM?
A defined workflow can carry enquiry context into the CRM with the required fields and permissions. Qualification and message sending remain subject to the rules agreed for that workflow.
Define the first piece of workPrepare a useful
Prepare a useful
project brief.
Capture the objective, current process and first boundary. Download a working brief or copy it for a project discussion.
This tool runs in your browser. It does not submit or store your information.