AI legal compliance

How I integrated generative AI into Intapp Time, saving lawyers hours and firms millions by streamlining time entry narrative compliance

225,000 users | $150 b/yr in billing | 6 weeks to design

Lawyers and other professionals billing hourly track their time by creating time entries with a narrative detailing their work. Narratives must comply with firm rules and client contracts—and bills can have +400,000.

Enter Intapp Time’s AI-powered narrative rewrite. With one click, narratives become compliant with firm rules and client expectations.

Overview

Domains

Industries: Legal, consulting

Technology: AI (LLMs)

Time frame

6 weeks

My role

Senior designer, working with a lead designer, 2 product managers, and 26 engineers

UX design for end user & admin experience

UX research

Goals

For the business: Demonstrate Intapp Time’s superiority over competitors by saving firms millions in contested bills and non-compliant narratives

For the user: Quickly write compliant narratives without having to check firm rules, client rules, and grammar rules

Problems

There were 4 design problems to solve for, 3 for the users and 1 for my team:

End users (attorneys, paralegals) need to quickly draft and release time entries without having to learn complex UI

Law firms need to maintain consistent branding and prevent revenue loss from non-compliant or client-contested entries

Admin users need a way to configure what rules are followed—at firm, client, matter, and jurisdictional levels

Our team needed to ensure the AI was as accurate as possible, without blowing through token usage

Higher stakes: The team needed a quick turnaround to keep our biggest client—who’d recently started a pilot program with our biggest competitor because of their AI features

Research & Requirements

Personas

There were two types of users to account for: admin users and end users (“timekeepers”). Timekeepers are anyone tracking time, either their own or on behalf of someone else.

Timekeepers write narratives for their time entries to bill clients. Timekeepers include attorneys, paralegals, and legal assistants.

A timekeeper should be able to:

  • Know as quickly as possible if their narrative is breaking a rule

  • Quickly fix a non-compliant narrative

  • Be able to edit or revert to a previous version of their narrative

Admin users determine what rules narratives should follow. Rules can be firm-wide or client- and matter-specific.

An admin should be able to:

  • Find what rule they’re looking for

  • Immediately see what rules are enabled

  • Configure these aspects of the rule:

    • Whether it’s enabled

    • Its enforcement level

Design Iterations

Admin experience

The admin experience went through various iterations before settling on its list layout. For the product MVP, the team decided the admin experience would be for firm-level rules only.

V1: Basic data grid

Lists out all firm rules

Shows enablement status

Does not allow configuration

Long horizontal text = difficult to scan

V2: Expandable cards

Lists out all firm rules

Shows enablement status

Allows configuration

Easy to scan

Configuration experience not scalable

Too much information for one page

V3: List view with overview

Lists out all firm rules

Shows enablement status

Allows configuration

Easy to scan

Configuration scalable (user clicks into new page)

Tolerable information density

Didn’t test well with users (was hard to scan)

V4: List view with testing (the winner!)

Lists out all firm rules

Shows enablement status

Allows configuration

Easy to scan

Configuration scalable (user clicks into new page)

Tolerable information density

Tested well with users

Added widget to test configurations (after feedback)

The Edit a rule page was added when I realized that rule configuration could be very complex.

Here, the user can add good and bad examples of narratives to train the AI on what to look for.

The timekeeper end user needs a way of knowing what rules or guidelines their narrative violates. I used existing design patterns around tool tips, warnings, and loading states.

Timekeeper end user experience

User testing was done ad hoc in 30-minute feedback sessions with users, product management, and client success.

Final design

Admin experience

Admins can:

  • Immediately see all their firm rules

  • Configure a rule’s

    • Enablement

    • Enforcement level

  • Add example narratives to train the AI

From user feedback, we added:

  • Test narratives with their configured rules

  • Add example narratives and additional instructions

Timekeeper experience

Timekeepers can:

  • Know as quickly as possible if their narrative is breaking a rule

  • Quickly fix a non-compliant narrative

  • Be able to edit or revert to a previous version of their narrative

From user feedback, we added:

  • Timekeepers can see a tool tip of the exact rules their narrative violates

Impact

Better time entry governance

Admins can set rules that maintain consistency and compliance across all narratives.

Less revenue loss

Fewer non-compliant entries means fewer contested bills from firm clients.

Improved AI trust

Using AI transparently and effectively demonstrated its value to apprehensive users

More competitive advantage

Intapp’s new AI features prove its mettle against less robust competitors.

In the press

“The Intapp Time Horizon release is very intuitive and won’t require us to retrain our lawyers… Our users really like the quick add functionality [and] the ability to use AI to create narratives.”

— Brian Donato, CIO at Vorys

“In professional and legal services, every billable minute contributes to revenue generation, and that’s why you need a reliable timekeeping platform like Intapp Time. It integrates with key systems and enforces compliance from the moment of entry.”

— Megha Agarwal, in a review for NeosAlpha

Lessons learned

Every project teaches me something new about design. Here are some takeaways from working on Intapp’s AI compliance:

  • Nail terminology first, then architecture, then everything else. There was a lot of back-and-forth about the admin experience. What would we call it? Where in the setting architecture would it live? I should have aligned the team sooner on those basics before diving into the design.

  • New technologies exacerbate old habits. Product management was ambitious about finishing this feature to compete with a startup. But as the team discovered AI’s limitations, we had to pull back on some goals—for example, AI hallucinates often enough that we added the Additional Instructions + Examples sections to keep prompts as specific as possible.

  • AI can’t solve fundamental design flaws. For the timekeeper experience, the real problem people had was knowing what rules their narratives violated. I added a tool tip listing the rules last minute, but I wish I’d had the time to rethink the structure of the entry form as a whole.

Next project

Watson Orchestrate

How conversation design and user research saved watsonx before it was watsonx

IBM | 0-1 in B2B SaaS