Role
Product Designer, System Redesign
CASE STUDY
Product Designer · Legal Practice Management SaaS
CasePro: rebuilding matter creation around what a client says, not what a form demands.
Where intake and matter creation were the same overloaded form — until they became two connected systems, one AI-assisted and one fully manual.
Domain
Legal Practice Management · CRM
Core Problem
Intake Merged Into Matter
Outcome
2 Linked Modules, 2 Creation Paths
01 / OVERVIEW
What is CasePro
A legal CRM where intake and matter stopped being the same object
CasePro is the practice-management layer law firms use to run a case from the first client call to close — matters, parties, incident facts, documents, deadlines, and legal-team staffing, all in one system.
Designed around two different jobs, not one form. Before this redesign, opening a new case meant filling out a single "New Matter" wizard that mixed a client's raw intake story with case metadata, incident facts, legal-team assignment, and outcome tracking — six tabs deep, with the client's intake reduced to a free-text box labeled Notes or links to intake form.
Rebuilt as two linked systems, not one mega-form. Intake became its own structured, firm-customizable questionnaire. Matter creation became a deliberate second step that could pull from that intake automatically, or start from a blank slate when a firm chose to build the record by hand.
The Opening Insight
Sitting with intake specialists, I watched the same sequence three times in one morning: transcribe a client call into a text box, open a brand-new matter, then manually retype half of what was just written into six more tabs. The form wasn't slow because of weak UI polish — it was doing the job of two different tools, and doing neither one well.
02 / THE PROBLEM
The Problem
Every "New Matter" started the same way: guessing what the client already told someone
The legacy matter form, still visible below, treated intake as an afterthought and treated every case — a two-car accident, a medical malpractice claim, a commercial dispute — as one identical, generic slog.
Intake Was a Text Box
- The only place client-provided information lived inside a new matter was one freeform field: Notes or links to intake form.
- Nothing structured, nothing reusable, nothing a downstream workflow could act on.
- Every fact a client had already shared still had to be re-typed by hand, field by field.
One Form, Every Practice Area
- The same generic fields applied whether the matter was an auto accident, a slip-and-fall, or a commercial dispute.
- Firms couldn't tailor intake questions to their own region, county, or practice group.
- Questions that mattered for one case type showed up as clutter on every other case type.
Nothing Carried Forward
- If a client had already answered questions during intake, none of it flowed into the matter.
- Paralegals re-typed names, dates, and incident details a second time, by hand, from scratch.
- There was no signal for who filled in a field, or how confident anyone was that it was correct.
Where It Broke Down
Two structurally different jobs — capturing what a client says, and building the legal record — were forced through one form. Every matter, regardless of size or complexity, hit the same rigid eight-section sequence, with no way to skip a step, customize a field, or reuse anything a client had already told the firm.
03 / RESEARCH & DIAGNOSIS
Root Cause
The bug wasn't the form. It was the data model underneath it
I sat with the people who processed new matters every day — intake specialists, paralegals building files after client calls, and the ops lead who owned the CRM configuration. Rather than starting from a blank redesign, I traced the existing form field by field: which ones were duplicated between "intake" and "matter," which ones nobody ever filled in, and which decisions — statute-of-limitations deadlines, legal-team staffing — had no system support at all and lived entirely in someone's memory.
How I confirmed this was a real cost, not an opinion
A redesign is easy to justify with a strong opinion and one bad screenshot. Before proposing anything, I wanted three independent signals to agree with each other, not just my own read of the form.
Shadowing, Not Just Interviews
- Sat through live matter creation with intake specialists and paralegals over two weeks, across several new matters end to end.
- Timed where they paused, backtracked to a client's notes, or pulled in a colleague to ask "wait, did we already ask this?"
Reading the CRM's Own Data
- Pulled a sample of recently created matters and checked which fields were actually populated versus left blank or default.
- The Intake Questionnaire field was empty, or just a name with no real content, on the large majority of them — meaning the structured intake tab downstream was barely ever opened at all.
Structured Interviews, Reframed
- Separate 1:1s with the ops lead and two attorneys — not "what's wrong with the form," but "walk me through the last matter you opened."
- That framing surfaced friction people had stopped noticing, because they'd built quiet workarounds for it.
All three pointed at the same root cause from three different angles: the form wasn't badly laid out. It was structurally answering two different questions — what did the client tell us and what is the legal record — as if they were one question, and everyone touching it had quietly adapted around that instead of naming it.
That trace surfaced the actual bug. It wasn't a layout problem, and more validation rules wouldn't have fixed it.
Intake and Matter aren't two steps of the same thing. Intake is what a client tells you, once, in their own words. A Matter is the legal record your firm builds from it — and firms need to build that record two different ways, depending on how much of the client's story is already captured.
