The decisions: what work AI does here, what it never does, what client information it may see, who checks its work, and which tools you already use. Adoption is a separate job. It starts with one piece of work nobody has time for, and ideally one that never touches client data.
The question keeps coming up
On Monday morning the office manager reads each inquiry that came in over the weekend. She works out what kind of matter it is, and sends it to the right person. That takes most of her morning, every week.
On Tuesday a client ends a call by asking whether the practice has an AI strategy. On Thursday a vendor emails asking the same thing. The partner who got the email has no idea what to say, so it sits open in a browser tab until Friday.
Look up AI for professional services and every page you find is about the profession in general, plus one list of tools. None of them tells a practice what to put in its own strategy.
A strategy is decisions. Adoption is a job.
A strategy is the set of decisions. What AI is allowed to do in this practice, what it isn't, and who answers for it.
Adoption is how AI actually comes into the work. Which job goes first, who learns it, how you know it's working. That part starts small, and it has a right place to start.
Most articles mix the two, so you end up with a twelve-page policy and no idea what to do in practice. It's much better if you keep them separate.
Five questions to guide your AI strategy
Here are five questions to guide your AI strategy.
What work will AI do in our business? List the tasks and be specific. In a small practice the list usually looks like:
- reading new inquiries and sorting them by matter
- first drafts of routine letters and emails
- summarizing a long document so a person can read the summary first
- meeting notes turned into a to-do list
- research that a person then checks
Each one takes time, follows a pattern, and gets checked by a person afterward.
What work do you want to make sure AI never touches? Most owners can write this down in a minute:
- giving advice to a client, in any form
- having the hard conversation, whether that's bad news, a complaint or a fee dispute
- deciding which clients and matters the practice takes on
- sending anything to a client that a person hasn't read first
- putting its name on work, because a person signs and a person is responsible
Writing that list down does two things. It protects clients, and it tells your staff that AI isn't here to take the part of the job they're good at.
How does AI interact with client information, if it does at all? I find most business owners already have rules for this, even if they're not written down. When you bring in an outside bookkeeper, a contract paralegal or an IT person, there are things you'd share with them and things you wouldn't. Client names, maybe. Medical records or financial statements, only with a signed agreement and a good reason. Use the same rule for a tool. If you wouldn't hand a detail to an outside contractor without an agreement, don't give it to a tool without one either. That covers the chat box and the tool you connect to your inbox, your documents or your calendar, where it can read things on its own.
Most practices sort information into four kinds:
- Public: website copy and published articles. Anything goes.
- Internal: your own templates, procedures and meeting notes. Fine in tools the practice pays for and controls.
- Client: anything that identifies a client or a matter. Only in tools with a written agreement that the vendor won't train on your data or share it.
- Highly sensitive: medical and financial records, anything under a court order or a confidentiality agreement. Usually nowhere until a lawyer has read the vendor's terms.
Who checks AI's work before a client sees it? A specific person, not a full team and not a tool. At the end of the day, as a business owner, you're accountable for what goes on in your business. If the tool gets it wrong, the practice owns the mistake, not the software.
Which tools do you already reuse, and who uses them? Keep a list: the tool, what it's for, who has a login, and what kind of information it's approved for. Three or four lines per tool is plenty. Two rules come with the list:
- Work accounts only, so no client work in anyone's personal account.
- A new tool gets asked for, and somebody checks the vendor's terms before a client's name goes anywhere near it.
Most practices skip the list and find out a year later that six people are using four tools and nobody knows which one holds what.
Put AI on your org chart
Draw the org chart you already have and put AI on it. It shows up in two ways.
The first is a whole job. One piece of work that a tool does from start to finish, with a person it reports to. Take the Monday inquiries. The office manager used to sort them. Now a tool sorts them, and she checks the result. On a chart that's one more box, reporting to her.
The second is people using AI inside their own job. An associate drafting a letter faster. A partner summarizing a long document before reading it properly. That doesn't get its own box. It gets a note next to the person: what they use it for.
One rule covers both. A specific person checks AI's work before it reaches a client, and that person answers for it.
Disclose your AI use
Add one plain sentence to your engagement letter. The practice uses AI tools for drafting, research and administrative work, a professional reviews everything before it reaches you, and your confidential information only goes into tools that meet our security standards.
Some clients will ask for more. A few, like government agencies or clients in regulated industries, will say no to AI on their matters. Write down who said no, so nobody forgets.
Talking about AI at work can be uncomfortable for employees
In most practices, people won't tell you how much they use AI. They worry it makes them look lazy, or replaceable, or like they broke a rule nobody explained. So they use it quietly, on their phones, and your strategy describes a practice that doesn't exist. And they may not want to say how much they're not using it either.
The fix isn't a stricter policy. It's making the rules plain enough that using AI inside them is normal, not a confession. Train everyone, the administrative staff included. Say out loud which uses are fine. Give people somewhere to ask a question without being judged. Then the list of who uses what is honest, because there's nothing to hide.
Start adoption where it expands your business's capacity, not with what people are doing right now
The instinct is to take something a person already does and have AI do it instead. That's the fastest way to make your staff defensive, and it rarely saves what you think, because the person was already doing it fine.
Start somewhere else. Every practice has work it should be doing and isn't, because nobody has the hours and you can't justify a hire for it:
- following up with inquiries that went quiet
- sending a plain-language summary after each meeting
- keeping the website current
- checking in with past clients once a year
- writing the article a partner has meant to write for two years
Nobody's job shrinks. The practice does more than it could before, and the person who owns that new work is glad of the help instead of worried about it.
The number everyone quotes about AI failing comes from projects where the only thing that counted was replacing a cost, and it is worth reading properly before you plan around it.
Once you've picked the job, the thinking goes in this order:
1. What is the job, start to finish? Not "use AI for marketing" but "publish one article a month and send it to past clients".
2. What tools does that job need? A writing tool, the website, the email list. Not every tool is a chat box you type prompts into. Some sit inside your calendar, your inbox or your website and work in the background.
3. What does doing it well take? What a good article looks like, who reads it before it goes out, how often it goes out.
4. How do you do it safely? What the tools can see, and what they can't.
Pick a first job that doesn't involve client information at all. Articles, the website, internal procedures, summaries of your own meeting notes. You learn how the tools behave, your staff get comfortable, and nothing confidential is at risk while you learn. Client-facing work, like sorting the new inquiries, comes once one job is running well and you know what your tools do with what you give them.
Regularly review AI use and your AI strategy
Put a date in the calendar. The first look is a month out, and every three months is about right after that in the first year, because the tools change faster than your strategy will.
At that review, three questions. Is the tool list still accurate? Has anything gone wrong? And is the first job you handed over actually working?
How I do this in my own practice
I run a small practice myself, and I use AI the way I've just described. I have digital employees, which is what I call a tool set up to hold one whole job and report to a person. Some people say AI employee. Same thing. One of mine writes first drafts. One posts on LinkedIn. One manages the ads. Each one does its one job and reports to me. None of them talks to a client, and I read everything before it goes anywhere.
When I work with a practice, I don't start with the tools. I start with the list. What should be getting done around here that isn't, because nobody has the time? That list is where the first digital employee goes.
Then I ask one more question. If you were going to hire a person to do that one job, what would you pay them for a month? That's roughly what a digital employee costs for a year.
If you want to see the shape of working together first, what to expect is written down here.
Book Let's Chat AI for 30 minutes and bring the list. I'll tell you where I'd start and whether this is a fit.