Customer service and support
The same questions tie up your best people. The rest is exactly the part no system should take over.
What people search for: automate customer service, take load off support
The situation
Eighty percent of enquiries are five questions that come back every week.
The answer sits in an email from two years ago.
Anyone new in service asks a colleague on every second case.
What we build
/ 01
The recurring cases first
We count which questions actually arrive, then build. Not the other way round. Five well-answered cases take more load off than fifty half-answered ones.
/ 02
Answers from your material
The basis is your documents, not a language model's world knowledge. Every answer carries its source so service can check rather than trust.
/ 03
A handover that works
When it hands over, the whole case goes with it: what was asked, what was answered, what is open. A customer who has to tell the story twice is angrier than before.
What we do not automate
The most honest part of any project. On the left, what a system can take over. On the right, what stays with a person.
| The system handles it | A person keeps it | |
|---|---|---|
| Recurring questions | Answer from approved material | Answer when the source gives nothing |
| Complaints | Record fully, document, pass on | Promise goodwill or a refund |
| Contractual matters | Point to the version in force | Interpret deadlines and claims itself |
| Disclosure | Appear recognisably as an AI assistant | Carry a name that suggests a person |
Common questions
- Why does a chatbot rank so low with you?
- Because it is rarely the biggest lever and almost always the first thing requested. A chatbot is visible, so it gets asked for. Measured in hours saved, the inbox, quoting and document processing usually sit well ahead of it. If service is your bottleneck, the order is of course different.
- Do we have to disclose that it is not a person?
- Yes, as soon as it is not obvious that someone is talking to a system. The EU AI Act sets that out as a transparency duty in Article 50. In practice: a recognisable name, a notice on first contact, and a visible route to a human.
- What if the assistant says something wrong?
- There are two safeguards: it only answers from approved material, and every answer carries its source. If it finds nothing, it hands over. On top of that you see which questions went unanswered, and that list tells you what belongs in the material next.
Where to start
Hand off the routine
For companies whose teams do the same motions every day.