Case study

Designing for Revenue Integrity at Scale

Helping Charge Master Analysts find important changes, understand their impact, and act safely with AI.

Hospitals use a Charge Master to manage the codes and prices behind their services. Those prices never stop changing.

I led the design of an AI experience that puts expert judgment where it matters most.

Role
Lead Designer
Team
2 Designers · 2 PMs · 19 Developers · 4 Data Scientists
Year
2025
Impact

38% faster decisions

33% fewer verification checks

35% lower analyst touch time

88% appropriate AI reliance

Watch the quick overview

A quick overview of how an automation brief became a way to share work between AI and experts.

01 / Context

Think of the Charge Master as a hospital's master menu

Every test, drug, and procedure has a code and a price. Unlike a restaurant menu, this one never stops changing.

Updating the menu wasn't the hardest part. Knowing what needed to change was.

A Charge Master Analyst reviewing a hospital's charge master binder, a medical menu of tests, imaging, surgery, and supplies, beside a restaurant menu with soup, steak, and wine.
Charge Master Analyst

Meet the person behind every change

Charge Master Analysts keep this system accurate.

But they do much more than update records.

They watch for changes, figure out what they mean, find the right experts, understand the money involved, get approvals, and make sure the final update is safe.

This is not a data-entry job. It is a decision-making job.

Watches
Regulations · Codes · Costs · Supply issues
Figures out
What changed · What it affects · What to do
Works with
Coding · Finance · Clinical teams · Revenue Integrity
02 / Problem

It looked like a data-entry problem. It wasn't.

Analysts spent more time finding, understanding, and coordinating than updating.

  1. Important changes were hard to find

    They often started outside hospital systems, so analysts checked many sources by hand.

    The problem was noticing change early enough.

    Before
    Many sources
    Regulations, vendors, costs, coding
    Analyst
    Checks each one by hand
  2. Understanding a change meant chasing people

    Every change meant explaining it to one expert after another.

    Experts kept rebuilding the same context.

    Before automation
    50 min
    to reach a decision
    22%
    of changes needed rework
    Before
    Analyst
    with one change
    • Coding expert
    • Pharmacy
    • Department leader

    Explained again and again.

  3. Every change got almost the same attention

    Routine changes got the same detailed review as risky ones.

    The most important work did not get enough.

    Before automation
    65 min
    of analyst time per change
    Before: the same review for everyone
    • Low riskFull review
    • Medium riskFull review
    • High riskFull review

Research revealed two kinds of work

Reactive

The change is already known.

  • New regulation
  • New code
  • Vendor price increase
Proactive

The analyst has to discover the problem.

  • Rising costs
  • Supply disruption
  • Margin pressure
An analyst seen from behind at an open charge master binder. On the left, labeled Reactive, a regulation document arrives in an inbox and flags a row for review. On the right, labeled Proactive, vendor supplies and trend lines show cost signals the analyst must investigate.

Different problems. Same decision pattern.

  1. Detect
    What changed?
  2. Understand
    Does it matter?
  3. Assess impact
    What is affected?
  4. Decide
    What should we do?
  5. Act
    Who needs to help?

The real bottleneck was not updating data. It was turning change into a safe decision.

Research changed the question

We started with
“How might we automate Charge Master maintenance?”
We ended with
“How might we help analysts find meaningful changes, understand their impact, and safely share work with AI?”
03 / Goals

Four things needed to improve

  1. Goal 1

    Find the right changes

    Help analysts notice meaningful changes without manually watching everything.

  2. Goal 2

    Understand them faster

    Bring the change, evidence, affected charges, and impact into one place.

  3. Goal 3

    Reduce unnecessary back-and-forth

    Give the right people the context they need before asking them to review.

  4. Goal 4

    Use expert attention wisely

    Let AI handle more routine work while people focus on uncertainty, risk, and important decisions.

One rule for all four

Keep important decisions accountable

Evidence, approvals, and the final commit must stay clear and traceable.

04 / Strategy

AI does the busywork. Experts keep the judgment.

Not every change needs the same amount of human attention.

We called this variable delegation: the amount of work AI does changes with the risk of the decision.

AI does more
Human does more
AI can do more when
  • the evidence is clear,
  • the situation is familiar,
  • and the risk is low.
People do more when
  • the situation is unclear,
  • the change is unusual,
  • or a mistake could have a large impact.
05 / Solution

Eight changes that put the idea to work

The first five answer the problems above. The last three make AI work usable and safe.