04 / SOLUTIONS CONSIDERED
Options on the Table
Three ways to fix it. Two were faster to ship — and wrong
Once the root cause was clear, the fix wasn't obvious yet — it rarely is. I brought three real options to stakeholders, not one polished recommendation, because the fastest option and the AI-forward option were both genuine temptations that would have shipped a version of the same bug under a nicer coat of paint.
A — Polish the Existing Form
- Reorganize the same fields into cleaner steps, add validation, keep intake as a field inside Matter.
- Fastest to ship — no data-model change, no migration.
Rejected — a better-looking version of the same bug. Intake still couldn't be customized per firm, still couldn't be reused, still only existed inside a matter that already had to be created first.
B — One Button, Fully Automatic
- A single "Generate Matter" action that always builds the record from whatever intake data exists. No manual path at all.
- The most "AI-forward" pitch — the one stakeholders liked first.
Rejected — assumes every matter starts from a complete intake. Many don't: walk-in clients, a five-minute phone note. Forcing AI-only creation onto those cases meant either blocking the user entirely or letting the model quietly fill gaps in a legal record with no human in the loop.
C — Separate Intake, Let the User Choose
- Intake becomes its own customizable entity, linked to a matter instead of buried inside one.
- Matter creation forks: AI-assisted where intake data supports it, manual where it doesn't.
Chosen — the only option that fixed the data-model root cause and matched how matters actually start: sometimes with a client's full story already captured, sometimes with almost nothing.
Why Not Just Ship the AI-Forward Version
Option B demoed the best. It was also the one where a bad guess would go straight into a legal record with nobody checking it — the CRM data review had already shown that a large share of matters start with incomplete or missing intake. Shipping it as the only path would have optimized for the confident-data case and quietly broken on the common one.
05 / DESIGN THESIS
Two Principles
Two decisions that reshaped the entire system
Principle 1 — Intake and Matter are separate, linked entities
Intake became its own object, with its own lifecycle, instead of a field inside a matter. That unlocked something the old model structurally couldn't: a firm can now define its own intake questions — the accident-type list for a state that regulates it differently, the statute-of-limitations triggers specific to a county, the extra fields a personal-injury practice needs that a commercial-litigation practice never will — without an engineer touching the matter schema, and without every firm on the platform being forced onto one generic questionnaire.
To reduce the time it takes a client's story to become usable data, intake capture itself was designed to be AI-assisted at the source: a voice agent that can answer or join a client's call, transcribe it, and populate the structured intake fields directly — instead of a paralegal typing rough notes into a blank box after the fact and hoping nothing was missed. Whether a field lands there from that call, a web form the client filled in, or a paralegal's manual entry, it becomes a candidate the matter can pull from later.
Principle 2 — Matter creation needs two honest paths, not one silent assumption
Not every matter starts with a complete intake. Some arrive from a thorough client call with every field captured; others start from a five-minute phone note or a walk-in client with almost nothing on file yet. A single "AI-generate the matter" button that assumes perfect intake data isn't a feature — it's a bet the product would force the user to take on every single case, whether or not the data supported it. So matter creation became a deliberate fork: let the person creating the matter tell the system how much is already known, and route them accordingly.
An "AI-generate the matter" button that silently assumes perfect intake data isn't a feature. It's a bet the product forces the user to take on every single case.
06 / MATTER CREATION
The Entry Point
Every new matter starts with one honest question: how much do we already know?
Instead of hiding the AI-vs-manual decision inside a settings toggle or defaulting silently to one mode, it's the very first screen after "Create New Matter" — presented as two equally legitimate paths, each with its own tradeoffs stated up front rather than discovered halfway through the flow.
Create with AI Agent
Best when intake exists- Pulls available intake information automatically.
- Organizes it into the right matter sections.
- Saves time and reduces manual re-entry.
- Still stops for review and edits before anything is created.
Create Manually
Best without intake data- Full control over every field, from a blank slate.
- The same guided six-step workflow as the AI path.
- Add details at the user's own pace, save a draft anytime.
- Still reviewed in full before the matter is created.
07 / PATH A — CREATE WITH AI
Letting the Agent Work, in the Open
The AI path never hides how confident it is
The risk with any "let AI build it for you" flow is that it either overpromises silently or leaves the user with no way to tell what came from real data versus a guess. The AI path was built to make that distinction visible at every step, from the moment it starts reading the intake record to the final review screen before a matter is created.
1. Understand intake
→
2. Prepare draft
→
3. Review & edit
→
4. Ask the assistant
→
5. Review before create
1. Understanding the intake, before touching the matter
Before drafting anything, the agent scores exactly how usable the linked intake record is — how many fields are ready to use as-is, how many need a human glance, and how many are missing outright. That score is shown to the user, not buried in a log.
