Case study

Reimagining Payer Contract Management with Agentic AI

Helping contract experts find what matters, understand the financial impact, and safely share predictable work with AI.

Payer contracts define how insurance companies pay hospitals. A small change to a rate or rule can affect millions of dollars.

I led the design of an AI-assisted contract experience that helps experts find the changes that matter, model their impact, and focus their attention where mistakes cost the most.

Role
Lead Designer
Team
4 Product Managers · 37 Developers · 6 Data Scientists
Year
2026
Impact

62% faster financial insight

62% less rework

78% more contracts per analyst

~$8M new booking value

Watch the quick overview

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

01 / Context

The contract is the map from care to payment

A payer contract tells the hospital how it will be paid.

It holds rules for rates, thresholds, carve-outs, and fee schedules. Those rules become system settings. If the reading or the setup is wrong, the expected payment is wrong too.

A contract is not just a document. It becomes the logic behind reimbursement.

The route from care to payment as a map. A hospital and an insurer sit at either end. A claim travels along a path marked by four points, access, pricing, process, and accountability, and returns as reimbursement.
Contract Manager

Meet the Contract Manager

Contract Managers turn complex payer agreements into something the hospital can use.

This is not a document-processing job. It is a financial decision-making job.

Understand
Rates · Thresholds · Carve-outs · Fee schedules · Payment rules
Evaluate
What changed · What matters financially · Which claims are affected
Configure
Turn contract terms into system rules
Validate
Test expected payment · Investigate differences
Negotiate
Use financial evidence for renewals and payer talks
02 / Problem

It looked like a document problem. It wasn't.

Contracts are long, and AI can read documents, so automating extraction seemed natural. Research showed that reading the contract was only the beginning.

Along the contract's journey, five things got in the way
  1. The terms that matter were buried

    A contract can hold hundreds of terms. Only some change the payment, but experts still reviewed most of them.

    Find the few terms that change the money.

    Before automation
    ~82%
    of surfaced terms reviewed by hand
    Before: most terms get reviewed
    Hundreds of termsA few change the payment
  2. A found term did not show its cost

    Knowing a threshold moved from $200K to $225K is useful. Experts still had to work out which claims it hits and by how much.

    Extraction tells you what changed. Modeling tells you whether it matters.

    Before automation
    95 min
    to reach a useful financial insight
    2.8 days
    to prepare model-ready analysis
    Before
    Threshold
    $200K → $225K
    • How many claims cross it?
    • How much does payment change?
    • Which services are exposed?
  3. Setup was fully manual

    After signing, analysts typed contract language into structured system rules. A wrong setup could still look valid.

    Setup errors were easy to make and hard to see.

    Set up by hand
    RatesThresholdsCarve-outsFee schedulesExclusionsEffective datesPayer rules
    A wrong setup can still look valid.
  4. Mistakes were found too late

    A wrong threshold or mapping may not show an error until test pricing looks wrong or a payment gap appears.

    The problem was finding mistakes downstream.

    Before automation
    21%
    of analysis needed rework
    Before: fixing it meant going around again
    1. Contract
    2. Interpret
    3. Configure
    4. Test
    5. Problem found
    Go around again
  5. Automation raised a new question: when should AI be trusted?

    AI could prepare work much faster. But one wrong reading of an important term could cause a large error later.

    The challenge was deciding how much responsibility AI should have.

    Before automation
    3.1 / 5
    decision confidence
    What experts told us
    Comfortable delegating
    • Predictable work
    Less comfortable
    • Ambiguous language
    • Unusual structures
    • Low-confidence readings
    • Big financial decisions
What the research showed

Analysts did not want AI to review contracts for them. They wanted AI to help them know what deserved their review.

Research changed the question

We started with
“How might we automate contract management?”
We ended with

“How might we help Contract Managers:

  • find the changes that matter financially,
  • understand their impact,
  • set up contracts safely,
  • and share predictable work with AI?”

The goal was not maximum automation. It was better use of expert attention.

03 / Goals

Four things needed to improve

  1. Goal 1

    Find what matters

    Spot the financially important terms without reviewing everything.

  2. Goal 2

    Reach financial insight faster

    Connect contract language to claims, scenarios, and payment impact.

  3. Goal 3

    Configure correctly earlier

    Catch wrong readings before they become rework.

  4. Goal 4

    Focus human review

    AI handles predictable work. Experts handle ambiguity and big financial impact.

One rule for all four

Keep important decisions accountable

AI prepares the work. The analyst can still inspect, correct, override, and commit.

04 / Strategy

Automate the predictable. Surface the consequential.

Five principles guided the design.

  1. 01
    Prevent, don't repair

    Catch errors upstream, not downstream.

  2. 02
    Automate predictable work

    Don't spend expert attention on routine answers.

  3. 03
    Make ambiguity visible

    Show where AI is unsure.

