The E-Billing System Docks 6% of Every Invoice. Reading All 4,200 Time Entries First Now Costs About $9 a Month.

You tried this in 2024. Your billing coordinator ran a bill-review tool over the pre-bill, it flagged four hundred entries, three hundred and sixty of them were nonsense, and the billing partner went back to skimming. That is the honest history of automated time-entry review at most insurance defense firms, and it is why the pitch below deserves your suspicion.

Here is what actually changed, and it is not that the software got smarter about your practice. It is that reading became close to free. Frontier AI is roughly ten times cheaper to run than it was in 2025, with strong models priced at around $2 for a million words' worth of reading. At that price, having every single time entry read carefully — with its narrative, its task code, its timekeeper, its rate, and the carrier's guidelines all in view — costs your firm less per month than one associate's parking validation. In 2025 that was an idea you priced out and abandoned. In 2026 it is a rounding error on the office coffee.

Where the 6% actually goes

Run your last four months of e-billing through the reduction report. In most defense firms on carrier panels, the gap between billed and approved sits somewhere between 4% and 9%, and it is not one big thing. It is a hundred small ones, and they are the same hundred every month:

• Block billing — "Review file; telephone conference with adjuster; draft correspondence to client; begin outline of MSJ" on one 3.4-hour line.
• Vague verbs the guidelines specifically prohibit. "Attention to." "Work on." "Review file."
• Clerical and administrative time nobody is allowed to bill: organizing exhibits, updating the calendar, scanning, e-filing mechanics.
• The wrong UTBMS task code — a deposition summary coded L120 instead of L330, an entry that belongs in L190 sitting in L110.
• Two timekeepers at the same deposition without the pre-approval the outside counsel guidelines require.
• Travel billed at full rate when the carrier pays 50%.
• A partner rate on a task the guidelines say is associate or paralegal work — the classic $410-an-hour records summary.
• An entry that pushes the matter past the approved budget with no budget revision on file.

Every one of these is knowable before the invoice leaves your building, from the text of the entry and the carrier's own written guidelines — which means the reduction is a self-inflicted wound, not a negotiation. The reason it kept happening is that nobody could afford to read 4,200 entries carefully every month.

What a dollar of checking used to cost

Do the old math. A billing coordinator reading and correcting entries carefully manages maybe 120 an hour once she is looking things up. Four thousand two hundred entries is 35 hours a month — most of a week, every month, on top of her actual job. So firms did what you did: the billing partner skims the fee summary Thursday night, catches the obvious block-billed monsters, sends the rest, and eats the reduction. The write-down becomes a line item everybody has stopped arguing about.

The 2024-era tools failed for a different reason. They matched keywords. "Review" in an entry got flagged whether the entry read "review file" or "review Dr. Patel's IME report and compare to plaintiff's Rule 26 disclosure." A tool with a 90% false-alarm rate does not save the billing partner time; it costs him time, which is why it lasted six weeks.

flowchart TD
  A["Timekeepers close the month in Aderant"] --> B["Draft pre-bill: 4,200 entries"]
  B --> C["Each entry read against that carrier's guidelines"]
  C --> D["Clean entries pass to the LEDES 1998B file"]
  C --> E["Flagged: block billing, clerical, wrong task code, rate mismatch"]
  E --> F["Billing partner approves or rewrites each flag"]
  F --> D
  D --> G["Invoice submitted to Legal Tracker or CounselLink"]

Friday of close week, with the checking turned on

The pre-bill comes out of Aderant, Elite 3E, or Centerbase on the first business day after month end. Before anything reaches the billing partner, every entry gets read once against the specific guidelines for that carrier — not a generic ruleset, the actual PDF the carrier sent you, the one that says paralegal time on document organization is not compensable and that any deposition over four hours requires notice.

The output is not a flag list. It is three columns: the original entry, what is wrong with it, and a rewritten narrative that says the same true thing in compliant form. "Review file re: MSJ" becomes "Analyze plaintiff's responses to Requests for Admission Nos. 4-11 and outline argument for partial summary judgment on the negligent entrustment count." Same work, same time, same truth — described the way the guidelines require.

