How to Make AI Actually Do Work for Your Business
Recorded live at our client mastermind, this episode skips the chatbot tips and asks a harder question: how do you get AI to take a real, repeatable task off someone's plate, and how do you know it did the job right?
The Short Answer
Pick one repetitive computer task that someone in your business already does and nobody enjoys, then hand it to AI the way you'd hand it to a new employee. It needs a clear goal, context about your business, a step-by-step process, the ability to see and click around (eyes and hands), and a way for you to verify the result. Start small and low-stakes, watch what it does, and correct it until it's reliable. Treat it like a very capable toddler: it will sometimes say it's done when it isn't, so keep a person on the final step, like approving a payment.
Start With Real Work, Not Just Chat
Most owners already use AI to draft an email or talk through an idea. That's useful, but this session asked a different question: how do you get AI to do real work in your business, the kind that creates value and that someone is already spending hours on? For trades and in-person businesses, that means computer work for now. AI isn't swinging a hammer anytime soon.
When you're looking for your first task, look in three places.
Work Someone Already Does
A repeatable task a person on your team handles every week, the same way every time.
Work That Should Happen but Doesn't
Follow-ups, reviews, and checks that keep slipping because nobody has time to do them well.
Work You Have to Do and Don't Enjoy
Our favorite category. The tedious admin that sits in the back of your head and drains your energy.
Here's a simple exercise from the episode. Think of one repetitive process you do on your computer that kind of sucks. When you do it, your brain is following a goal, holding some context, and running a set of steps. If you can describe those, a computer can probably repeat them, and more consistently than you can.
Hand It Off Like You Would to a New Employee
For AI to do real work, it needs the same six things a new hire needs. If you've learned to delegate to people, you already know this process.
1. A Goal
What does done look like? Describe the end result, not just the first step.
2. Context
What your business does, how this task fits into the bigger picture, why it matters, and what good looks like.
3. A Process
Do this, then this, then this. The clearer the steps, the more reliably it follows them.
4. Eyes
It has to be able to see what it's doing: the inbox, the folder, the screen.
5. Hands
It has to be able to act: click around, enter text, download and upload files.
6. Verification
You need a way to check that it did the job correctly. This part depends far more on you than on the AI.
At the time of recording, AI had long been decent at goals and context, had gotten much better at following a process start to finish over the past year, and had only become reasonably good at the eyes and hands part in the last few months. It was also still slow at that part. Verification is where most owners need to put in the effort.
A Real Example: Getting Contractor Invoices Off the Plate
One of us had a building project with several contractors sending invoices by email and text, all paid through a bill pay app called Melio. Every invoice meant typing the vendor's details into the app, double-checking them, saving the invoice to Google Drive with a specific naming format so it wouldn't get lost, and uploading it again to the payment app. That's a very computery process: repetitive, easy to get wrong, and nobody's favorite job.
Here's how the handoff went:
- The request was talked out loud, not typed. It described what done looks like: watch for invoices, download them, label them the right way, file them, and get them ready to pay.
- It started from context. The work happened inside a project folder that already held everything about that building project, so the AI didn't start from zero.
- It planned before it built. The AI laid out the design for the whole thing first, and only then got the go-ahead to build it.
What worked: within about a day, an automated check was running roughly every hour, finding invoices in email, filing them, and queuing them up in the payment app. What didn't: it took a lot of prodding. It kept stopping to ask permission or to ask what was next, and the text message side was still giving it trouble. Longer, once-a-year jobs were even harder. A playbook built for tax season worked, but it was still painful.
The AI does the gathering, filing, and data entry. A person still opens the payment app, double-checks the details, and hits pay. Anything that moves money or can't be undone should end with a human review.
Treat AI Like a Superhuman Toddler
"I think the most important perspective that I've adopted is that AI is basically a superhuman toddler."
From the episodeIt's very capable, and it thinks it's smarter than you. It also tells you what you want to hear. It will say the job is done and every check passed, and then you try the thing and the main button doesn't work. So the approach is teach it, watch it, correct it. Test it where it can't do irreparable damage, and repeat until it gets the right result over and over. That's how you build trust.
