Insurance AI agent · Legal-bill review
Legal-Bill Review & Litigation-Spend AI Agent
Read each outside-counsel invoice against the litigation guidelines for that matter, flag entries that don't match, track spend against reserve, and hold every adjustment for the litigation manager who decides.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Manager review
- Pre-adjustment
- After launch
- Agent Care
What this agent does
Prepares the adjustment, not the decision to make it
In
- 01
An invoice arrives from outside counsel, and the agent pulls the guideline set for that matter.
- 02
A matter opens, and the agent maps entries, task codes and expense lines against the guideline.
Reason
- 03
Each entry is compared to the guideline provision it falls under, not judged on its own.
- 04
The guideline applies only once the matter's jurisdiction is confirmed to enforce it.
- 05
Every flagged entry is bound to the guideline clause and the invoice line it was read against.
Decide
- 06
Spend against the matter's litigation budget and reserve is tracked using the erosion rule the policy sets.
- 07
The flagged review is routed to the litigation manager, with any reservation-of-rights conflict escalated separately.
Out
- 08
The entry, the guideline it was read against, the adjustment reason and the consent record stay on the matter.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent prepares the flagged review and its reason; the litigation manager decides on any adjustment, and outside counsel is told and can dispute it.
Example workflow
One invoice, receipt to adjustment
- 1Invoice receivedOutside-counsel invoice, matter number and the guideline set that applies to it
- 2Context assembledTime entries, task codes, expense lines, reserve and the matter's guideline, each with a source
- 3Review draftedFlagged entries, proposed adjustment, reason and confidence
- 4Controls appliedGuideline-match checks, budget-and-reserve checks, privileged-narrative handling and confidence threshold
No human action required
Stages 1 to 4 run unaided, and no adjustment is made at any of them — the agent is drafting the review, and the manager's lane opens at the confidence gate.
Goes to the litigation manager to decide.
Adds a legal-ops read before the decision.
Manager review
The review is held with its flagged entries, the guideline and the confidence.
Approve · Edit · Escalate for conflict review
Approved — counsel notified
- 6Matter record updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedAdjustment accuracy, dispute outcomes, reserve movement and post-adjustment corrections
Edits
Every manager edit before the decision is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Adjusting, reducing or rejecting an invoice line.
- Withholding or releasing payment to outside counsel.
- Setting or approving the matter's litigation budget.
- Instructing counsel on strategy, staffing or how to litigate the matter.
Automation boundary
Agent acts unaided
- ✓Read each invoice against the guideline that applies.
- ✓Track spend against the matter's litigation budget and reserve.
- ✓Flag entries that do not match the applicable guideline provision.
- ✓Prepare the adjustment and its reason, and hold it.
Any write happens inside the boundaries agreed at implementation, never ahead of the decision.
- Deciding a coverage or reservation-of-rights question.
- Deciding whether independent counsel is owed.
- Sending detailed invoices to a bill-review vendor absent documented insured consent.
- Changes to the guideline library or adjustment thresholds.
Example output
One entry in the invoice, annotated
Everything the agent flags is attached to the guideline it was read against.
Invoice-review output · single invoice
Illustrative example- Matter
- Products-liability defence, RoR matter
- Flagged entry
- Two associates billed the same deposition prep on the same day
- Stated amount
- $1,240.00
- Guideline clause
- Block-billing guideline
- Confidence
- 89%
- Attribution
- Litigation manager of record
As received
Taken from the invoice and the guideline citation given — nothing on this side is written by the agent.
Guidelines applied
- Block-billing clause
- Task-code requirement
- Prior invoice history
Why this flag
It states what the guideline requires, not a judgment on the work.
What the score decides
Below the configured threshold, the review picks up a legal-ops read before the manager sees it.
Action
ApproveEditEscalate for conflict reviewValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every invoice
From the matter record
Read against the guideline
Draw on the guideline that applies and the matter's configured budget and reserve rules.
The insured is the client
Routine entries that match the guideline are cleared without a flag.
Point the manager at what
Entries that don't match the guideline are flagged, so review starts where spend is at risk.
Build an evidence trail
The entry, the guideline it was read against and the manager who adjusted it stay on the matter.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
E-billing & matter systems
- Legal Tracker
- CounselLink
- Collaborati
- Passport
Guideline & budget data
- Litigation guidelines
- Reserve and budget records
Claims & policy systems
- Claim file
- reserve ledger
- Guidewire
- Duck Creek
Agent
Legal-bill review & litigation spend
- Reads the guideline
- Flags the entry
- Holds for the manager
Counsel correspondence
- Dispute tracking
- Adjustment correspondence
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the invoice
Each layer contains the next. What escapes all of them is named in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether the flag may stand
- L3
- Puts the adjustment in the manager's hands
- L4 – L5
- Keep the entry and the guideline behind it
- L6
- Falls back to budget reporting when signals degrade
How Nestack evaluates it
Evaluate the review workflow — not only the finished adjustment.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the adjustment counsel sees
- E1Final-output evaluationDid every flagged entry match the guideline clause it was read against?
- E2Step-level evaluationDid the agent use the right guideline, budget and reserve configuration?
- E3Tool evaluationDid it read and write the correct matter and the correct invoice field?
- E4Confidence calibrationDo low-confidence flags actually attract more manager edits?
- E5Slice evaluationHow does performance change across specific matter types?
- E6Business outcomeHow many flags needed a manager edit or drew a dispute from counsel?
Floor — the outcome the carrier answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, placed at the stage each one originates.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- LB-03
Superseded guideline version
The matter's litigation guideline is read from a version the carrier has since revised.
