~£8M
improvement in cash collection
~80%
reduction in client-rejected timesheets
Billing cycle shortened by eliminating rejection loops
First firm-wide visibility into rejection patterns
An AI-assisted pre-submission compliance tool recovered approximately £8M in cash collection for a Magic Circle law firm by catching billing errors before invoices reached clients. The system reduced timesheet rejections by roughly 80 percent, shortened billing cycles, and gave partners their first firm-wide view of rejection patterns by client, matter, and team.
The Problem
Major law firm clients impose detailed billing guidelines dictating how time entries must be written, coded, and structured before they will be approved for payment. These rules cover narrative format, task and activity codes, permitted time increments, staffing restrictions, and which activities can be billed at all. The guidelines vary by client and sometimes by matter, and they change without much warning.
The Solution
We built an AI-assisted pre-submission review system integrated into the firm's time entry workflow. Rather than waiting for the client to reject non-compliant entries, the tool checks each narrative and its associated coding against the relevant client's billing guidelines at the point of capture, before the entry ever reaches a partner for review or an invoice for submission.
The system retrieves the applicable billing rules for the specific client and matter from a configuration-driven rules engine, then analyses the narrative text using NLP to detect common rejection triggers: vague descriptions, prohibited terminology, block billing, missing context, and formatting violations. Simultaneously, a rules engine validates task and activity codes, time rounding, staffing restrictions, and no-charge activity coding. Where it identifies a problem, it suggests a compliant rewrite that preserves the original meaning, with a plain-language explanation of the specific rule violated. Client rule changes are updated in configuration, not code, so the system keeps pace without development cycles.
The fee earner always has final say: every suggestion is presented for review, and the fee earner can accept, modify, or override it. Override decisions are logged, and if an overridden entry is subsequently rejected, that feedback refines the model's confidence calibration. All compliance checks, flags, and decisions are surfaced through dashboards, giving partners and business development teams their first firm-wide view of rejection patterns by client, matter, and team.
Results and Impact
| Metric | Outcome |
|---|---|
| Reduction in rejected timesheets | ~80% fewer rejections post-implementation |
| Estimated improvement in cash collection | ~£8 million through fewer write-downs and faster billing cycles |
| Partner review time | Materially reduced: fewer rejected entries to chase and rewrite |
| Billing cycle duration | Shortened by eliminating rejection-resubmission loops |
| Compliance visibility | First firm-wide view of rejection patterns by client, matter, and team |
| Audit risk | Reduced exposure to client-triggered audits of historical timesheets |
The £8 million figure reflects the combined effect of time entries that would previously have been written off now being corrected and collected, invoices moving through the approval cycle faster, and the volume of rework consuming partner hours dropping substantially.
Key Takeaways
-
Billing compliance is a behaviour problem, not just a technology problem. The solution had to sit inside the time-entry flow, intercepting errors at the moment they are created, because retrospective checks arrive too late to change habits.
-
Suggestions outperform red flags. Telling a fee earner "this narrative will be rejected" without telling them how to fix it creates frustration, not compliance. Providing a compliant rewrite alongside a plain-language explanation drove significantly higher adoption.
-
Rejection analytics revealed problems nobody knew existed. The reporting layer surfaced systemic issues (particular guidelines misunderstood firm-wide, specific clients with unusually complex rules) that could be addressed through training and process changes, multiplying impact beyond what automation alone could deliver.