Where AI helps most
- Amplifying expertise you already have, because you know what good looks like and can spot shortcuts
- Repetitive computer tasks with clear steps
- Brain dumps, when you talk instead of type
- Explaining its own work back to you in plain language
Where to be careful
- Work outside your expertise that you can't check
- Outsourcing your critical thinking about whether something is a good idea
- Outsourcing your creativity, which can quietly drain the joy from work you love
- Outsourcing your understanding of what it actually did
Two prompts help a lot. First, ask it to explain what it just did like you're an idiot (or a fifth grader). Then ask it to go back over how it designed the thing and see if it can make it simpler or clearer, or whether it actually follows the principles it just explained. It often finds its own mistakes. It also tends to over-engineer: too much text on a page, too many ideas on a slide, a spreadsheet full of extra math that misses the core answer.
Keep It Organized So It Stays Useful
The more you use AI, the more scattered it gets. A few habits keep it manageable:
- Use projects. Both ChatGPT and Claude have project folders, so you might keep one for business strategy, one for clients, and one for personal. Be careful what you put in them if you're subject to HIPAA or other privacy rules.
- Build skills and playbooks. Skills are small reusable recipes, like writing emails the way you write instead of the way the AI does. Playbooks are longer chains of steps for bigger jobs, closer to a full SOP.
- Teach by showing. The ChatGPT app has what we called a record and replay plugin, which watches you do a task and learns the clicks. For simple tasks it does a good job.
- Talk to it. Voice dictation (we use an app called Whisperflow) is far faster than typing and great for getting everything out of your head.
- Pin conversations and turn on notifications so you don't lose track when several things are running at once.
Ideas From Owners in the Room
This episode was recorded at a live session with our clients, and some of the best examples came from the room:
- A daily triage. Connect AI to your email, calendar, and messages and have it list what needs to happen first each day, plus messages you sent that never got a reply. Start with very little and add more as you get comfortable.
- A daily numbers summary. One owner gets a morning update on what was collected against goal and whether any new customer calls were missed overnight. (If you want help choosing which numbers to watch, see our guide to weekly financial KPIs.)
- Scheduled team reminders for recurring weekly tasks, so the owner stops being the reminder system.
- Call and meeting review. Tools that record customer conversations can help you coach your team on sales and service, and pull action items out of meetings. Check privacy, consent, and liability rules for your area first.
- A first-pass review. One idea we were about to test: after a team member prepares payroll, have AI run the usual review checks before the final human review.
One caution: a lot of what CRMs do is automation, and that's fine. But AI can go further, into analysis, interpretation, and holding a standard. And if you're not using any of this yet, you're not behind. Plenty of profitable owners aren't. Pick one next step that would make your week easier and try it.
Want More Time for the Work Only You Can Do?
Handing off tedious work is easier when you know where your time and money are actually going. We help owners get clear numbers and a plan, so you can decide what to hand off, what to hire for, and what to stop doing.
Book a Free Intro Call Browse our free workshopsFrequently Asked Questions
Pick one repetitive task done on a computer that someone already does the same way every week and nobody enjoys, like filing invoices or prepping a routine report. It should be low-stakes, so a mistake can be caught and fixed. Skip anything that moves money or can't be undone until you trust the setup, and keep a person on the final approval.
Not at first. AI often reports that a job is finished and every check passed when something is still wrong. Teach it the task, watch it do it, and correct it in a low-stakes setting until it gets the right result repeatedly. Even then, keep a human review on anything important, and only rely on it in areas where you know what good looks like.
Not really. If you've ever delegated to an employee, you already know the process: give it a goal, context about your business, clear steps, access to see and do the work, and a way to check the result. Talking to it out loud, instead of typing, makes it easier to explain what you want. Expect to nudge it along a lot in the beginning.
It depends on the tool, your industry, and what you connect. Start with very little access and add more as you get comfortable. Be careful with customer or patient information, especially if you're subject to HIPAA, and check consent and privacy rules before recording customer calls. Keep a person in charge of anything that sends money or can't be reversed.
No. Plenty of owners who don't use AI run profitable businesses, and some people spend lots of time and money on AI without getting any real value from it. Approach it with curiosity, not guilt. Pick one tedious task you'd love to stop doing, try handing it off, and decide from there whether it's worth going further.