2. Preparing the draft, with visible, cancellable progress
The agent narrates its own work step by step — finding the intake questionnaire, reading client information, identifying parties, extracting incident details, checking legal information — and states plainly that it's running in the background, so the user is never stuck staring at a spinner or forced to babysit the screen.
3. A prefilled matter that labels its own sources
This is the detail that makes the whole path trustworthy: every field in the drafted matter carries a provenance tag — From Intake, AI Matched, or AI Suggested — so a reviewer knows in one glance whether a value is a verified fact, a confident inference, or a guess that deserves a second look before the matter goes live.
4. An assistant to close the remaining gaps
Rather than sending the user back to the intake readiness score to figure out what's missing, an in-context assistant offers direct next actions — review and suggest changes, fill missing information, generate a case summary, or find similar past matters — right where the reviewer is already working.
5. Nothing is created without a full review
Regardless of how confident the draft is, the last step of the AI path is identical to the manual path: every section, grouped and labeled, with an explicit Ready to Create Matter confirmation before the record is written. AI assistance ends where the legal record begins.
Why Provenance Tags Mattered Most
Every other decision in the AI path — the readiness score, the narrated progress, the review gate — exists to support one rule: nothing in a matter should look identical whether a human confirmed it or the model guessed it. That rule shaped the field-level tagging system more than any single screen.
08 / PATH B — CREATE MANUALLY
Same System, No Shortcuts
Manual creation isn't the fallback path — it's the same system, minus the head start
A common failure mode in "AI-assisted" products is treating the non-AI path as a second-class, unmaintained fallback. Here, both paths share the exact same six-step information architecture — Matter Overview, Parties Involved, Incident Details, Legal & Insurance, Team Assignment, Images & Outcome — so a user's mental model never breaks depending on which button they clicked first.
09 / SHARED DESIGN SYSTEM
Craft Details
Six small decisions that hold the whole system together
None of these are headline features on their own. Together, they're the difference between a redesign that looks clean in a screenshot and one that a paralegal actually trusts on a real case.
Field-Level Provenance
Every AI-touched field is tagged From Intake, AI Matched, or AI Suggested — never presented as one undifferentiated "auto-filled" block.
Confidence-Based Readiness
Intake completeness is scored per category before the AI path even starts drafting, so the user chooses a path with real information, not a guess.
Grouped Cards Over Flat Forms
Both paths replace the legacy's undifferentiated field list with labeled sections — Core Details, Classification & Tracking — so a long form still reads as a set of decisions.
Smart Autofill & Legal Calculations
Selecting an incident address auto-fills city, state, and zip; the statute-of-limitations deadline auto-calculates from the incident state and date, with a manual override always available.
Round-Robin Staffing
Team-assignment roles surface the firm's own round-robin logic as a suggestion, so staffing a matter isn't starting from twelve empty dropdowns.
One Review Gate, No Exceptions
Both paths converge on the identical grouped review screen before "Create Matter" — AI assistance speeds up the draft; it never skips the confirmation.
10 / RESULTS & IMPACT
Impact on the People Who Use This
Back to the paralegal retyping the same story into six tabs
That's where this case study started — someone doing real, careful work on a client call, then watching almost none of it survive into the matter they had to build next. Here's what changed for each of them, specifically.
Intake Specialists
What they capture on the client call is no longer a side note someone else re-types later — it's the record the matter itself is built from, structured to their firm's own questions from the start.
Paralegals
No more silently reconciling "what the client said" against "what the matter form wants." Either they confirm an AI-drafted record field by field with visible provenance, or they fill in one clean guided form — never both, never twice.
Attorneys & Firm Admins
A matter arrives at the same reviewable, grouped record no matter which button someone clicked to create it — so trust in the system doesn't depend on knowing which path was used. Firms can also reshape their own intake questions without opening an engineering ticket.
What Changed Structurally
1 → 2
Linked entities instead of
one merged intake-as-matter object
one merged intake-as-matter object
2
Honest entry paths
instead of one silent default
instead of one silent default
3
Field provenance states —
From Intake, AI Matched, AI Suggested
From Intake, AI Matched, AI Suggested
1
Shared six-step IA
across both AI and manual creation
across both AI and manual creation
Key Design Decisions
Fix the Model, Not the Screen
The redesign started with the data model — separating intake from matter — before a single new screen was drawn, so the UI changes actually resolved the root cause.
Never Hide the Confidence Level
Readiness scores, narrated progress, and field-level provenance tags all exist for the same reason: an AI-assisted legal record has to show its work.
Treat the Manual Path as a Peer
Manual matter creation shares the AI path's information architecture and its smart-field assists, rather than shipping as the unmaintained fallback.
The old form failed for the same reason most "combine two workflows into one screen" decisions fail: it optimized for fewer clicks in a demo, not for what a paralegal actually needs to trust on a real case. Splitting intake from matter meant more screens — and, for the first time, a system firms could actually make their own.
