The best legal invoice review process does not force every decision through the same layer. Use deterministic e-billing rules for objective checks with no contextual exception, contextual AI when the answer depends on written billing or engagement terms and requires interpretation, and human judgment when a finding is material, ambiguous, or sensitive to an outside counsel relationship.
That division of labor creates a more practical operating model. Each layer does the work it is best suited to perform, while the legal team remains responsible for governance and the final business decision with greater accuracy and less manual effort than before.
This framework builds on the discussion in The Right Invoice: Rethinking Legal Invoice Review about assigning business rules, AI, and human judgment to the work each handles best.
| Type of decision | Best review layer | Examples |
| Objective, with no contextual exception | Deterministic rule | Duplicate invoice number, budget threshold |
| Depends on written engagement context / requires interpretation | Contextual AI review | Appropriate level resource, fee arrangements, daily hour threshold |
| Material, ambiguous, or relationship-sensitive | Human judgment | Review/Act on findings, worked performed in alignment with goal |
1. Use deterministic rules for fixed, objective checks
Conventional e-billing rules remain an important first layer of legal invoice review. They are well suited to binary checks that should produce the same answer every time and do not change based on the matter, vendor, or surrounding documents.
A duplicate invoice number is the clearest example. If two invoices carry the same number, the system does not need to interpret a narrative, compare negotiated terms, or weigh an exception. A deterministic rule can identify the condition efficiently and consistently.
The practical test is simple: Could valid context change the answer? If the answer is no, use a business rule. This keeps stable controls stable and avoids adding complexity where interpretation offers no benefit.
2. Use contextual AI when written terms can change the answer
Not every billing decision is binary. General outside counsel guidelines may prohibit a charge, while a matter engagement letter permits it. An approved rate agreement may supersede a standard rate. A negotiated fee arrangement or travel exception may apply only to one engagement.
These decisions require the reviewer to read multiple sources, understand which terms apply, and resolve conflicts between general and matter-specific guidance. That is where contextual AI review can add value. It can evaluate invoice lines against the relevant written context and surface an explanation for review and is also best positioned to interpret a line item where the coding is not necessarily accurate.
The source hierarchy matters. Matter-specific terms should take precedence when they conflict with general billing guidelines. Relevant engagement documents must also be captured in a supported file, uploaded to the matter, and associated with the applicable vendor. Contextual AI should not be described as automatically retrieving terms from a user’s mailbox.
3. Keep people in control of material and sensitive decisions
Finding a potential variance is not the same as deciding what to do about it. A legal team may need to consider the amount at issue, the clarity of the supporting language, the strategic importance of the matter, and the relationship with outside counsel.
Human judgment belongs at this decision point. Reviewers can determine whether to accept a recommended adjustment, request more information, make an exception, or address a broader pattern with the firm. This is especially important for high-value invoices, ambiguous guidance, and sensitive vendor relationships.
Governance should reflect those differences. For example, Spend Agent gives legal teams vendor-level choices for automated action, human confirmation, or no AI review. Those choices are not maturity rankings. They are controls that should align with risk tolerance, billing guidance, and the relationship involved.
Build a layered legal invoice review model
The goal is not to choose business rules, AI, or people as a single answer. It is to route each decision to the right review layer.
Keep deterministic e-billing rules for objective controls. Use contextual AI to interpret applicable written guidance and engagement terms. Reserve human attention for the findings that require materiality, discretion, or relationship judgment.
Spend Agent is designed to complement established e-billing controls and support this layered approach. Deterministic rules handle fixed controls, contextual AI evaluates complex line-item descriptions against applicable written terms, and people make the decisions that require discretion. Together, those layers give legal teams a more governed and practical approach to invoice review.
