AI Assisted Intake Form
At Intapp, I led the design of an AI-assisted intake form for Intake, our business intake product used across professional services firms. Intake forms are something people have to fill out, and the information they need is often already in an email they've received. Instead of retyping it, they drop that email in and the system fills in the form. They review it and stay responsible for what's submitted.
An intake request is a long form, around 200 fields, and it becomes the matter record. In professional and legal services, that record is a compliance artifact, so it has to be right. People have to fill it out, and much of what it asks for is often already in an email they've received, so they were copying it over field by field.
The brief was simple: let AI fill out the form. But talking to users, I realized automation alone wasn't the answer. They were anxious about it. Who extracted this? Is it correct? What did the AI change?
A form that fills itself creates a new job: checking. If checking takes as long as typing, nothing is saved.
How might we turn an email into intake data people can verify at a glance?
- The worry was about where things came from. If people can't tell what the AI touched, they re-check everything.
- Ambiguity is information. If an email names two possible lawyers, silently picking one is the worst option.
- The person signs it. Whatever the AI suggests, the submitter is accountable, and the design should make that feel natural.
I tested two directions:
- Inline suggestions: the AI's suggestion sits beside each field, right where people type.
- Summary panel: a panel at the top lists everything the AI filled, before people touch the form.
Testing settled it in favor of inline.
Here's what testing showed about each:
Suggestions appear right where people expect to type. They felt in control, even though the system did most of the work, and that made them far more willing to trust what they saw.
The AI's work is spread across a 200-field form, so it's harder to see everything it filled at once.
All of the AI's work in one place. On paper, it looked like the more transparent option.
People skimmed it and still questioned what had changed. It added a step without adding confidence.
Showing the AI's work without cluttering the form. Going inline meant the AI's suggestions live on the form itself. But a 200-field form is already a lot. Badging every AI field would make it worse, and hiding them brings the anxiety back. So I split the job:
- Reviewing: a highlight toggle tints every AI-filled field and adds a "(pre-filled)" label, so it doesn't rely on color alone.
- Working: toggle it off and the form returns to its normal, tighter layout.
- Unsure: the system shows "2 suggestions available" instead of guessing. The AI narrows the choice, and the person makes it.
Drag an email into the form. The system reads it, pulls out what's relevant, and pre-fills the fields. Same workflow, just faster. Every suggestion can be edited in place.
AI-assisted should never mean AI-controlled. The system does the first draft. The person does the thinking.
Email ingestion removes roughly 10% of the ~200 fields a user would otherwise fill by hand.
An email only holds so much, and the rest still needs the person.
In a form like this, speed comes from trust, and trust comes from being able to tell who did what. Making that line visible was the design.