The policy should cover passive time capture, matter suggestion, duration suggestion, narrative generation, billing-code suggestion, compliance warnings, prebill analytics, and any predictive billing functions. Identify which tools are approved.
Approval and Accuracy Controls
Set review requirements for decisions, accuracy, and the evidence behind each suggestion.
Review area | What to check or do |
Human Approval | Define decisions that require human confirmation: matter, duration, billing status, narrative, codes, exceptions, and release. The final fee remains the lawyer's responsibility. |
Accuracy | Require field-level testing, pilot validation, error tracking, sampling, correction metrics, and revalidation after material changes. Do not rely on one vendor accuracy percentage. |
Source Evidence | Generated entries should be traceable to relevant activity where practical. Reviewers should be able to understand why the system made a suggestion. |
Confidentiality and Fee Reasonableness
Confidentiality
Document data types, model provider, retention, training use, subprocessors, access, encryption, location, and deletion. Apply professional confidentiality obligations to the AI workflow.
Fee Reasonableness
AI should not be used to inflate time or convert predicted effort into actual billable time. ABA Model Rule 1.5 continues to apply regardless of the technology used.
AI-assisted time entries and reasonable fees
The policy should distinguish a suggestion about work from the decision to bill it. Under the ABA's guidance on generative AI, hourly billing should reflect actual work performed rather than hypothetical time that a manual process might have required. Applicable jurisdictional rules, engagement terms, and client requirements still govern the firm's decisions.
ABA Formal Opinion 512 addresses professional responsibilities when lawyers use generative AI. Model Rule 1.5 provides the reasonable-fee framework. These are ethics sources; task-code lists and LEDES formats serve different purposes.
The firm's responsible lawyer or ethics adviser should interpret applicable obligations and approve policy wording. A software feature or generated narrative does not determine compliance with professional-conduct rules.
Turn the policy into review decisions
Use controlled examples in training so reviewers understand how to apply the rule to a draft. An example should identify the actual work and the decision required, without inventing client results or implying automatic billability.
Draft situation | Required review decision | Record to retain under firm policy |
AI suggests a duration from incomplete activity | Confirm actual work and appropriate duration | Approved duration and relevant correction |
A generated narrative includes sensitive context | Remove unnecessary confidential detail | Approved client-facing description |
AI makes a task faster | Apply the fee arrangement and actual-work requirements | The basis for the final billing treatment |
Client rules restrict an activity | Check the applicable guideline and any authorized exception | Reviewer decision and required approval |
Do not treat client permission, professional duties, and firm policy as interchangeable. The final process must satisfy the applicable requirements together.
Client Requirements
Check engagement terms, outside counsel guidelines, AI restrictions, disclosure requirements, billing rules, and security requirements. Client-specific instructions may override firm defaults.
Explainability and Audit
Preserve source, AI suggestion, user edit, final entry, approval, exceptions, and release status so the firm can review how the record was created.
Change Control
Reassess after model updates, prompt changes, integration changes, new data sources, code-set changes, or client-rule changes. A material change should trigger targeted retesting.
Roles and Ownership
Assign owners for legal/ethics, privacy, security, billing, finance, technology, vendor management, and training. Governance fails when every issue belongs to 'IT.'
Assign governance responsibility to real roles
The timekeeper or authorized reviewer confirms the entry. Billing maintains the approved interpretation of operational client rules. IT/security owns the approved collection and access configuration. The firm's designated legal or ethics owner handles questions of professional responsibility.
When a material source, model, integration, or policy changes, review the affected workflow before relying on prior acceptance results. Keep approved rules and training examples current, and record who can authorize an exception.
Use the confidentiality guide, human-review guide, and outside counsel guidelines article for related controls. This section is general educational guidance and requires firm-specific review before adoption as policy.
Incident Process
Define escalation if AI generates unsupported time, leaks confidential data, applies the wrong client rule, creates repeated coding errors, or produces misleading narratives.
Training
Users should understand AI limitations, review responsibility, confidentiality, billing rules, how to correct suggestions, and when to escalate.
How MIRA Fits
MIRA supports AI-assisted timekeeping with human review. Current materials describe connected activity capture, AI-assisted narratives and task/activity codes, and user controls to edit, merge, exclude, save, and release entries.
Related Resources
Authoritative References
This guide is for educational purposes only. Billing, pricing, accounting, AI, privacy, client, and professional-conduct requirements vary by firm, engagement, and jurisdiction.
Frequently Asked Questions
Should AI be allowed to submit time automatically?
A safer model requires an authorized human to confirm the final entry before release.
What should an AI billing policy cover?
Scope, data, accuracy, human approval, fee reasonableness, client rules, audit, security, and change control.
Should law firms tell clients they use AI for billing?
The answer depends on applicable duties, engagement terms, client requirements, and how AI affects the work or fee.
Can governance eliminate AI billing risk?
No, but it can make responsibilities, controls, and monitoring explicit.
