Legal billing review automation uses AI to check each pre-bill against the client's outside counsel guidelines before the invoice goes out. It matches rates and timekeepers to what the client approved, flags block billing, vague descriptions and wrong UTBMS codes, and routes each issue to a person. The billing partner still decides every change.
Why do law firm invoices get rejected or written down?
Usually because the bill doesn't match what the client agreed to pay for. Each institutional client has its own outside counsel guidelines (OCGs), and a time entry that is fine for one client can break another client's rules.
The gap between work done and money collected is real. In Clio's 2025 Legal Trends Report benchmarks, law firms invoiced 88% of their billable work on average and collected 93% of what they invoiced.1 By our arithmetic, that leaves roughly 82 cents collected per dollar of billable work. Not all of that gap comes from guideline problems, but guideline problems are among the few causes a firm can fix before the invoice goes out.
Line-by-line review isn't going away soon either. The Thomson Reuters Institute's 2026 State of the US Legal Market report found that "90% of legal dollars still flow through hourly billing arrangements."2 Hourly bills are the ones clients and their e-billing tools read entry by entry.
What can AI check before an invoice goes out?
Anything a careful billing coordinator would check against the guidelines, on every entry, for every client. The work is mostly matching, which is where AI is reliable when people review its flags.
| Check | What AI does | Who decides |
|---|---|---|
| Rates | Compares each timekeeper's billed rate to the rate the client approved for this matter | Billing partner, if a rate needs an exception |
| Timekeepers | Flags people the client hasn't approved, or roles the client won't pay for | Billing partner |
| Task and activity codes | Checks that each entry carries the UTBMS code the client requires and that it fits the description | Billing coordinator |
| Block billing | Finds entries that lump several tasks under one time charge and suggests a split | Timekeeper and billing partner |
| Vague descriptions | Flags narratives like "review file" or "attention to matter" that guidelines often reject | Timekeeper rewrites |
| Banned or capped charges | Spots charges the guidelines exclude or cap, such as certain admin work or expenses | Billing partner |
| Staffing and timing rules | Flags too many attendees at one meeting, or time entered later than the client allows | Billing partner |
The hard part is usually the guidelines themselves. They live in PDFs, engagement letters and old emails, and they change. The first step is pulling each client's rules into one place the review can check against, with a person confirming each rule. That matches the most common surprise in our assessments at Soren AI, an AI-native engineering firm: how scattered a team's information is before any automation starts.
What does the billing team still decide?
Every decision about money and client relationships. AI prepares the flags and suggested fixes; people decide what goes on the invoice.
- Whether to write down or write off time, and how much
- How to reword a narrative so it is accurate and still meets the rule
- Whether to ask the client for an exception, such as an extra timekeeper or a rate change
- Whether a flagged entry is actually fine because of something the AI can't see, like a phone call where the client approved the work
- Final sign-off on the pre-bill before it is released
Timekeepers should fix their own descriptions where they can. An AI that silently rewrites narratives would put words in a lawyer's mouth on a document the client relies on.
How does AI billing review work with e-billing systems?
It runs earlier, on the pre-bill in the firm's own billing system, before the invoice is exported and uploaded. Many corporate clients receive invoices in the LEDES format through an e-billing platform, which applies the client's rules on its side and can reject or reduce line items.
Those LEDES invoices often require Uniform Task-Based Management System (UTBMS) codes, a series of codes used to classify legal services. The code sets were developed by bodies including the American Bar Association and the LEDES Oversight Committee, which hosts the central reference for them.3 Task codes describe the type of work performed, activity codes describe what the timekeeper was doing, and expense codes classify expenses.4 Checking those codes on the pre-bill means fewer surprises after upload.
On our page about AI for law firms, we describe the billing agent we build: it works inside Clio the way a person would, checks each pre-bill against the client's guidelines, and waits for the billing partner to approve every change. Time entries carry client confidences, so the same care applies as with any AI tool; our guide to using ChatGPT with client documents covers the ethics questions. This is general information, not legal advice.
How do you know whether billing review automation is working?
Measure the same numbers before and after, by client, for a few billing cycles.
- Rejected or reduced line items in the clients' e-billing systems, and the dollar amount.
- Write-downs taken before billing, split by reason: rate, timekeeper, description, code or banned charge.
- Hours spent on pre-bill review by billing partners and coordinators each cycle.
- Days from month-end to invoice sent, since review is often where bills wait.
- Realization and collection rates for the clients with the strictest guidelines.
At Soren, a billing review workflow starts with the people who review pre-bills today, and a first version is typically live within two to three weeks. See how we work. Billing review pairs naturally with law firm intake automation, because the rates and billing terms set up at intake are what billing review checks later.
When is billing review automation not worth it?
When there are few guidelines to check. A firm that bills mostly flat fees, or whose clients are individuals and small businesses without e-billing requirements, may get more from a clear billing checklist and a careful pre-bill review than from AI.
The case gets stronger with each institutional client that sends its own guidelines and requires e-billing, because that is where the rules multiply faster than any one person can keep in their head.
See where your pre-bills lose money
Start with a 30-minute call with our team. We'll look at how billing review runs today, which systems it touches, and where AI can catch problems before your clients do.
Frequently asked questions
Can AI catch block billing and vague time entry descriptions?
Yes. Finding entries that combine several tasks under one time charge, or narratives too vague for a client's rules, is a pattern-matching job AI does well. It should flag the entry and suggest a split or a clearer description. The timekeeper or billing partner decides on the final wording, since the description goes to the client.
Does billing review automation work with guidelines from many different clients?
Yes, and that is where it helps most. Each client's outside counsel guidelines are turned into a set of rules that a person confirms, and every pre-bill is checked against the right client's set. When a client updates its guidelines, someone updates the rules and confirms them before the next billing cycle.
Will AI billing review replace our billing coordinators?
Usually not. It takes over the line-by-line matching against guidelines, so coordinators spend their time on the flagged entries, client exceptions and conversations with timekeepers. The usual goal is fewer rejected line items and faster invoices with the same team, and a billing partner still approves every change before an invoice is released.
Is it safe to run time entries through an AI tool?
It can be, with the right setup. Time entry narratives often describe privileged work, so the AI should run in systems the firm controls, under terms that keep the data confidential and out of model training. Check the ethics guidance in your state, including ABA Formal Opinion 512 on generative AI tools, and confirm with your ethics counsel.
Sources
- Law Firm Key Performance Indicators: Insights from the Legal Trends Report, Clio, 2025 Legal Trends Report benchmarks
- State of the US Legal Market 2026, Thomson Reuters Institute, January 7, 2026
- UTBMS: A Global Resource on UTBMS, hosted by the LEDES Oversight Committee, checked November 5, 2026
- About UTBMS Codes, Thomson Reuters Legal Tracker help, April 2026
