Treadstone Associates
Article · Practice metrics & pricing

How to model a fee increase with AI

A model will build the scenario table in minutes from data you already hold. What it cannot supply is the one input that matters — what your clients will actually accept — and it has no view at all on the notice your engagement letter requires.

Treadstone Associates · Updated 2026

Key takeaways

  • • AI is a scenario calculator over your own billing history. It has no information about your market that you did not give it, so never let it generate the acceptance assumption.
  • • Whatever you land on has to be disclosed in time. For an Ontario lawyer, rule 3.6-1 of the Law Society of Ontario’s Rules of Professional Conduct requires a fee to be fair and reasonable and disclosed in a timely fashion.
  • • If you add late-payment interest, section 4 of the Interest Act caps recovery at five per cent per annum where a rate is expressed for a period shorter than a year without an express statement of the equivalent yearly rate.
  • • Tax follows the fee. subsection 165(1) of the Excise Tax Act sets the federal rate at 5%; the provincial component is a prescribed rate, so check the current one for the province where the supply is made rather than quoting a number from memory.

The short answer

Model a fee increase by running your own realisation, write-off and WIP data through a scenario table: current effective rate by client, proposed rate, and the revenue outcome at several attrition assumptions that you choose. A language model is good at building and explaining that table and terrible at populating the attrition column, because it has no knowledge of your clients. Then check three things that are not arithmetic at all — what your engagement letter says about changing fees, what notice you owe, and whether your regulator imposes a disclosure standard on top.

The reason firms get this wrong is that the modelling feels like the hard part. It is not. The hard part is that a rate change is a variation to a live contract with every client on the book at once.

What the model is actually computing

Three quantities do almost all the work, and all three come out of your practice management system rather than out of a model:

  • Effective rate, not standard rate. Standard rate times hours is a fiction in most firms. Effective rate is what you actually collected divided by the hours actually recorded, per client and per service line. A five per cent increase on the standard rate can be worth nothing if it is absorbed by write-offs.
  • Realisation by client, not in aggregate. The firm-wide number hides the two or three relationships that are subsidising everything else. Ranking clients by realisation is the single most useful output of the exercise, and it is a sort, not a model.
  • Unbilled work in progress at the changeover. Work performed before the effective date is billed at the old rate unless your engagement letter says otherwise. Firms routinely forget this and then either lose the margin or invoice something they cannot support.

What AI adds is speed and legibility. It will reconcile exports from a time system and a billing system, flag clients whose recorded hours and invoiced amounts diverge, generate the sensitivity table, and write the plain-language explanation the partners will actually read. Every figure in that table traces to a row you can point at. None of it is a judgment.

The column you must not let a model fill

Attrition. A model asked how many clients will leave at a nine per cent increase will produce a confident number, and that number is invented. There is no source for it. Fill it yourself with a range — a low, a central and a pessimistic case — and read the revenue outcome at each. If the increase only works in the optimistic case, it is not an increase, it is a hope.

The same applies to any benchmark the model offers about “typical” professional-firm rate rises. If you want a real comparison, use published industry data: Innovation, Science and Economic Development Canada’s Financial Performance Data tool reports industry averages and financial ratios drawn from tax and survey data, with quality indicators showing how much imputation sits behind each figure. That is a source. A model’s recollection is not.

What the increase touches contractually

A fee change is a change to the terms you are engaged on. Whether you can make it unilaterally, and on what notice, is a question about the document you signed. Treadstone Law’s note on key clauses in an Ontario service agreement covers where fee, term and variation clauses usually sit and what a bare “rates may change from time to time” line does and does not achieve.

If you bill hourly and are considering moving part of the book to fixed fees at the same time, treat those as two separate decisions. Treadstone Law’s comparison of hourly rates and flat fees sets out the trade-offs; conflating a pricing-model change with a price increase is how a straightforward letter turns into a renegotiation.

