Where does AI actually help a professional services firm?
AI for professional services firms earns its place in the preparation, not the advice. Gathering, sorting, extracting, drafting, chasing and formatting are where a practice loses hours it cannot bill for. The judgement, the advice and the signature stay exactly where they are, with the person whose name is on the engagement.
That division is not a compromise, it is the design. A firm sells judgement. Everything that surrounds judgement is overhead, and overhead is what a machine is for.
What are the honest use cases?
The honest use cases are all unglamorous ones, which is usually a good sign that somebody has actually done this before.
- Document intake. Pulling the figures and dates out of what a client sends, in whatever format they sent it, and putting them where they belong.
- Client chasing. The outstanding information list that nobody enjoys following up, sent on a schedule, in your firm's voice, with what is still missing.
- First draft preparation. Working papers, file notes, engagement correspondence, standard schedules. Drafted for review, never issued.
- Searching your own history. Finding the three matters where you dealt with this issue before, which is knowledge you already paid for and currently cannot retrieve.
- Deadline and obligation tracking. Watching what is due and telling the responsible person before it is a problem.
- Practice reporting. Utilisation, work in progress, realisation, kept current instead of assembled monthly.
Nothing there is advice. Everything there is time.
What must never be automated in a practice?
Anything you sign, anything you advise, and anything a regulator would expect a named person to have considered.
The output of a professional engagement is a considered opinion attributable to a person with obligations attached to it. A machine can prepare the material that opinion rests on. It cannot form the opinion, and no efficiency argument survives a professional standards complaint. Your professional body's rules are the authority here, and they were written for you specifically. Read them before you build, not after.
Client confidentiality sits alongside that. Putting client material into a third party service is a disclosure, and the ordinary obligations apply on top of your professional duties. The Australian Privacy Principles set the floor.
What about the billable hour?
This is the question nobody asks out loud, and it deserves a straight answer: if you bill by the hour, saving hours reduces revenue.
Pretending otherwise is dishonest, so here is the actual case. Most firms are not short of demand, they are short of capacity, and capacity is people. Time recovered from preparation is capacity redirected to work you could not previously take, or to the advisory work that carries a better rate than compliance.
There is a second effect that matters more over a few years. Firms with a lower cost to serve can price fixed fee work profitably where hourly competitors cannot, and clients increasingly want the fixed fee. The efficiency shows up as a pricing option, not as a smaller invoice.
If neither of those is true for your practice, do not do this. That is a legitimate conclusion.
How does it fit with the systems you already run?
It sits behind them, reading and writing where you already work, rather than becoming another place staff have to check.
Your practice management system stays the record. The machine reads from it and writes back to it. A build that requires your team to work somewhere new will be abandoned in a quarter, no matter how good it is, because a practice under deadline reverts to the path it knows.
What has to be got right is access. Least privilege per matter, an audit trail of what the machine touched, and the ability to show a client or a regulator exactly what happened. The ACSC has practical guidance on the security side, and we go further in AI data security for Australian businesses.
What does a sensible first project look like?
One process, one team, thirty days, with numbers written down before it starts.
Document intake is the usual first choice, because the volume is high, the rules are clear, the output is checked by a person anyway, and the time saved is easy to count. Client chasing is a close second, for the same reasons and because nobody will miss doing it.
Take the baseline first: hours on that task per week, turnaround time, and the proportion of machine output your reviewers changed. Then judge it honestly at thirty days. The full process is in how to run a 30-day AI pilot.
What should you ask a supplier?
Ask how client confidentiality is handled, where the data goes, what the audit trail looks like, and who is accountable when the output is wrong.
A supplier who has worked with regulated practices will answer those without hesitating. One who talks about capability instead of accountability has not thought about your obligations, and your obligations do not transfer to them. How to choose an AI implementation partner covers the rest.
A conversation, not a platform
If you run a practice and the preparation is eating the advice, that is a solvable problem and a narrow one. Start with the process you would happily never do again.
Get in touch and describe it, or read our method for how it gets scoped and priced before anything is built.