  4. 04
    Design authority

    Decide when AI leads and when people do.

  5. 05
    Scale the learning

    Reuse the rules across the lifecycle.

05 / Solution

How the design answered each goal

One answer for each goal, then one trust model across the whole lifecycle.

Goal 1 · Find what matters

AI surfaces the terms that deserve attention

AI compares the proposal with the current contract and flags the few terms that matter.

The analyst focuses on what deserves judgment.

Goal 2 · Reach financial insight faster

Connect contract language directly to financial impact

AI prepares the analysis. The analyst checks the assumptions and weighs the trade-offs.

Finding the term is not the outcome. Understanding the financial consequence is.

Goal 3 · Configure correctly earlier

AI prepares the setup. Experts check the risky parts.

  1. 1Upload contract
  2. 2Review terms
  3. 3Resolve issues
  4. 4Test price
  5. 5Activate

AI maps terms to settings, flags the uncertain ones, and tests expected payment. If AI is wrong, the analyst fixes that one decision and keeps going.

Catch reading errors before they become payment problems.

Goal 3 · When AI is wrong

Recovery should be local

If AI keeps misreading one decision, the analyst takes more control for that decision only. Once it is fixed, the workflow returns to normal.

AI can lose authority locally without losing usefulness everywhere.

Goal 4 · Focus human review

Move from verify everything to verify what matters

Review shifts toward ambiguous terms, low-confidence readings, unusual structures, and big financial changes.

Routine work gets less expert time. High-impact work still gets scrutiny.

How much gets reviewed by hand
Before~82%
After~37%

Tiles stand for surfaced terms. Review moved to the ambiguous and high-impact ones.

One trust model across the contract lifecycle

  1. Should we accept this proposal?

    1. 1AI: Finds changes
    2. 2AI: Prepares scenarios
    3. 3AI: Models impact
    4. 4Person: Interprets trade-offs
    5. 5Person: Chooses the negotiation strategy
  2. Did we translate the signed contract correctly?

    1. 1AI: Extracts terms
    2. 2AI: Maps settings
    3. 3AI: Flags issues
    4. 4Person: Resolves ambiguity
    5. 5AI: Prepares test pricing
    6. 6Person: Checks the consequential terms
    7. 7Person: Activates
  3. Is the contract still behaving as expected?

    1. 1AI: Sorts deviations
    2. 2AI: Links variances to likely terms
    3. 3AI: Prepares the investigation
    4. 4Person: Investigates exceptions
    5. 5Person: Finds the cause
    6. 6Person: Decides the fix
06 / Results

Did we achieve the goals?

Goal 1

Find what matters

Less review overall. More attention where it mattered.

~82%
reviewed by hand, before
~37%
reviewed overall, after
91%
of high-impact terms still reviewed
90%
of material terms surfaced by AI
Goal 2

Reach financial insight faster

Contract language reached financial analysis much faster.

−62%
time to relevant financial insight
Before 95 minAfter 36 min
−61%
time to model-ready analysis
Before 2.8 daysAfter 1.1 days
Goal 3

Configure correctly earlier

More problems were caught before they traveled downstream.

−62%
analysis needing rework
Before 21%After 8%
Goal 4

Focus human review

Experts relied on AI more selectively and felt more confident.

81%
appropriate AI reliance by month 3
65% → 74% → 81%
+1.1
decision confidence points
Before 3.1 / 5After 4.2 / 5
Capacity

The same team handled more contracts

An observed before and after result.

+78%
contracts per analyst
Before 9 / monthAfter 16 / month
Business impact

New booking value

Easier to read, check, set up, model, and act on.

~$8M
new booking value

What changed in the work?

  1. 1

    Find

    Before: Read everythingAfter: Surface what matters

  2. 2

    Understand

    Before: Interpret by handAfter: AI prepares the reading

  3. 3

    Model

    Before: Assume by handAfter: AI prepares scenarios

  4. 4

    Configure

    Before: Write rules by handAfter: AI sets up, flags doubts

  5. 5

    Validate

    Before: Errors found lateAfter: Fix doubts early

  6. 6

    Review

    Before: Verify everythingAfter: Check what is risky

  7. 7

    Recover

    Before: Rework downstreamAfter: Fix one decision

  8. 8

    Decide

    Before: Rebuild the pictureAfter: Compare with evidence

Before, experts spent their time finding and verifying contract information.

After, they spent more of their time deciding what the contract meant financially.

07 / Reflection

The goal was never maximum automation

It started as a question about automating contract management. It ended as a model for when automation is right.

AI does more when the work is predictable. Experts take control when the language is unclear or the money at stake is high.

The outcome was not more trust in AI. It was better-calibrated trust.

AI handled more of the predictable work.

Experts focused on ambiguity, financial consequence, and final decisions.