Interest and tax follow the fee

Firms often bolt a late-payment charge onto the same letter. Two constraints apply. First, section 4 of the Interest Act: where a written contract makes interest payable at a rate for any period shorter than a year, no more than five per cent per annum is chargeable or recoverable unless the contract expressly states the equivalent yearly rate. “1.5% per month” on its own is therefore worth five per cent a year. Treadstone Law’s piece on late payment interest clauses on Ontario invoices explains how that plays out.

Second, if you are a licensee, your regulator may restrict interest independently — rule 3.6-1.1 in the Law Society of Ontario’s rules prohibits a lawyer charging a client interest on an overdue account except as permitted by the Solicitors Act or otherwise by law.

On tax, subsection 165(1) of the Excise Tax Act fixes the federal rate at five per cent of the value of the consideration. The provincial component in a participating province is a prescribed rate, so state it from the current published rate for the province in question rather than from memory or from a model.

The regulator sits above the model

Name yours and read its fee rule before you send anything. For Ontario lawyers, rule 3.6-1 states that a lawyer shall not charge or accept any amount for a fee or disbursement unless it is fair and reasonable and has been disclosed in a timely fashion; the commentary lists the factors that bear on fairness, including the time and effort required, the difficulty and importance of the matter, whether special skill was required, the amount involved and the results obtained. A professional engineer setting expectations on price is engaging the same instinct the Engineers Canada guideline on the code of ethics expresses in Principle 2 — care and clear communication in accepting and interpreting assignments and in setting expected outcomes. CPAs work to their provincial body’s code. The common thread is that a fee is a professional communication, not a marketing message.

Worked example (illustrative)

A six-partner accounting practice in Ontario has not moved its rates in three years and suspects it is losing money on compliance work for a handful of long-standing clients.

The exercise takes two evenings. An export of recorded hours and issued invoices for the last two fiscal years is reconciled by a model into one table: client, service line, hours recorded, amount invoiced, amount collected, effective rate. Twelve clients fall below the firm-wide effective rate by a wide margin, and nine of those twelve are on the same legacy fixed fee agreed before a change in scope. That is the finding. It did not require a model to be clever, only to be fast.

The partners then set the attrition assumptions themselves and read three revenue outcomes. They decide on a general increase from the next engagement anniversary, individual conversations with the nine legacy clients, and no change at all to two relationships where the write-off is a deliberate investment. The model drafts the client letter; a partner rewrites the second paragraph, because a model’s instinct is to apologise for the increase and the firm’s instinct is not to. The unbilled WIP at the changeover date is billed at the old rate, deliberately, and the letter says so.

Measure the countable things afterwards: realisation by client before and after, the number of clients who asked for a call, the number who left, and the number of invoices queried. Not “hours saved” — that is unverifiable and makes a poor business case.

Where this sits in the firm

This page is written for a firm that delivers work to a book of clients. If the question is really about the front desk — intake, scheduling, recall and reminders — that lives on the professional practice owners page. If it is about your own month-end, reconciliation and payables rather than what you deliver to clients, that is accounting automation, or bookkeeping automation for coding and receipt capture. The tooling overlaps; the risk almost never does.

Questions we get asked

Can AI tell me what other firms in my city charge?
No, and a model that answers is guessing. Published Canadian data works at industry level, not at firm level — Financial Performance Data is the practical source for industry averages and ratios by industry and revenue band. Rate cards are not public.

Do I have to give notice before increasing fees?
It depends on your engagement terms and, if you are licensed, on your regulator’s disclosure standard. The Ontario position for lawyers is disclosure in a timely fashion under rule 3.6-1; the contractual position is whatever your service agreement says. Treadstone Law’s service agreement note is the place to start, and a lawyer should read the actual clause.

Should the increase letter be AI-drafted?
Drafted, yes; sent unread, no. A generated letter tends to over-explain and under-commit. Keep the structure, cut the justification to two sentences, state the effective date and how work in progress is treated, and have a partner sign it.

Find out what your book is actually earning.

A 30-minute call is enough to tell you whether AI pays for itself here.