Service 01/03
AI Agents
Support and sales agents that answer, qualify and book — and know when to stop and hand over.
An agent is software that reads a message, looks things up in your systems and replies — or passes the conversation to a person with everything they need. We build agents for one channel and one job first, measure them against your real conversations, and only then widen the scope.
What you get
Support agent on web chat, email or WhatsApp
Lead qualification, routing and booking
Voice agent for appointments and call-backs
Human hand-off with the full conversation
Eval set of 100+ real cases and a weekly accuracy report
Runbook, prompts and logs in your accounts
An example workflow
Hover or focus a step to see what it does. Every workflow is adapted to your tools.
AI Agents · example run
- A customer writes in. The agent sees the message and who sent it — nothing else until it needs it.
- The model works out what's being asked and reads only the records it needs: the order, the account, the relevant policy page.
- Plain rules decide: answering and tracking are fine; refunds, cancellations and anything unusual go to a person.
- The person sees the conversation, the proposed reply and the reason it was escalated. One click to approve, edit or take over.
- The reply goes out, the ticket is tagged and the run is logged for the weekly accuracy review.
How the project runs
Week 1
Audit: pull 300 past conversations, sort them by intent, agree what the agent may do alone.
Weeks 2–3
Prototype on your data, tested against 100+ real cases; weekly demo on Friday.
Week 4
Live in one channel with hand-off, logging and approval steps on.
Weeks 5–8
Daily review of hand-offs and failures; the eval set grows with every miss.
Case studies
Related work
Illustrative results from fictional clients. Results vary.
Questions
Questions about this service
Will customers know it’s an AI?
Yes. The agent introduces itself as an AI assistant and offers a person at any point. It’s the honest default and, in the EU, a legal duty for chatbots.
What does it do when it doesn’t know?
It says so and hands over — with the conversation, what it looked up and why it stopped. We’d rather it hand over too often in week one than guess.
Can it work in more than one language?
Yes, if your knowledge sources exist in those languages. We test each language against its own eval set before switching it on.
Ready when you are
Your first agent.
Four weeks.
Start with a free 30-minute automation audit. You leave with a shortlist, a cost estimate and no obligation.