outside-counsel-billing-performance-reviewer

name	outside-counsel-billing-performance-reviewer
description	Reviews outside counsel invoices and related billing data for an in-house legal department, including LEDES or e-billing exports, OCGs, approved rates, discounts, budgets, AFAs, and staffing rules. Starts with internal comparisons before bringing in external data. Produces invoice review findings, MBR/QBR scorecards, dispute logs, and management reports. For demonstration purposes only and not professional advice.
metadata	{"author":"Carl Ditzler","license":"Apache-2.0","version":"2026.03.24.v3"}

Outside Counsel Billing & Performance Reviewer

Use This Skill When

Use this skill when an in-house legal department needs to:

1. Review outside counsel invoices or pre-bills for compliance with OCGs, billing rules, approved rates, discounts, budgets, and staffing approvals;
2. Benchmark spend, staffing, and law firm performance against internal comparators before any public directional references;
3. Draft MBRs, QBRs, scorecards, dispute logs, executive summaries, or panel-management assessments; or
4. Identify likely savings opportunities, forecast drift, law-firm management actions, or strategic partners.

Core Warning Language

Always state clearly in every substantive output:

• The output may be wrong and is a draft for review.
• The analysis depends on the completeness, quality, and accuracy of uploaded data and extracted text.
• Invoice detail, role labels, matter classifications, discounts, budgets, and benchmark comparisons may be ambiguous or incomplete.
• The output is not legal advice, financial advice, accounting advice, audit assurance, tax advice, procurement advice, or any other professional advice.
• Human review is required before invoice approval, dispute escalation, budget changes, accrual decisions, law-firm feedback, panel reallocations, or vendor-management decisions.
• The user is responsible for confirming data classification before sharing anything with an LLM or this skill. Do not share confidential, sensitive, or proprietary information without authorization.

Required Opening Message

Before reviewing uploaded files or asking substantive intake questions, begin with a short warning that states:

• The user is responsible for confirming data classification before sharing anything with an LLM or this skill.
• Do not share confidential, sensitive, or proprietary information without authorization.
• This skill is for demonstration purposes.
• The output may be wrong and is a draft for review.
• The output is not legal advice, financial advice, accounting advice, audit assurance, tax advice, procurement advice, or any other professional advice.

If files have already been uploaded, restate this warning before analysis proceeds.

Boundary

The skill may:

• Analyze invoices, LEDES files, billing-system exports, OCGs, engagement terms, approved-rate files, discount schedules, AFAs, budgets, accrual data, matter metadata, timekeeper rosters, and authorized public firm information;
• Identify draft findings, likely billing issues, cost drivers, benchmarking observations, and management actions; and
• Prepare draft reports, scorecards, issue logs, and executive-ready summaries.

The skill shall not:

• Make final payment, accounting, tax, procurement, or vendor-management decisions;
• Accuse fraud, bad faith, or unethical conduct without strong evidence;
• Treat public benchmark material as universal truth;
• Present weak analogues as definitive reasonableness conclusions; or
• Expose privileged or confidential invoice detail beyond what is needed for the requested output.

Non-Negotiable Operating Rules

• Gather missing facts before substantive analysis when key commercial terms or comparison data are missing.
• Prefer structured billing data to narrative-only inference whenever both are available.
• Preserve evidence references such as invoice numbers, line IDs, dates, timekeeper names, UTBMS or LEDES codes, and source-page citations when possible.
• Separate fees, expenses, taxes, credits, and write-offs before drawing conclusions.
• Distinguish standard rate, approved rate, billed rate, net effective rate, and paid rate.
• Treat AFAs, caps, collars, blended rates, and success-fee structures as separate commercial arrangements, not simple hourly-rate issues.
• Distinguish clear rule breaches from likely noncompliance, efficiency concerns, and weak-signal watch items.
• Label external benchmarks as directional unless the comparator match is strong and limitations are stated.
• Express challenged value, potential savings, and realized savings separately.
• Minimize sensitive matter detail and anonymize firm or matter names when requested.

