Reviewing invoices before they go out, with AI help
By the ChamberOne team 4 minute read
Few moments in a law firm are as uncomfortable as a client ringing to query a line on an invoice. The narrative says 'attendance on matter', the rate is not the one agreed, or a week of work is missing and will appear next month. Invoice review is tedious, and it is usually done late, at speed, by the person with the least time. This article looks at how billing partners and finance staff can use AI to review drafts, and which parts of the decision must stay with a person.
Key points
- Most invoice queries come from vague narratives, wrong rates, missing time and mismatched expectations; test drafts against those.
- AI is good at spotting patterns and inconsistencies, and poor at knowing the client relationship or what is fair.
- Treat flags as questions for the fee earner, and reconstruct missing time honestly from real activity.
- Keep invoice approval with a named person who sees the content and the flags, and keep a record of it.
What clients actually query
Most invoice disputes come from a short list of causes. The bill does not match what the client expected. The description of the work is too vague to justify the time. The rate or the fee arrangement is wrong. Time appears that the client did not know about, or work they know about does not appear. Fixing these before the invoice leaves is far cheaper than explaining them afterwards.
Check what your engagement letter and any client billing guidelines say about format and level of detail, and what your regulator or bar says about the contents of a bill. The review should test the invoice against those documents.
What AI can check well
Pattern-spotting across many lines is where software helps most. A tool can read a draft invoice and the time entries behind it and raise questions for a person to answer.
- Vague narratives such as 'work on file', 'attendance' or 'review', with a prompt to add the subject.
- Entries with unusually long durations, or a long day recorded on a matter that was quiet.
- Rates that differ from the rate card or the agreed rate for that fee earner and matter.
- Duplicate or near-duplicate entries across days or people.
- Gaps: periods with correspondence or documents on file but no time recorded.
- Inconsistent descriptions, such as the same task described differently by two fee earners.
- Entries outside the scope of the engagement or phase.
What a tool cannot know
A model does not know the relationship. It does not know that the client agreed informally to a cap, that the partner decided to write off a week's work or that a difficult stage deserves a discount. It cannot judge whether the bill is fair, only whether it is internally consistent.
It can also be wrong about its own findings. A flagged entry may be perfectly proper, and a clean invoice may hide a problem the tool had no data to see. Use it as a second pair of eyes that raises questions, not as an inspector that passes or fails.
- Vague narratives
- Unusually long durations
- Rates that differ from the rate card
- Duplicate or near-duplicate entries
- Gaps with no time recorded
- An informal agreement to a cap
- A partner's decision to write off time
- Whether a difficult stage deserves a discount
- Whether the bill is fair
Catching missing time
Unbilled time is a quiet leak. Fee earners forget to record a call, a short drafting job or travel. A review can compare activity on the file, such as documents created, emails sent and calendar entries, with the time recorded, and highlight days with activity and no time. The fee earner then decides whether time was missed.
Be careful about the direction of the nudge. The aim is accuracy, not inflation. Reconstructing time from activity should be done honestly, and any added entry should reflect work actually done. Check the firm's own policy on late entries.
Better narratives
A tool can suggest clearer descriptions from the entry and the documents involved, for example turning 'review documents' into 'review the lease and note break provisions'. The suggestion should be checked by the fee earner who did the work, since only they know what was done. Drafting narratives after the fact from document names alone invites inaccuracy.
The best fix is upstream: encourage fee earners to write a clear narrative when they record the time, perhaps with a short template for common tasks. Review then becomes a light check rather than a rewrite.
A review workflow
A workable sequence, from draft to sent, looks like this.
- 1
Generate the draft invoice from recorded time and agreed arrangements.
- 2
Run the automated checks and list the questions raised.
- 3
The fee earner answers each question and corrects the entries concerned.
- 4
The billing partner reviews the invoice as a whole, against the engagement letter and the client's expectations, and decides on any write-down or write-off.
- 5
Finance checks tax treatment, currency, client details and payment instructions.
- 6
The invoice is approved by a named person and then sent.
- 7
Questions from the client are logged and used to improve the checks.
- 1Generate the draft invoice
- 2Run the automated checks
- 3Fee earner answers and corrects
- 4Billing partner reviews as a whole
- 5A named person approves and sends
- 6Client questions improve the checks
- Then back to step 1.
Keep approval human
An invoice is a statement to the client and, in some systems, a document of regulatory significance. Someone accountable should approve it, having seen the content. An 'approve all' button that skips the reading defeats the exercise. Make sure the approver sees both the invoice and the flags that were raised and how each was resolved.
Keep a record of who reviewed and approved, and when. If a client later disputes a bill, that record is the first thing you will want.
If the firm bills on fixed fees or milestones, include a check that the invoice matches the agreed stage and amount, since an automated draft can trigger a charge before the milestone is actually reached.
Confidentiality of the billing data
Time narratives often reveal the substance of the work: names, strategy, even the existence of a transaction. Before sending them to an outside tool, check what is shared and how it is handled. Consider whether narratives can be reviewed within the firm's own system or with identifying details removed. Follow your firm's policy and any client instruction on confidentiality.
Start small. Run the checks on a sample of invoices for a month, count how many flags were useful and how many were noise, and tune the rules. Share the findings with fee earners as practical help and not as surveillance, since the tool will work only as well as the habits of the people recording time.
Billing in ChamberOne
- Records time with a timer in the top bar, or with a finder for work you did but did not log.
- Applies rate cards, fixed fees, per-appearance fees or milestones, and puts the invoice through approval.
- Sends statements and payment reminders.
- Records every read and write in the audit trail.
What it does not do
- It does not decide for you whether an invoice is fair; a person approves each invoice.
- The client-money ledger has not been assessed against any professional accounts rules and does not by itself make a firm compliant with them.
- Prices shown on the ChamberOne site exclude tax.
This article is general information about the law, legal practice and technology, current when it was published. It is not legal advice. Laws and guidance change: check the sources linked above and take advice on your own circumstances from a lawyer qualified where you are.
Read next
AI and the billable hour: what actually changes and what does not
How faster drafting puts pressure on hourly billing, and which fee models suit which work.
5 minute readFixed-fee pricing with AI-assisted scoping
How to use past matter data and AI summaries to scope a fixed fee, define exclusions and set a review point.
5 minute readRecovering unrecorded time: how AI time capture works and its limits
How suggestions from calendar, email and documents help fill timesheet gaps, and why the fee earner must still confirm each entry.
5 minute read