Written by: Agata Wojtas, Chief Commercial Officer, Digital Colliers
Your associates are using AI. That's not the question anymore. The question your managing partner should be asking is whether the six-minute increments on this month's invoices line up with what the tools actually did. Because as of 2024, that reconciliation is an ethics obligation, not a nice-to-have.
What Opinion 512 actually says
ABA Formal Opinion 512, issued in 2024, is short and awkward for anyone billing by the hour. Lawyers can't charge clients for time that AI actually saved. If a partner used to spend three hours on a memo and Copilot cuts it to forty minutes, the bill reflects forty minutes plus honest review time. Not three hours. Not two hours with a shrug.
The UK side of this isn't quieter. The SRA issued its AI guidance back in November 2023, and it points in the same direction: competence, confidentiality, and honest billing when tools are in the loop. Regulators on both sides of the Atlantic have decided that the burden of proof sits with the firm.
And the deadline isn't next year. It's the invoice you're about to send.
The reconciliation nobody has run
Here's the mechanical problem. You have two data sources that have never talked to each other:
- Time entries in your practice management system, keyed to matter numbers and narratives.
- Tool logs from Copilot, Harvey, ChatGPT Enterprise, CoCounsel, Lexis+ AI, and whatever else associates opened this week.
If an associate billed 2.4 hours to matter 41221 on Tuesday afternoon, can you show what AI activity happened on that user's account during that window? Most firms cannot. The time entry lives in Aderant or Elite or Clio. The AI logs live in a Microsoft admin console, or a vendor dashboard, or nowhere the finance team can reach. Nobody has built the join.
That's the audit exposure. Not that associates are using AI. That you can't reconstruct what they did with it.
Why this keeps getting punted
A few reasons, all boring, all real.
- The logs are inconsistent. Copilot tells you a prompt was submitted; it doesn't tell you which matter it belonged to. That mapping has to be inferred, usually from timing or from a matter tag the associate hopefully entered.
- Retention windows differ. Some vendors keep prompt history for 30 days, some for a year, some depend on your tier. If you get a client audit request in month four, half the evidence may be gone.
- The people who understand billing don't understand tool logs, and the people who understand tool logs don't touch billing. There's no single person who owns the reconciliation.
- It feels like surveillance. Partners don't love the idea of pulling associate prompt history, and associates like it less. So it doesn't get built.
Meanwhile the risk keeps rising. Court cases involving AI-fabricated citations jumped from 87 to over 1,300 in eleven months in 2024. Every one of those is a discoverable event where somebody's tool use is now on record. If a client's litigation team can subpoena that pattern, your own compliance team should be able to see it first.
Building the join
The firms doing this well treat it as a data problem, not an ethics problem. The ethics answer follows once the data is in one place. What that looks like in practice:
- Pull tool logs into a warehouse on a nightly job. All vendors, one schema, one timestamp column.
- Pull time entries with matter numbers, narratives, and timekeeper IDs.
- Join on timekeeper plus time window. It won't be perfect. It doesn't have to be. You need to be able to answer the question "show me AI activity on this timekeeper during this billed block."
- Sample. Pick five entries a week per practice group. Have a partner review the pairing. Adjust narratives where the AI clearly compressed the work.
That's it. It's not glamorous. It's a warehouse, two connectors, a join, and a review cadence.
The framing that gets budget
The reason firms don't fund this is that it gets pitched as an ethics project, which reads as cost. It's better framed as a realisation project. Clio's Legal Trends Report has long put average billable realisation around three hours out of every eight. AI use, honestly reflected, changes what those three hours look like and what clients will pay for them. Firms that can show clients a defensible narrative about AI-assisted work tend to hold rates. Firms that can't tend to get the rate conversation forced on them later, with less leverage.
Opinion 512 didn't create a new obligation so much as it made the old one auditable. The winning move is to make it auditable on your terms, before a client does it for you.