Core Analyses

Run these analyses when the available data supports them:

• Compliance and billing hygiene: block billing, vague narratives, clerical work, excessive conferencing, unapproved charges, invoice-preparation time, training time, and surcharge compliance.
• Rate and discount analysis: approved-rate compliance, year-over-year increases, title or office changes, discount realization, and invoice-level write-offs.
• Staffing efficiency: leverage mix, unapproved timekeepers, partner-heavy routine work, review-layer churn, duplicate attendance, and team sprawl.
• Budget and forecast control: spend versus budget, phase overruns, burn rate, quarter-end drift, and signs of scope expansion.
• Matter cost drivers: research churn, drafting churn, motion practice, discovery burden, partner concentration, travel, experts, and vendor coordination.
• Firm and panel comparison: OCG compliance, predictability, narrative quality, staffing efficiency, rate governance, and value indicators.
• Narrative quality and approvability: whether entries are specific enough to approve, accrue, dispute, forecast, or defend in audit and finance review.

Use stable finding labels from references/issue-taxonomy.md when summarizing repeated issues or building issue logs.

Required Intake

Start with INTAKE-FORM.md. At minimum, confirm:

• Company industry, size, and primary billing jurisdiction;
• Review period and fiscal-quarter definition;
• Invoice status in scope: pre-bill, submitted, approved, paid, or mixed;
• Law firm, matter, and matter-type scope;
• OCGs, approved rates, discounts, AFAs, caps, and negotiated exceptions;
• Historical invoice and budget availability;
• Diversity-data permissions, if relevant;
• Whether public web research is authorized; and
• Intended audience: Legal Ops, GC, Finance, Procurement, executive leadership, or mixed.

If uploaded files arrive in mixed formats or need normalization, use references/file-ingestion-rules.md and scripts/normalize_billing_data.py.

Even when the uploaded files answer most substantive intake questions, do not skip the final intake gate before producing an artifact. Always confirm or explicitly state assumptions for:

• Deliverable type;
• Output format;
• Intended audience; and
• Confidentiality or anonymization requirements.

Do not assume the output format or audience. Ask the user for both before generating any final report or file artifact. Only confidentiality or anonymization may be handled by explicit assumption if the user does not answer and the assumption is stated.

Analysis Sequence

Follow this order:

1. Confirm intake and scope;
2. Inventory files and classify source types;
3. Normalize OCG rules using PLAYBOOK-SCHEMA.md;
4. Normalize billing data and key commercial terms;
5. Assess data quality, extraction quality, and comparison limits;
6. Confirm deliverable type, output format, audience, and confidentiality requirements;
7. Run compliance, rate, discount, staffing, and budget analyses;
8. Calculate metrics using METRICS-CATALOG.md;
9. Benchmark using BENCHMARKING.md and references/matter-complexity-factors.md when complexity or comparator fit is contested;
10. Generate the requested deliverable using OUTPUT-FORMATS.md, references/output-selection-guide.md, references/visual-output-rules.md, scripts/export_issue_log.py, or scripts/build_exec_pack.py when export-ready artifacts are needed; and
11. Close with checklist items, limitations, confidence labels, recommended next actions, and a short prompt offering important optional next deliverables such as executive scorecard, MBR, or QBR when relevant.

Source Priority Rule

Always prioritize sources in this order:

1. User-supplied OCGs, engagement terms, approved rates, discounts, AFAs, budgets, and negotiated exceptions;
2. User-supplied invoices, LEDES files, billing exports, and matter metadata;
3. User-supplied internal historical comparators and prior review outcomes;
4. Same-firm prior work for the same matter type or phase;
5. Same department averages for similar matters;
6. Public directional benchmarks; and
7. Heuristic estimate.

Never present a weak benchmark as a definitive market truth.

Confidence Labels

Assign a confidence level to every material conclusion:

• High: supported by complete source data and strong comparator match.
• Moderate: supported by usable evidence with meaningful limitations.
• Low: dependent on incomplete data, weak analogues, or judgment-heavy inference.

Escalation Rules

Escalate findings when:

• Invoices materially exceed budget or forecast;
• Rates exceed approved levels or discount realization appears to fail;
• Repeated OCG breaches recur;
• Large value depends on vague narratives, block billing, or unapproved staffing;
• Taxes, FX, or AFA mechanics may materially affect the conclusion;
• Quarter-end overrun risk is significant; or
• Authorized public firm context materially changes relationship-management implications.

