Two hosts work the reservation phone at a busy neighborhood restaurant we'll invent for this example. One writes each booking in the paper book first and enters it into the reservation system later; the other books straight into the system but never asks about high chairs. The owner wants an AI assistant to take calls during the rush. The assistant would have to copy one of them, and nobody has decided which.
That's what readiness really comes down to. It's less about your technology than about whether the work is settled enough to hand to someone else. If you could hand a task to a new hire with a one-page sheet, you can probably hand it to software. Here's how to check.
Seven signs you're ready
Each sign comes with a check you can run this week. Score signs 1 to 4 for one task only, the one you'd most like to hand off; signs 5 to 7 are about the business as a whole. Give yourself a point for each sign that holds.
1. The same task shows up many times a week
Check: open your call log, inbox or job list and count one task over a week. Double digits means there's volume worth automating.
2. You could explain the rules on one page
Check: write the steps the way you'd brief a temp. If it fits on a page with a short list of exceptions, it's rule-based enough. If every other line reads “it depends,” it isn't yet.
3. The information is already digital
Check: list what the task needs (names, dates, prices, availability) and where each piece lives. Software, email and shared drives count. A paper book and a manager's memory don't.
4. One person owns the process
Check: ask “who decides how this is done?” If one name comes back quickly, good. Three names or a shrug is the first thing to fix.
5. Logins and access are in order
Check: does each person have their own login to the systems this work touches, and could you say who can open client or customer files? Shared passwords and a folder everyone can see become a bigger problem once software can read that folder too. Tax and accounting practices covered by the FTC Safeguards Rule should already have a written information security plan, and any AI tool belongs in it.
6. Your core software is current and can connect
Check: list the two or three systems your daily work depends on, such as the reservation system, practice-management software or calendar. Is each on a version the vendor still supports, and does it offer an integration, an API or at least a clean export? One email to each vendor's support line will tell you. Software with no way in means someone still retypes at the end, which limits what automation can do.
7. Your team wants the help, and someone has time to supervise
Check: ask whoever handles the work day to day what they'd hand off tomorrow. If they name it before you do, adoption will be easy; if they're worried about their jobs, talk about that openly first. Then find someone with an hour or two a week to review the system's work for the first month, the way you'd check a new employee's.
Three signs you aren't ready yet
None of these is a failure. Each is groundwork, and each has a clear first step.
The process changes every week
If the rules are still being worked out, automating them freezes the confusion in place. The restaurant with two hosting styles is here. First step: agree on one way to take a booking, write it down, and have both hosts follow it for two weeks. Then automate the version that worked.
The records live on paper
Picture a small law firm where intake happens on a printed form that sits in a tray until a paralegal has time to type it up. No AI system can read the tray. First step: don't scan years of history. Start capturing new intakes digitally, with a web form or a fillable PDF, and let the archive wait.
You're hoping AI will fix a people problem
If calls go unanswered because someone ignores the phone, or a partner refuses to record time, a new system will be ignored the same way. First step: settle the expectation with the person directly. Automate once the process works when people actually follow it.
Flowchart
A quick readiness check for one task
- StartPick the task you'd most like to hand off
- DecisionDo the steps fit on one page that everyone already follows?NoA personOwner agrees one way of doing it; staff follow it for two weeksYes
- DecisionIs everything the task needs already in software or email?NoStepStart capturing new records digitally; leave the archive for laterYes
- DecisionIs one named person responsible for how it's done?NoA personOwner names someone and gives them authority to change itYes
- DecisionIf it worked, would the people doing the task actually use it?NoA personOwner settles the expectation with the person before automating anythingYes
- StepCount how often it happens and what it costs today
- DoneReady: scope a small, supervised first project
Before and after
Automating a messy process vs. a settled one
Automated as it was
- 01Assistant copies one host's habits and misses the other's
- 02Bookings go into the system; the paper book stays in use
- 03Large-party rules depend on who answers
- 04Staff overrule the assistant and stop trusting it
ResultTwo booking records and a tool nobody trusts
Settled first, then automated
- 01One written booking routine, tested for two weeks
- 02Paper book retired; the system is the single record
- 03Large parties and allergy notes go to the manager by rule
- 04Assistant follows the same routine as the hosts
ResultOne record, and an assistant that books like the team
What your score means
| Your result | What it means | Do this next |
|---|---|---|
| 6 or 7 signs, no warning signs | Ready for a first project | Pick the highest-volume task and scope it tightly |
| 4 or 5 signs, no warning signs | Ready for something narrow | Choose a task where mistakes are cheap, such as reminders |
| 3 or fewer, or any warning sign | Groundwork first | Fix the first warning sign, then recheck in a month |
Once you're ready, here's how to pick your first project from several candidates. If you'd like an outside read on where you stand, our opportunity reviews open with these same questions and say plainly when the answer is “not yet.”
A four-week plan for a “not yet”
If your score says groundwork first, here's a month that gets most businesses to a clear yes or no. It costs nothing but a little attention, and every step helps even if you never buy anything.
- 01Week 1: write it down. The owner of the task writes the steps on one page, including the five most common exceptions and what happens with each.
- 02Week 2: follow it. Everyone who does the task follows that page exactly. Each time it doesn't fit, they note why in one line.
- 03Week 3: count it. Keep a tally of how often the task happens, how long it takes and what goes wrong. For the restaurant, that's calls during service, missed calls and booking mistakes.
- 04Week 4: decide. Revise the page with what weeks 2 and 3 taught you, then rerun the seven signs. If the score has moved to five or more, you have a task ready to automate and the numbers to judge it by.
A note for law firms
For a law firm, readiness includes understanding a tool well enough to supervise it. The American Bar Association's Formal Opinion 512, published in July 2024, expects a lawyer using generative AI to have a reasonable understanding of what the tool can and can't do, and confirms that the usual duties around confidentiality and supervision still apply. State bars have added guidance of their own.
In practice, a firm is ready when someone can explain what the system does with client information and who reviews what it produces. One concrete check: find out whether a tool is self-learning, meaning it may use what you type to shape answers for other users. Opinion 512 says a client's informed consent is required before information about their representation goes into such a tool, and boilerplate in an engagement letter doesn't count.
That's a useful bar for any business, not just law firms. Treat this as general information rather than legal advice; state rules differ. For how firms typically start, see our page on AI for law firms.