Stage gathers
Invoice, matter number, guideline and reserve
02 · Reasoning
2 modes
- LB-04
Guideline misapplied
A guideline is applied in a state where it may not be enforceable against counsel.
- LB-06
Adjustment overstated
A proposed adjustment is presented as decided, not as a flag for the manager.
Stage proposes
Flagged entries, adjustment and confidence
03 · Tool / write
2 modes
- LB-02
Confidence dropped in transit
A flagged entry reaches the manager without its confidence attached.
- LB-05
Duplicate invoice processed
The same invoice is reviewed twice under two different matter numbers.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- LB-01
Privilege flag dropped
A narrative entry carrying privileged detail is surfaced without the privilege marker.
Stage returns
The adjustment the manager decides and counsel sees
05 · Change / Version
1 mode
- LB-07
Silent guideline drift
A guideline or reserve-configuration change widens what the agent will flag unreviewed.
Stage tracks
Model, prompt, guideline and reserve configuration
- Sev-1 · adjustment made outside the boundary
- Sev-2 · counsel not told of an adjustment
- Sev-3 · guideline degrades, routes to review
Affected slices
A clean docket can hide where spend concentrates
A portfolio adjustment rate can look controlled while a handful of firms or matter types carry most of the flagged spend. Nestack reports the adjustment-dispute rate by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Matters under a reservation of rights | 7.8% | 4.0× | Review | |
| Block-billed and grouped time entries | 5.3% | 2.7× | Review | |
| Expense and vendor pass-through lines | 3.5% | 1.8× | Watch | |
| Task-coded time inside the guidelines | 1.9% | 0.8× | Normal |
Bar: adjustment-dispute rate lift vs. the task-coded baseline · scale 0–4.0× · tick at 2.0×
2 of 4 slices over threshold
Evidence-linked improvement
Every cycle closes on one more matter
A cycle is closed when the wrong adjustment is a case the next release must survive. That suite is what the next invoice reviewed is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Adjustment-dispute rate rises in a matter slice.
- 02
Diagnose
The line that was cut without anyone telling counsel is traced back through the guideline and the invoice until the cause narrows to one entry.
- 03
Improve
The change ships against a version, with the matters that exposed it attached.
- 04
Verify
Release is blocked until the affected adjustment cases pass again.
- 05
Learn
The case joins the standing suite and the guideline mapping moves with it.
Learn → Detect
The return edge. Detection next time runs against a suite one matter longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, guideline sources, review workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Review workflow discovery and scope and boundary definition.Week 1
- 02Matter and e-billing system assessment.Week 1, Week 2
- 03Guideline library and reserve mapping and rule mapping.Week 1, Week 2
- 04Invoice ingestion and normalisation.Week 2
- 05Flagging logic and guideline binding.Week 2, Week 3
- 06Confidence scoring and escalation routing.Week 3
- 07Litigation-manager review workflow.Week 3, Week 4
- 08E-billing and claims-system integration.Week 5
- 09Adjustment and privilege cases.Week 4, Week 5
- 10Guardrails and dispute controls.Week 4
- 11Matter-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne guideline set, one matter | ProductionProduction e-billing systems | AdvancedMultiple guideline sets / firms |
|---|---|---|---|
| Introduced at Pilot | |||
| Flagging to your guideline and reserve rules | ✓ | ✓ | ✓ |
| Manager review | ✓ | ✓ | ✓ |
| Adjustment-accuracy baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by firm | — | ✓ | ✓ |
| Manager workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Matter-system integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-jurisdiction guideline complexity | — | — | ✓ |
| Multi-stage litigation approvals | — | — | ✓ |
| High invoice volume | — | — | ✓ |
| Multi-jurisdiction defence controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, invoice volume, review controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your litigation guideline library and reserve rules
→Guideline ingestion and matter mapping
Week 102Representative past invoices, including disputes
→Flagging baseline and guideline-binding baseline
Week 203Your budget thresholds and reserve rules
→Guideline library and reserve mapping
Week 104Access to relevant APIs, feeds or exports
→E-billing and claims-system assessment, then integration setup
Week 205Adjustments you would not want made
→Guideline cases and failure-mode testing
Week 406What no adjustment may be made without
→Confidence scoring, escalation routing, guardrails and dispute controls
Week 307Named litigation managers to review flags
→Manager review workflow, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Each phase sits over the weeks it really occupies, and week 5 carries evaluation alongside the pilot.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Review workflow discovery, guideline mapping and the automation boundary
W2E-billing and claims-system integration and the flagging baseline
W3Flagging workflow, confidence logic and manager-review controls
W4Evaluation suite, privilege-handling checks and failure-mode testing
W5E-billing integration, pilot invoices and targeted corrections
W6One billing cycle reviewed under the litigation manager, then handover
- Reading the band
- A bar covers the weeks its work is named in, and nothing else. The week 5 overlap is real, not padding.
- At the end of W6
- The last checks clear on live invoices and Agent Care picks up monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Insurance AI agent
Build a legal-bill review agent around your litigation guidelines.
Show us your litigation guidelines, your reserve rules and who decides an adjustment. Not a cost-cutting tool. A guideline match held for the litigation manager, with counsel told and able to dispute it.
- Typical delivery
- 4–6 weeks
- Deployment
- Your e-billing systems
- Manager review
- Pre-adjustment
- After launch
- Agent Care available
Nestack Agents · Legal-bill review & litigation spend
AGT-INS-18 · Agent Care available after launch