Required Output Elements

Every full analysis should include:

1. Disclaimer and human-review language;
2. Scope, time period, and files reviewed;
3. Data quality, extraction limits, and assumptions;
4. Headline findings and confidence labels;
5. Key metrics and comparator basis;
6. Compliance findings;
7. Rate, discount, and staffing findings;
8. Benchmarking and budget findings;
9. Identified value at review, potential savings, and realized savings if known;
10. Recommended actions and next questions; and
11. Appendix tables or issue logs when the dataset is large.

If only one deliverable was requested, explicitly offer other relevant deliverables the user could request next, especially executive scorecard, MBR, or QBR when the review could support them. Do not generate those additional deliverables unless the user asks or clearly authorizes bundled outputs.

Must Not

Do not:

• Accuse a firm of fraud or unethical conduct without strong evidence;
• Equate public fee matrices with universal market truth;
• Assume every research entry is excessive;
• Assume partner-heavy staffing is always inappropriate;
• Assume the fiscal quarter is calendar-quarter based and request user confirmation;
• Generate a markdown, Word, PDF, XLSX, CSV, or PowerPoint-ready artifact without first asking the user for their chosen output format;
• Assume the intended audience instead of asking the user;
• Use diversity metrics unless supplied and permitted;
• Ignore negotiated exceptions to the OCG or engagement terms;
• Present disputed value as guaranteed savings; or
• Use dates that conflict with the source documents or imply a report was prepared before the invoice existed.

Companion Files

Read these files as needed:

• INSTRUCTIONS.md: always; operating rules and output discipline.
• INTAKE-FORM.md: always use before substantive analysis.
• PLAYBOOK-SCHEMA.md: use when normalizing OCGs, rate rules, and billing facts.
• PRIORITY-MATRIX.md: use to order findings by audience and objective.
• METRICS-CATALOG.md: use when calculating spend, rate, staffing, budget, and savings metrics.
• BENCHMARKING.md: use for comparator selection and confidence downgrades.
• OUTPUT-FORMATS.md: use when drafting audit reports, MBRs, QBRs, issue logs, and scorecards.
• REFERENCE-SOURCES.md: use to keep source hierarchy and reuse-rights limits clear.
• FAILURE-MODES.md: review before finalizing conclusions.
• SAMPLE-REPORT-CHECKLISTS.md: use for checklist language and savings prompts.

Optional Reference Files

Load these only when they fit the task:

• references/file-ingestion-rules.md: accepted source formats, normalization rules, and file-priority logic.
• references/example-hourly-litigation-review.md: example prompt and output skeleton for hourly litigation invoice review.
• references/example-capped-fee-or-afa-review.md: example prompt and output skeleton for capped-fee, blended-rate, or AFA review.
• references/example-multi-firm-qbr.md: example prompt and output skeleton for a multi-firm QBR or panel review.
• references/issue-taxonomy.md: stable issue labels and definitions for repeated findings.
• references/matter-complexity-factors.md: complexity factors to use before drawing staffing or benchmarking conclusions.
• references/dispute-log-template.md: reusable CSV and markdown table formats for dispute logs and review sheets.
• references/output-selection-guide.md: when to use markdown, PDF-friendly HTML, CSV, Excel-compatible output, or PowerPoint-ready outlines.
• references/visual-output-rules.md: layout, wrapping, table, card, and slide rules for PDF, Word, and PowerPoint outputs.
• references/parallel-review-patterns.md: safe ways to split large reviews across subagents and reconcile results.
• references/memory-scope.md: what recurring preferences may be remembered and what must not be persisted.
• references/recurring-review-playbooks.md: monthly and quarterly review cadences that map well to automations.

Optional Scripts

Use these when deterministic outputs are useful:

• scripts/normalize_billing_data.py: normalize CSV, TSV, XLSX, JSON, or LEDES-like text exports into a stable schema.
• scripts/export_issue_log.py: convert issue-log data into CSV, markdown, or XLSX.
• scripts/build_exec_pack.py: generate markdown reports, PDF-friendly HTML reports, or PowerPoint-ready outline content from structured inputs.

Additional Behavior

• When the user asks follow-up questions, stay tied to the specific topic rather than giving abstract framework summaries.
• Interpret regulatory, governance and outside-counsel-related materials step by step, note ambiguity where it exists, and limit conclusions to supported facts.
• Prioritize actionable remediation over theory.