LexHero

A lawyer starts drafting a contract in one platform. Then comes the familiar tour: the document is downloaded, opened in Word, sent by email, checked against another file, copied into a legal research tool and eventually uploaded somewhere else for signature.

LexHero already offered useful tools for this work. Users could analyse a document, generate content or search for information. What they often could not understand was where to begin, which tool to use or how one part of the product related to another.

The problem was not a lack of features. It was orientation. The interface had become a map of the company’s feature history rather than a reflection of the customer’s work.

The opportunity was not to add another stop. It was to reduce the number of stops.

I used user interviews, sales calls, product demonstrations and support requests to understand how independent lawyers, law-firm employees and in-house legal teams actually worked.

A legal professional reads a document, notices a problem, checks a source, rewrites a clause, compares two versions, adds client information and reviews the result. A new finding may change an earlier paragraph. Editing and analysis continuously affect one another.

The workflow is a document that gradually becomes more complete. The product architecture needed to reflect that.

One possible strategy was to separate LexHero’s AI capabilities into standalone products: a document analyser, a research assistant, a contract generator and other specialised tools. Each would have a clear name and a tidy landing page.

But the user would still be left assembling the workflow—moving information between environments, repeatedly explaining the same context and deciding which tool to open next. The product would look more organised while the work remained fragmented.

Adding AI this way would not solve the problem. It would automate the fragmentation. Instead, we organised the product around the object legal professionals were already working on: the document.

The redesigned experience was built around two connected elements. Total Chat offered one conversational interface for questions, analysis, legal research, generation and changes to the active document. Canvas provided a persistent place to preview, edit, structure and personalise the result.

Chat provided speed. Canvas provided control. Together, they let users move from an initial request to a usable legal document without switching between standalone AI tools.

A user could upload a contract, identify a risky clause, inspect the relevant source, request an alternative and edit the final wording directly. The context stayed in place. The user stayed in place too.

Chat is effective for direct questions and initial drafts. It is much less effective when the result is forty paragraphs long, contains several sections and needs repeated revision.

Legal documents need a full structural view. Users must be able to edit one clause without regenerating everything, personalise parties and obligations, compare new content with original material and understand what changed.

Canvas removed the gap between generation and evaluation. AI could propose a change through Total Chat while the user inspected and edited the document directly. The result was immediately visible, structured and usable.

The workspace was organised into three areas: document navigation for uploaded files and supporting material; Canvas for the active document and direct editing; and Total Chat for analysis, questions, generation and modifications.

This relationship meant the interface no longer asked users to understand LexHero’s internal organisation. It followed the structure of their work.

• Document navigation
• Canvas
• Total Chat

Trust is not an optional layer in legal technology. An AI response can sound polished and precise while still being incomplete or wrong. Fluency is useful. It is not evidence.

Legal professionals need to understand where information came from, what the system changed and which parts still require their judgement. AI-assisted changes therefore needed to remain inspectable and reversible.

• Source references
• Verification cards
• Document comparison
• Clear content ownership
• Change history

The existing Figma files had no shared component system. Similar elements used different layouts, behaviours and visual treatments depending on where they appeared.

I created a common foundation for typography, colour, navigation, forms, document states, chat components, verification elements, cards, panels, feedback, loading states and reusable layouts.

The design system established shared rules for how LexHero behaved. Future capabilities could enter through existing patterns instead of becoming entirely new sections with their own interaction language.

Products rarely become complicated through one dramatic mistake. They become complicated one reasonable feature request at a time.

The most important result was not a new visual style. It was a new product model: analysis, research, writing, editing and verification became connected parts of the same document workflow.

There were no reliable baseline analytics that would support a precise productivity claim. The impact was evaluated through observable product changes: reduced architectural fragmentation, a clearer primary workflow, a consistent interaction system and a stronger foundation for AI-assisted legal work.

• One primary workspace replaced disconnected entry points
• AI capabilities shared the same document context
• Generation and manual editing worked without export
• Trust and verification became part of the interaction model
• The system created a scalable foundation for future development
• Sales and product teams gained one coherent product story

Adding AI does not automatically make a product simpler. It can easily create more tools, buttons, entry points and decisions for people who already have enough decisions to make.

For LexHero, the strongest solution was not a catalogue of specialised AI products. It was a shared workspace where different capabilities could operate on the same document, preserve context and remain under the user’s control.

The product became less about demonstrating what the technology could do and more about supporting how legal professionals actually work.

That distinction usually separates a useful AI product from a very impressive demo.