Product
Lawli.ai — AI legal assistant and document drafting platform
LAWLI.AI · LEGALTECH AI PLATFORM · PRODUCT REDESIGN (1.0 → 2.0)
Lawli.ai 2.0: From Six Features to Two Intents
Redesigning an AI legal platform around two connected tasks: legal research and document drafting.
Project status: 1.0 was shipped. This case study documents the 2.0 design; production rollout of the redesign has not been independently verified.
- 2.0 design for an existing LegalTech AI platform
- Six destinations reduced to two intents
- A mandatory gate removed from the highest-intent moment
- Shipped 1.0 screens and 2.0 design compared
The 30-Second Version
62
Sidebar modules → intents
Lawli.ai is an AI legal assistant for lawyers, paralegals, law students and in-house counsel. Version 1.0 shipped as a feature dashboard: six sidebar modules, a card-grid home, and an assistant that made users pick a legal domain from a dropdown before they could ask anything. I redesigned the product around the two dominant intents identified in the workflow analysis — understand something, or produce a document.
1.0 was organised around what the product could do. 2.0 is organised around what the user came to do.
Product Objective
Four business goals set the terms of the redesign, and each one is answered by the two-intent architecture rather than by styling.
- 01Reduce time-to-first-value
- 02Make the AI the product surface, not a module
- 03Connect research and drafting into one workflow
- 04Create a structure that absorbs new capability without growing the navigation
Project Overview
A full redesign of a shipped LegalTech product, carried out as design owner across both versions: from the heuristic review of 1.0, through the two-intent architecture and the chat-to-draft bridge, to the screens, states and recovery paths that 2.0 needed to work as a product rather than a demo.
Project
Product redesign (1.0 → 2.0) · 2025 · 2.0 design documented
Role
Senior UI/UX Designer — design owner across both versions
Users
Lawyers, paralegals, law students, legal consultants, in-house counsel
Tools
Figma (design, prototyping, QA) · Figma AI · Claude
Business goals
Reduce time-to-first-value · make the AI the product surface, not a module · connect research and drafting into one workflow · create a structure that absorbs new capability without growing the navigation
I owned
UX architecture, the two-intent model, the chat-to-draft bridge, all screens, states, onboarding, auth and error design
I did not own
AI behaviour, drafting engine and template generation (engineering) · scope and release sequencing (founders, with my input)
What Changed Structurally
AreaLawli 1.0Lawli 2.0
Navigation
Six feature destinations in a sidebar
Two intents in one top-level switch
Landing
Dashboard-first: a card grid of eight competing destinations
Work-first: the user lands in the mode they came for
First question
Domain gate: select a legal domain before typing
Immediate input, with suggested prompts as guidance
Drafting entry
Template-first: browse a filtered gallery
Prompt-first: describe it, with templates still below
Workflow
Research and drafting as separate destinations
Intent bridge: an offered handoff between the two
Coverage
Happy paths only
Onboarding, auth recovery, error and empty states
The Product Before the Redesign
A heuristic review of the shipped 1.0 screens surfaced four recurring problems:
01
Navigation mapped to features, not goals.
Six sidebar modules required users to translate their goal into the product’s vocabulary before doing anything.
02
A gate before the first question.
The assistant required selecting a legal domain from a dropdown before the user could type — a taxonomy decision users may not be able to make confidently, placed at the product’s highest-intent moment.
03
A dashboard that delayed the work.
Eight competing destinations, including a client list and a stock-photo article carousel, with no dominant action.
04
Browse-first drafting.
The Drafter opened on a template gallery, so users who already knew what they needed had to find it in a catalogue.
Discovery
Five inputs informed the redesign:
Heuristic review of the shipped 1.0
screen-by-screen analysis of navigation, hierarchy and first-session path, which produced the four findings above
Founder and stakeholder discussions
where 1.0 underperformed commercially and what the product needed to become
Role and workflow mapping across the five user types
a law student researches, a paralegal drafts repeatedly, in-house counsel does both in one sitting
Competitive pattern audit of general assistants and LegalTech tools
how they structure conversation, templates and editing
Engineering constraint sessions
on drafting output, template structure and editor capability
Reframing the Problem
The brief was "modernise the UI." The review pointed somewhere harder: the interface was not dated, it was mis-structured. A visual refresh would have left every navigation decision exactly where it was.
Across the workflows I mapped for the five roles, the majority of tasks clustered around two intents: understand something, or produce a document. Six modules were six answers to a question users were not asking.
Constraints
Product
- A shipped product with existing users — recognisable, not a different application
- 1.0 capabilities could be reorganised or absorbed, but not silently deleted
Domain
- Legal work is high-stakes: generated drafts are starting points requiring human review, and the interface must not imply finished legal advice
- Five roles with materially different workflows share one surface
- Legal documents have real structure — parties, clauses, fillable terms — so the editor could not be a plain text box
Technical and scope
- Drafting output depends on the model and template library, so the UI had to leave room for editing rather than present output as final
- Export had to survive formatting, which shaped the editor toolbar and PDF export
- A small team and phased delivery — the core loop first, secondary states specified for later build
Comparing Three Directions
Visual refresh of 1.0
What it wasKeep six modules and the dashboard; restyle components and typography
Outcome and whyRejected — leaves the gate, the card-grid home and feature-based navigation untouched.
