Case Study
01 · Project
·
Internal Systems Portal
·
B2B Internal Systems
MY ROLE
Product designer, end to end, research, UX, UI, and prototype
TEAM
1 PM, 4 engineers, 1 policy specialist, 1 product designer
TIMELINE
12 weeks, from shadowing study to pilot rollout
CONSTRAINTS
Client data privacy rules, no AI-made decisions, live queues untouched
02 · Evidence
Five experienced reviewers were observed over three working days. 62% of review time went into finding records, repeating checks, and chasing missing documents. This became the baseline for the redesign
Three-day observation log, daily average across five reviewers
Day 1 · 68% searching
Day 2 · 61% searching
Day 3 · 54% searching
62%
of review time was spent preparing the case, not making the decision
Three-day average across five experienced reviewers
The rule was clear: AI could gather records, flag issues, and prepare a summary. The reviewer still had to check the evidence, make the decision, and record the reason
03 · Decision
Before
1 · Search records
2 · Review documents
3 · Check policy
4 · Draft the explanation
5 · Send for approval
After
1 · Evidence gathered, policy gaps flagged, sources linked
2 · Reviewer decides
3 · Decision recorded
What testing changed
The first concept showed an AI recommendation beside an Accept button. Reviewers accepted it before reading the evidence. The recommendation was removed. AI now prepares and flags while the reviewer chooses the action and explains why
The final flow became simple: AI prepares the case, the reviewer checks the evidence, the reviewer decides, and the reason stays with the record
04 · Outcome
Evidence, policy checks, sources, and prepared context now appear together. The reviewer can verify the case, choose an action, and record the reason without moving between five systems
The result came from removing repeated search work, not from letting AI make the decision
Every action still belongs to a named reviewer and every decision keeps its reason
The record stays on the case so the evidence, owner, and reasoning remain clear during review
Next step: expand the pilot to two more request types and check that faster reviews do not lead to more overturned decisions
20 minutes
Five systems and one notebook
Reviewers moved between disconnected tools, repeated the same checks, and wrote every explanation from scratch
6 minutes
One workspace and one recorded decision
Evidence, policy checks, and prepared context appear together. The reviewer checks the case and records the reason for the decision
→ Less time spent searching
→ Evidence easier to check
→ Clear ownership for every decision