AI cost recovery for small law firms — the modern Westlaw model.

Most small firms using AI today are eating the cost. The Claude or ChatGPT subscription goes on the firm credit card, the bill lands in Software & Subscriptions, and nobody connects the spend to the matters it generated. Every dollar of AI usage on a client's matter is a dollar of profit the firm absorbs.

This is a category error. The legal industry has known how to handle this for forty years.

The Westlaw precedent

When West Publishing introduced Westlaw to firms in the 1970s, billing it back to clients as a soft cost was an immediate and uncontroversial practice. Same with LexisNexis. Same with PACER fees, document copying, courier services, and specialty subscriptions. Firms built engagement-letter language and invoice line items around the principle that research and document costs incurred on a specific matter pass through to the client at actual cost.

The accounting infrastructure is already there. The IOLTA-adjacent trust mechanics, the soft-cost reporting columns in every legal billing system from Clio to PCLaw, the standard engagement-letter clauses — all of it was built around the assumption that some costs follow the matter, not the firm.

AI research, AI-assisted drafting, AI document review, and AI deposition prep are exactly that kind of cost. The tokens consumed answering a research question for the Henderson matter are no more "firm overhead" than the Westlaw search that question would have triggered five years ago.

What ethics opinions actually say

ABA Model Rule 1.5 has long permitted reasonable cost recovery with proper client disclosure. The newer wave of state bar AI-specific opinions — California, Florida, New York, and several others — has been remarkably consistent on the cost-recovery question. The common ground:

• Bill AI at actual cost, not marked up as if it were legal services performed by the attorney.
• Disclose the practice in the engagement letter, including the type of AI tools used and the basis for billing.
• Itemize on the invoice, so clients can see what they're paying for the same way they see Westlaw or printing.
• Maintain confidentiality posture — if the AI tool sends client data to a third party, that's a separate disclosure obligation regardless of billing.

None of this is controversial. The same principles already govern how every other research and document expense gets handled.

The numbers for a small firm

Consider a four-attorney practice running ~20 active matters in a given month. Across legal research, drafting assistance, contract review, and discovery work, AI usage averages roughly $35 per matter per month. The math:

• 20 matters × $35/matter = $700/month
• $700 × 12 months = ~$8,400/year

Today, that $8,400 lives in SG&A as a generic software expense, eroding profit invisibly. With per-user, per-matter attribution, the same $8,400 becomes a soft-cost line on client invoices — flowing back to the firm exactly the way Westlaw soft costs flow back today.

Net P&L impact moves from negative $8,400 to approximately zero. The firm has not made any new money — but it has stopped giving away $8,400 of margin per year. Scale that up linearly for a ten-attorney firm and you're talking about $20,000+ per year that was being quietly subsidized.

What the engagement letter needs

A defensible engagement-letter clause for AI cost recovery is short and unsurprising. It typically:

• Names the practice (the firm uses AI tools to assist with research, drafting, and document review).
• States the billing treatment (AI usage attributable to the client's matter is billed at actual cost as a soft cost, itemized on monthly invoices).
• Notes data handling (which AI providers are used, and what data leaves the firm, if any).
• Offers the client the ability to ask questions before signing.

Plain English. Same length as the Westlaw clause that's already in the document.

The piece that was missing

Until recently, none of this was operational, because there was no clean way to attribute AI tokens to a specific matter at the moment of use. Consumer AI subscriptions don't ship with matter-code tagging. Generic API keys generate one big monthly bill with no client-level breakdown.

That is the piece that has to be solved before any of this works in practice. The accounting model is forty years old. The technical attribution is new.

The accounting pattern is general — see the Cost Recovery overview for the full framework, or Interchange for the platform that makes per-matter AI attribution operational.