Single unified assistant
What it wasCollapse everything into one chat surface; drafting happens in the conversation
Outcome and whyRejected — legal documents need a structured editor with fillable terms and export. A transcript cannot carry that.
Two intents, one bridge
What it wasChat and Draft as peer modes, with an explicit handoff between them
Outcome and whySelected — matches the workflow patterns identified across the five roles, keeps the editor intact, and gives new capability a home without a new destination.
The Redesigned Experience
01
One switch, two modes. Chat with Lawli and Draft Document sit as a centred toggle at the top. The six-item sidebar is gone; the sidebar now holds the user’s own material — chat history grouped by Today, Last 7 Days and Previous Month, or recent documents and templates in drafting mode.
02
Immediate input, guided start. The domain dropdown is removed. Six suggested prompts — three research, three drafting — show what the product is good at without blocking anyone who already knows.
03
Prompt-first drafting. Drafting opens on a prompt field with examples, then templates and pinned documents below. 1.0 served only the browsing user.
04
A real document editor. Drafts open in a rich-text editor with legal structure intact — headings, parties, fillable blanks, inline comment markers and Export as PDF — so a draft is a starting point the user finishes.
05
Skippable onboarding. Three steps opening with the user’s role, skippable at each. Personalisation without recreating the gate the redesign removed.
The Intent Bridge
The most distinctive interaction in the redesign has no before/after, because 1.0 had nothing like it.
In 1.0, a user researching a question and then needing a document had to leave the assistant, find the Drafter, and start again — the product treated one task as two destinations. In 2.0, when the conversation indicates a drafting task, the interface offers an inline handoff card into Draft Document. The move is offered, never forced: the user stays in control of the mode switch, and declining it costs nothing.
- 1. Research: Ask a legal question in the conversation.
- 2. Choose the handoff: Accept the offered Draft Document card, or continue the conversation.
- 3. Edit the draft: Review and complete the document in the structured editor before export.
Research and drafting are one workflow, not two destinations.
States and Edge Cases
1.0 covered the happy path. The redesign extended the product into the states a working product needs:
First session
three-step onboarding, skippable at every step.
Account recovery
Sign Up, Login, Forgot Password and Reset Password as a complete set.
Server failure
a designed 500 state with a route back to work, instead of a dead end.
Reading vs working
collapsed-sidebar workspace and document preview.
Response feedback
thumbs up, thumbs down and copy, so users can signal quality on a system that will sometimes be wrong.
Accessibility requirements
These requirements are part of the design specification. Production accessibility conformance has not been independently verified:
- Visible focus states across the mode switch, composer, sidebar and editor toolbar
- Contrast targets for orange and its tints on light surfaces, including small text and icon-only controls
- Active mode and selection never signalled by colour alone
- Long chat titles and document names truncate without hiding their actions
- Editor toolbar operable by keyboard
The Business Case for Two Intents
These design decisions aim to make the product easier to extend and explain. Their effects on delivery cost, adoption and onboarding have not been measured.
- 01
New capabilities can reuse the two entry points.
In 1.0 every new capability needed a sidebar slot, a dashboard card and a place in the mental model. In 2.0 the seventh capability can enter an existing intent — a new template category, prompt type or editor action — without requiring another navigation destination.
- 02
Shared components limit new design and QA work.
Two modes share one composer, one suggested-prompt pattern, one history sidebar and one document surface. The design gives the team reusable patterns for new capabilities, with the aim of limiting additional screens and states.
- 03
A stable entry model as the feature set grows.
Research and drafting remain the two primary starting points as capabilities expand. New features can be introduced within those modes; whether this reduces learning effort still needs evaluation.
- 04
The product becomes describable in one sentence.
Ask a legal question, or draft a legal document. That is a clearer proposition for acquisition and sales than a list of six modules.
Impact & Evidence
Design delivered
- A complete 2.0 product design: two-intent architecture, chat and drafting modes, the intent bridge, document editor, onboarding, full auth set and error states
- Six destinations reorganised into two intents with no capability silently dropped
- The mandatory domain gate removed from the product’s highest-intent moment
- A state model covering onboarding, recovery, failure and empty states that 1.0 did not have
Not measured
Hypotheses the architecture is designed to produce, not results:
- Removing the gate increases first-session completion of a first question
- Prompt-first drafting reduces time-to-first-draft versus the template gallery
- The intent bridge increases the share of research sessions that produce a document
What I would measure next
GroupMetricQuestion it answers
Time to value
First login to first question asked, and to first draft produced
Did removing the gate speed people up?
First session
Share of new users who ask or draft in session one
Does the new entry point convert intent into action?
Intent bridge
Handoff appearance rate, click rate, drafts completed after it
Is the bridge noticed, wanted and useful?
Mode balance
Chat vs drafting split, and switches within a session
Was two modes the right structure?
Draft quality
Edit volume before export, regenerate or abandon rate
Are drafts usable starting points?
Onboarding
Completion vs skip by role, and whether role changes behaviour
Does the one question we ask earn its place?
Reflection
The two-intent model was committed without user validation. Next time I would test it with a handful of users from each role before building the architecture on it, and secure analytics access so the redesign could be measured rather than argued.
A redesign earns its name by changing what the product is organised around, not what it looks like.
