Case study
AI-drafted customer emails, with data kept in-house
An FCA-regulated debt collection firm, named here only by sector to protect the client's confidentiality.
100–200
customer emails drafted every day~95%
of drafts approved without changesOn-premise
no customer data sent to outside AI servicesThe problem
The firm receives a steady flow of customer emails every day. Each reply has to be accurate, polite and in line with FCA rules on treating customers fairly. Writing them all by hand took up a lot of skilled staff time.
Public AI tools weren't an option. Customer details are sensitive personal data, and the firm couldn't send them to an outside service.
What we built
- AI that runs on the firm's own servers. The language model runs on-premise, so customer data never leaves the building.
- A draft for every email. Each incoming email gets a suggested reply in the firm's tone.
- A person decides. Staff review every draft, then approve it, edit it or reject it. Nothing is sent automatically.
- Connected to existing systems. It works with the tools and processes the team already uses.
The result
The system now handles 100 to 200 emails a day. About 95% of drafts are approved without any changes, so staff spend their time on the cases that really need their judgement.
What this means for your business
The same approach works for any firm with sensitive documents, such as accountants, solicitors and regulated businesses. AI can do the first draft or the first read, a person checks it, and the data stays in the UK or on your own premises.
Talk it through on a free call