It depends how often you correct its work and how long it takes to get that work up to your standard. Nobody can give you a number of days. Your time goes into reading what it does and saying what's wrong. It needs the tools the job touches and examples of work you'd put your name to. Correct it often and it gets there sooner.
Say you're a mortgage broker. Monday, before the phones start, you open your inbox and there's a stack of drafts waiting. Your AI employee has written a reminder to every client who still owes you paperwork for their application. Most of them are fine. One asks a retired client for a pay stub, and he hasn't had one in years. You fix it, send the rest and get on with your morning.
That's what the first stretch looks like for a business owner who has just added an AI employee to his team. Under all of it sits one plain question: when can I stop opening those drafts.
How often you correct your AI employee's work decides how long it takes
When a business owner asks me how long, I give the same answer every time. It depends how often you correct its work and how long it takes to get the work up to your standard.
It sounds like I'm being dodgy. I'm not. It's the answer you'd give about a new hire, and for the same reason.
An AI employee does the job the way it was last told to. It won't work out on its own that a retired client has no pay stub. Somebody has to say so, and that somebody is you, because you're the one who knows.
When you correct it, that correction gets written into its instructions, and it works from those the next time. The next stack of drafts comes back closer to how you'd have written them, and the stack after that closer again, as long as someone keeps saying what was wrong. The reading ends when you open the drafts and stop finding things to fix. That's the only end point I can give you.
The training time runs on your corrections. It doesn't matter what the calendar says.
Say one business owner reads the drafts every morning and marks what's wrong. His AI employee gets a correction every working day. Another looks on Friday afternoon, once the week finally goes quiet, so his gets one a week. It's the same job either way, and the second owner will still be fixing the same mistake long after the first has stopped seeing it.
One of my own digital employees, Ora, keeps my business knowledge in one place. My other digital employees go to her for the current truth about the business, so she has to have it right. At the start it was a little painful. I kept thinking, she's not getting this. I trained her over and over on where to find files and how to put them together. She couldn't read certain things. She kept asking for permissions. It was exactly like a new employee. It took me three or four days to get her up to speed, and that wasn't twenty minutes a day with a coffee. That was long days, most of each one spent working closely with her. Every AI employee will have a stretch like that. And a few days of that doesn't mean she's fully out of the gate. There are still a few things I check on periodically, and I expect I always will.
The other half of my answer is the work itself. Some jobs come along fast because the standard is easy to write down: which documents, which clients, which day of the week. Others take longer because the standard lives in your head. How warm to be with a client who's had a hard year. When to chase and when to wait. Which clients you'd phone instead of email.
What a new AI employee needs from you before it starts
Drafts waiting in your inbox are where the setup ends. Before it writes a single reminder to a client who owes you paperwork, it needs a few things from you, and none of them can be skipped.
- Access to the tools the job touches. If it chases paperwork, it has to be able to open the client file and the inbox the reminders go out from. One that can't see your practice management software is guessing.
- Examples you'd put your name to. Reminders you've actually sent and were happy with. Real ones, like the note to the client who always sends things late, or the one to a new client who doesn't know how you work yet.
- The rules you already follow. Most business owners I talk to already have rules for this, even if they're not written down. You never chase a client the week of a funeral. You always copy both spouses on a joint file. Those go in.
- A line between what it sends and what it drafts. On day one, everything it writes waits for you. Later on, the reminders you never touch can go out without you.
- One person who reads its work. That's you. You're accountable for what leaves your practice, so you read what it writes until it comes up to standard. Who answers when it gets something wrong is covered in AI staffing is a hiring decision.
What happens in your office when nobody corrects the AI employee
An AI employee that nobody corrects stays exactly where you left it. Your practice keeps changing around it.
Say the lender asks for one new document from everyone refinancing a house, so you change your list. You tell your office manager. Nobody tells the AI employee. It keeps sending the old list out every week, and you stopped reading those drafts a while ago because they looked fine. A couple send in everything the reminder asked for and think they're done. Then the lender calls your office manager. The file is short one document, the closing is in a few days, and the missing document is the one the reminder never asked for. Your office manager is on the phone apologizing for a list that went out from your practice with your name at the bottom.
That is what "nobody corrects it" costs. The AI employee did the job it was last told to do. The person who deals with the fallout is on your staff.
Where your own time goes while an AI employee comes up to standard
Most of AI employee onboarding is you: reading the work, correcting the work, checking the work. It has drafted the week's reminders to clients who owe you paperwork, and the work is yours to check. The list above is what you give it once, before it starts. This is what you do every working day after that.
- Reading its drafts before they go out.
- Saying what's wrong in a sentence, and what right looks like. "Retired clients don't have pay stubs. Ask for the pension statement."
- Answering it when it stops to ask. A new client with no file yet. A reminder for someone who already sent everything in.
- Telling it when something changes. When you change how you do something, it hears the same day your staff do.
- Deciding, one task at a time, what it can send without you.
You don't have to sit down and write out the whole job before it starts. Your corrections become the instructions, one at a time.
The time this takes is however long the reading takes, every working day, and it runs until you've stopped finding things to fix. It shrinks as the corrections thin out. The stack you read line by line turns into one you skim, and then into one you check for a single kind of client.
You have to put the time in to get the result on the other side. Don't stop too early. If you do, it kicks you later, usually in front of a client.
Your AI employee takes over the job one task at a time
Not one day comes along when it goes from supervised to alone. Sellers who describe their products as autonomous AI agents tend to talk as if it switches on from day one. When you're doing it well, and you want an AI employee to do a whole job from start to finish, it happens in pieces and it takes time. The same way you'd train a regular employee.
Take the reminders to clients who owe you paperwork. The ones to clients who always send things on time go first, because you stopped changing those drafts a while ago. The ones to new clients go next. The refinancing files, where the lender keeps changing what it wants, may need your eye for a long time, because the standard keeps moving. That's how it should work. The tasks where your standard holds still go first, and the tasks where it keeps changing wait.
When you hand a task over, you still own everything that goes out of your practice. What changes is that you read it after it goes, not before.
It goes faster when the first job is narrow. If you haven't picked one yet, Nobody Needs a Morning Briefing: Your First AI Agent is about choosing a first job small enough to get right.
If you want to work out which job to hand over first and who in your practice will read its work, book a Let's Chat AI discovery call. It's 30 minutes.