The billing partner's Friday goes from skimming 4,200 entries to reviewing 190 flags, of which he accepts maybe 150 rewrites as-is, edits 25, and rejects 15 because the machine misread what the associate was doing. That is ninety minutes instead of a lost evening. Then the LEDES 1998B file goes up to Legal Tracker, CounselLink, or Collaborati on the 5th instead of the 12th, which matters on its own, because most carrier guidelines put a 60- or 90-day submission window on the invoice and a late bill is a 100% reduction, not a 6% one.

The arithmetic, with the assumptions on the table

Illustrative firm: 14 timekeepers, insurance defense, $340,000 billed per month across four carrier panels.

On the cost line: 4,200 entries at roughly 35 words each is about 150,000 words of reading, plus the guidelines and the rewritten narratives. At 2026 prices — around $2 for a million words read — that is single-digit dollars a month, which is why the calculation now works and did not in 2025. Even at ten times that cost it would still be immaterial against a $7,854 monthly recovery.

The number to be honest about is the 55%. Some reductions are not fixable by writing better narratives: a carrier that simply refuses to pay above $185 an hour for a partner is not going to be argued out of it by cleaner prose. Pull your own reduction codes from the last six invoices and sort them into fixable and structural before you believe anyone's percentage, including mine.

What this will get wrong, and why the billing partner still signs

The rewrite is the risk, and it is a real one. A model rewriting a time narrative is describing work it did not do, based on a fourteen-word entry. It can subtly overstate. "Review correspondence" becoming "analyze expert disclosure and prepare objections" when the associate actually spent six minutes reading a scheduling email is not a compliance improvement — it is a false statement on a bill, and it is the kind of thing that ends careers and produces fee disputes with a carrier's audit unit.

So the rule is simple and non-negotiable: the model proposes, the timekeeper or the billing partner approves, and nothing goes out unread. Do not let rewrites auto-apply. Keep a record of the original and the approved version — most billing systems will hold both — because if a carrier's auditor ever asks how your narratives got so consistent, you want to show them a supervised process, not a black box.

Two other limits. It will not catch a timekeeper who is simply billing time he did not work; that is a management problem and always was. And it cannot fix a matter that blew its budget in March. If you are getting cut for budget overruns, the answer is a budget revision filed at the time, not better wording in July.

Frequently asked questions

Does this touch our billing system directly?

It does not need to. Export the pre-bill to a spreadsheet, get the review back as a spreadsheet, and have the billing coordinator apply the approved edits the same way she applies partner mark-ups today. Firms that want it tighter can wire it into Aderant or Centerbase later, but the export-and-return version is what you should test with, because you can stop it in an afternoon.

Is sending client matter narratives out of the firm a confidentiality problem?

Time narratives contain client information, so treat it as you would any outside vendor: business-grade account, no training on your data, retention limits, and a line in your engagement terms or your carrier's guidelines review confirming it is permitted. Several carriers now require notice of AI use in the handling of their files. Read your outside counsel guidelines before you start, not after.

Can it tell us which carrier reduces us most, and for what?

Yes, and that is arguably worth more than the recovery. Feed it a year of remittance detail and reduction codes and you get a ranked answer: which panel cuts hardest, which guideline provision costs you the most, which timekeeper's narratives get flagged most often. That report is what you take into the next rate negotiation.

What about appealing the reductions we already took?

Most carriers allow an appeal within a defined window — often 30 to 60 days from remittance. Reading every reduced line, matching it to the guideline provision cited, and drafting an appeal letter is exactly the sort of grinding volume work that used to be uneconomic. Start with the last 60 days, since anything older is likely time-barred by the guidelines.

What to do on the first business day of next month

Take one carrier — the one with the worst reduction rate — and one month of entries for that panel. Run the read, do nothing with the output except compare it against what that carrier actually cut when the invoice came back. If the flags line up with the reductions, you have a business case in one cycle. If they do not, you have learned that your reductions are structural and no amount of narrative work will move them, which is also worth knowing.

One related note for firms whose defense work brings in a steady flow of adjuster calls, insured contacts, and status inquiries: CallSphere builds AI voice and chat agents that answer the firm's phone line and web chat around the clock, take the caller's matter details, and book call-backs on the right attorney's calendar. It has nothing to do with bill review — but the hours a billing partner gets back are worth more if the phone is not eating them.