Some profit leaks are obvious. A project overruns, a client pushes back on scope, or a fee is discounted too heavily. Other leaks are quieter. One common example in consultancies and professional firms is client-rechargeable cost that never gets recharged.
A train fare, search fee, courier charge, software licence, printing cost or specialist report might be recoverable under the client agreement. But if it is submitted as a general expense, tagged to the wrong matter, or left without enough detail, it can disappear into overheads.
Individually, these items may look small. Across a month or quarter, they can add up to money the firm has already earned the right to recover, but quietly writes off through admin gaps.
Example workflow: an expense recharge leakage assistant
This is an example workflow, not a real client case study.
Imagine a professional firm that has a written recharge policy. Some costs should be absorbed by the firm, while others should be passed on to the relevant client or matter. The finance team does a monthly review, but it is manual and easy to miss edge cases.
An AI assistant could review expense data against the firm’s policy and flag likely recharge leakage before the month is closed.
The assistant might use:
- an export from the expense system, such as Pleo, Concur or Xero;
- receipt images and OCR text;
- a client or matter reference list;
- the firm’s recharge policy rules;
- historic examples of correctly coded expenses.
It would not automatically rebill anything. Instead, it would produce a list of candidate expenses that look untagged, mis-tagged or possibly client-related.
What it might flag
A practical report might highlight items such as:
- a receipt that mentions a client name but has been coded to general admin;
- a travel expense submitted after a client meeting but not linked to that client;
- a specialist report fee that matches the recharge policy but has no matter code;
- expenses from one team that are consistently coded differently from similar expenses elsewhere;
- items that mention a matter reference in the receipt notes but not in the accounting category.
The output should be evidence-based. Each line should show the original expense, the relevant receipt or OCR clue, the policy rule that may apply, and the suggested client or matter code if one can be inferred.
The monthly report could include:
- a flagged expense list;
- an estimated monthly leakage figure;
- a draft correction list for finance review;
- a team or individual breakdown to spot training issues.
The aim is not to blame people for coding mistakes. It is to make the leakage visible while it can still be corrected.
The margin-recovery angle
This is one of the more practical uses of AI because the value is relatively easy to understand. If the assistant helps recover legitimate recharge income that would otherwise have been missed, the workflow has a direct commercial purpose.
It can also save finance time. Instead of manually sampling expenses or relying on perfect coding at the point of submission, the assistant does a first pass and narrows the review down to the items that need attention. For many small and mid-sized firms, that might save one to three hours a month.
The bigger benefit may be consistency. If one team interprets the recharge policy differently from another, the monthly report can reveal the pattern. That gives the firm a chance to clarify the rules, update training, or simplify the expense process.
Guardrails: AI flags, humans decide
This workflow sits in a medium-risk area because money is involved. That does not mean it should be avoided. It means the process needs to be designed properly.
The key rule is simple: AI only flags candidates. A finance manager confirms the correct client or matter code before any recharge is applied.
There should also be an audit trail. The report should keep the original expense, the AI’s reason for flagging it, the human decision, and any correction made. That protects the firm if a client queries a charge later.
It is also important not to over-automate. Some costs may be technically rechargeable but commercially unwise to pass on. Others may be covered by a fixed fee or a client-specific agreement. AI will not know all of that context unless the firm provides it, and even then a human should make the final decision.
A practical first step
If this sounds relevant, start with last month’s expenses. Pick a sample of claims, gather the recharge policy, and ask: which items could a sensible reviewer identify as possibly client-related?
You do not need to connect every system on day one. A simple pilot using exports, receipts and a policy document can show whether there is enough leakage to justify a more automated workflow.
If you would like help assessing whether this kind of AI assistant would work in your firm, I can help map the inputs, the policy rules, the approval process and the likely return. Book a short call, or send over the version of this problem you are seeing in your own finance process.
