Case study · a financial planning and tax firm
From a client meeting to a scored client profile
A client intake system that fills in a questionnaire from what was said in a meeting, then ranks who the firm should contact next.
The problem
The firm wanted every client's situation recorded the same way, so its advisors could see who to contact next and why.
What I built
- Two ways in. A client answers a 41-question questionnaire on a web form. Or the advisor uploads the meeting transcript and the system fills in what the client said.
- It knows who said what. The transcript keeps its speaker labels, so a statement by the advisor is never recorded as the client's.
- The client confirms. The pre-filled form goes to the client as a draft. Anything that was not said in the meeting stays blank.
- Scoring is plain code. The score comes from fixed point tables. The same answers always give the same score, and every point can be traced to an answer.
- Advisors get a ranked list. A dashboard sorts clients by score, and summaries arrive by email daily, weekly and monthly.
- It reads the firm's CRM every night. The import only reads. It never writes back.
Questions in plain English
An advisor can type a question, such as a request for every client who matches a description. The AI turns the question into a filter and ordinary code runs it. Counts are exact, and no client record is sent to the AI to decide who matches.
One optional step does send records to the AI: a written summary of the clients who matched. It goes to an enterprise AI service only.
Where the data goes
Client data goes through Google's enterprise AI service and never through a consumer chat tool. The system stores no access keys that could leak.
Where it stands
The system runs in production on the firm's own web address.