1 · Solves: hard to find

Help analysts notice what matters

AI watches the sources and flags possible changes.

  1. 01AI monitors all sources
  2. 02AI finds the signal
  3. 03Analyst decides
2 · Solves: chasing people

Bring the evidence together

Before asking an expert for help, AI answers five questions:

  1. 01What changed?
  2. 02Does it matter?
  3. 03What is affected?
  4. 04What should we do?
  5. 05Who needs to help?
3 · Solves: chasing people

Send the right context to the right person

Each reviewer sees the same change in their own terms.

One change. Different views. Shared evidence.

4 · Solves: same attention

Match attention to risk

Routine changes move faster. Risky ones get more human attention.

Experts don't disappear. Their time moves to where it matters most.

After: attention matches risk
  • Low riskQuick check
  • Medium riskStandard review
  • High riskDeep review with approvals
5 · Keeps it accountable

Keep important decisions human

AI can prepare the update. People approve it before it is committed.

AI prepares the decision. A person owns the decision.

6 · Keeps it safe

Explore freely. Commit carefully.

Analysts can explore options with AI as freely as they like. Once an action has consequences, approvals are required.

The gate is built into the product, not left to process.

7 · Keeps it usable

Show answers as they arrive

AI answers stream in, so analysts start reading before the full answer is ready.

Waiting feels shorter when something is already on screen.

8 · Keeps it trustworthy

Give AI a memory with limits

Analysts can see what the AI remembers, and correct it.

  • What will it remember?
  • For how long?
  • Where does it apply?
  • Can I correct it?

When AI is wrong, the fix is visible, not silent.

The same idea works differently depending on the problem

Reactive

The change is known. AI can do more of the preparation.

AI
  • Triages the batch
  • Prepares the update
  • Gathers reviews
Person
  • Handles exceptions
  • Approves
  • Commits

AI prepares the action.

Proactive

The problem is still emerging. AI helps explore it.

AI
  • Finds the signal
  • Gathers evidence
  • Explores options
Person
  • Weighs the trade-offs
  • Makes the call
  • Commits

AI prepares the decision material.

06 / Results

Did we achieve the goals?

Goal 1

Find the right changes

Less noise. Fewer misses.

91%
of flagged changes were relevant
86%
of important changes were caught
Goal 2

Understand them faster

Faster, and just as sure.

−38%
time to decision
Before 50 minAfter 31 min
−33%
verification checks
Before 9After 6
+28%
decision confidence
Before 3.2 / 5After 4.1 / 5
Goal 3

Reduce back-and-forth

Less repeating the same work.

−18%
rework rate
Before 22%After 18%
Goal 4

Use expert attention wisely

Less manual work. Better judgment.

−35%
analyst touch time per change
Before 65 minAfter 42 min
88%
appropriate AI reliance
Rule

Keep important decisions accountable

AI explains its work. People own the outcome.

95%
of recommendations traced to a source
Downstream impact

Room for the same team to do more

A modeled estimate, not a measured result.

+25%
analyst capacity
16 → 20 changes / analyst / week

What actually changed in the work?

  1. 1

    Detect

    Before: Watch many sourcesAfter: AI surfaces credible signals

  2. 2

    Understand

    Before: Rebuild the story by handAfter: AI prepares the evidence

  3. 3

    Assess

    Before: Connect charges by handAfter: Impact is ready for review

  4. 4

    Collaborate

    Before: Explain the same change againAfter: Context travels with the change

  5. 5

    Decide

    Before: Build the recommendation from scratchAfter: AI prepares options, the analyst judges

  6. 6

    Act

    Before: Prepare and commit everythingAfter: AI prepares, a person approves

Before, experts spent their time assembling the decision.

After, they spent more of their time making the decision.

07 / Reflection

The goal was never maximum automation

The project started with a question about automating Charge Master maintenance. It ended with a model for deciding when automation helps.

AI does more when the situation is clear and predictable. Experts take more control when it is unclear or important.

The result was not less human judgment. It was better use of human judgment.

What comes next

  1. 01

    Adaptive delegation

    Let the system recommend the right level of AI involvement based on risk, uncertainty, and past performance.

  2. 02

    Organizational memory

    Reuse prior decisions, policies, and corrections when similar changes happen again.

  3. 03

    Close the loop

    Watch what happens after a change is committed and use those outcomes to improve future recommendations.

Detect, Understand, Assess, Decide, Act, Learn