Using AI to Build an AI Chat Case Study

The Business Problem

Someone looking for help usually wants to know one thing: has wildesoft.net solved a problem like mine before? Finding out meant browsing dozens of pages across many clients and systems. At the same time, businesses are asking us how to use AI safely, and the best way to show them is with something real. An AI chat on a public website comes with real risks, though: it might invent clients or figures, be misused, or run up costs.

The Solution

We wrote a detailed design first, covering what the chat may and may not say, then used AI to help us build it in small, tested steps. The chat answers in plain English, drawing only on the published case studies, and every claim about past work links to the case study behind it. If nothing matches, it says so honestly, and it never quotes prices or timescales.

It's protected at every level. Visitors are limited to a set number of questions a day, off-topic requests are politely declined, personal details are removed before questions are logged, and a fixed monthly spending cap can never be exceeded. A set of test questions with known answers checks that the chat stays accurate after every change. The library of case studies rebuilds itself when they change, and old conversations are deleted automatically.

When a visitor is ready to talk, they can pass the whole conversation to us in one step, so they never have to repeat themselves. An "Ask about this" button on every case study starts the chat from that page.

The Result

Visitors get straight answers about our experience, backed by the real case studies. We learn what people are asking for, including the questions our work doesn't yet answer. The chat also shows, on our own site, the careful, well-guarded use of AI we offer clients through our AI Consultancy.

See all case studies for Website