AI for accounting, legal, and consulting firms
You bill for judgment. The day goes to typing.
The economics of a professional services firm are simple and slightly cruel: you are paid for judgment, and judgment is the smallest part of the day. The rest goes to intake, summarizing, formatting, chasing clients for documents, and writing the same status update for the eighth time.
That gap is where firms quietly lose money, and it is also the single best-understood use of AI available today. Reading a long document and pulling out what matters, drafting a first pass from your own templates, turning scattered notes into a clean narrative — this is the work the technology is genuinely good at, as opposed to the work it is marketed for.
The work that fills the day but doesn't bill
- New client intake that takes a week of back-and-forth before anyone does real work.
- The document grind — reading, summarizing, comparing, extracting — done by the most expensive person available.
- Client status updates written when someone finally gets to them, which is usually after the client asks.
- Proposals started from a blank page when the firm has written eighty of them already.
- Time entries reconstructed on Sunday night from memory and calendar entries.
The number most firms haven't run
Unbilled and under-captured time — the entries reconstructed from memory on Sunday night, the fifteen minutes that never got logged, the write-downs at pre-bill — typically runs 3–8% of a firm's realizable revenue. Not new work. Time already worked, leaking at the point of capture.
A ten-person firm billing $2M that recaptures even 3% recovers $60,000 a year. That's not an efficiency argument; it's found money, and it's the first thing I look for in the audit — because it usually pays for everything else on this page by itself.
What I build for professional firms
- New client intake — forms, conflict checks, and onboarding docs prepared before anyone touches them
- The document grind: summarizing, extracting, comparing, and drafting first passes from your own templates
- Client status updates written on schedule instead of when someone finally gets to it
- Proposals and scopes assembled from your past work, not from a blank page
- Billing prep — time entries cleaned up and narratives drafted before invoicing
- Document collection from clients, chased automatically without anyone nagging
Implementations are fixed-fee from $7,500, live in 2–3 weeks, with an optional $850/month retainer if you want me running it — month to month, no contract.
Start where the evidence is.
Most engagements start here: one week inside how your team actually works, every repetitive workflow mapped, costed, and ranked, delivered as a written plan you own whether or not anything gets built.
AI Operations Audit
$4,500, fixed fee, one week. The audit is priced to stand on its own, and it does not credit toward a build — which means when I tell you what's worth building and what isn't, you know the recommendation isn't selling you anything.
Start with the audit →- 01Where the hours actually goA map of the repetitive work across your team, ranked by hours consumed. It is rarely where people assume.
- 02What each one is costing youEvery workflow costed in staff hours and dollars, plus the revenue quietly leaking out of the ones nobody owns.
- 03What it would take to fixBuild effort, timeline, and price for the top candidates — so you can sequence them instead of guessing.
- 04A written plan you ownYours to keep. Build it with me, hand it to your own team, or sit on it until the budget is there.
The 5 places a firm's week leaks revenue — with the math
The five workflows on this page, each with the dollar math worked out. Built to be forwarded and to survive being printed in black and white.
Why me for this specifically
I run a business built on quoting, dispatching, and billing service work — the same realization-and-utilization economics as a firm, at higher volume and thinner margins. Fifteen years of production systems behind that, across startups and large enterprises. The engagement is the same one I run for associations: an audit that finds where the hours go, then a fixed-fee build that takes the worst of it over — and I do the core engineering myself rather than handing you to an account manager.
What firms ask first.
Won't this reduce our billable hours?
No — it converts non-billable hours into billable capacity. The intake, drafting, and status-update work this automates was never billed at full rate anyway; it was compressed, written down, or done at 9pm. What partners actually see is realization going up, because the hours that remain are the ones clients pay full freight for. If your compensation model rewards raw hours over realization, that's worth a conversation on the call — it's the most common reason firm technology fails, and I'd rather name it before you spend money than after.
What happens to privileged or confidential client material?
It stays in the accounts you already own wherever possible — your document management system, your inbox, your practice management software. I don't train models on your data, I use business-tier providers with no-training terms, and I will sign your NDA. For regulated work we scope the data handling before we scope the build, not after.
Our staff are already pasting client information into free chatbots. Is that a problem?
Usually yes, and it's more common than most firm leadership realizes. Consumer tiers of these tools have different data terms than business tiers, and the exposure is real. Part of what a first engagement does is give people a sanctioned tool that is actually better than the one they were improvising with — which works considerably better than a policy telling them to stop.
How do we keep AI-drafted work from going out wrong?
Nothing I build sends client-facing work without a human approving it. The value is in the first pass, not the final word — a draft on the screen when someone sits down beats a blank page, and it still gets a professional's review before it leaves the building. Any workflow that can't support that checkpoint is one I will tell you not to automate.
Which workflow is eating the most staff time?
Twenty minutes. One workflow. A straight answer.