Articles / AI in your business

How to Use AI in Your Business: The Right Way to Start, in 5 Steps

November 17, 20268 min read
In this guide

If you want to know how to use AI in your business, start with a boring process that repeats every week. Most owners do the opposite. They see a new AI tool, think it will solve everything, buy it, and then try to figure out what to use it for. A few months later nothing in the business has really changed.

One of our clients, a law firm, started with the most boring process in the office: background research for every case. That one process now gives them back about three days of paralegal time every week.

I've spent 16 years in business operations, helping over 300 companies, and deciding where AI belongs in a business is one of the things clients ask me for most. Below is the exact process I go through with them, in five steps, with the law firm as the example all the way through.

Want AI to take you through this? I built a free AI Consultant. Paste it into ChatGPT, Claude or any AI, and it goes through all five steps with your own business: it interviews you, scores your processes and writes the first instructions for the AI part.
Get the free AI Consultant

Where AI actually fits in a business

Every company has dozens of processes. When AI works well, it sits inside one of them, in the parts that repeat the same way every time and are easy to check.

At the law firm, that was background research. Their paralegals spent about 40% of their time doing the same research, on the same website, for every case. So we put AI on that one process. Now AI does the research and drafts the report, and a paralegal double-checks it.

3 hours โ†’ 10 minutesresearch time per case at the law firmArman's client, a law firm

No one lost their job. The firm got about three days of paralegal time back every week, and started taking on more clients. Here's how to find the same kind of process in your business.

Step 1: List every process that repeats

Write down every process that happens again and again. Write each one as a task with a start and an end, like "send the weekly client report" or "onboard a new client". A department name like "marketing" is too broad to work with.

In my experience, most of the time goes into the back end of a company: doing the client work, reporting, admin and the other support work. Sales gets the attention, but the hours usually sit behind it.

The easiest way to build the list is to go through one normal week, Monday to Friday:

  1. Open your calendar and note every meeting that comes back every week.
  2. Open your email and look for the same kind of request arriving again and again, and emails you keep writing from scratch.
  3. Open your task tool and look for tasks that keep coming back.
  4. Keep anything you or your team did more than twice. Drop the true one-offs.

If you don't want to do this alone, let AI interview you. Turn on voice input, talk through your week from Monday to Friday, and ask it to turn everything you said into the list.

Try this prompt

Interview me about one normal week in my business, Monday to Friday, one area at a time: winning clients, onboarding, doing the client work, reporting, invoicing, hiring and admin. Ask one question at a time. When we're done, give me a numbered table of every process that repeats, with who does it and where it happens. Keep anything done more than twice a week or for every client.

Step 2: Score the list and pick one

Now you need to choose one process. To do that, answer three questions for each process on your list:

  1. How often does it happen? Per week, or per client.
  2. How long does it take each time? All people included.
  3. What does it cost when it goes wrong, runs late or waits on someone? Low, medium or high.

Multiply how often by how long, and you get how many hours it costs you every week. Then move up anything that costs you money or a client relationship when it goes wrong. A late delivery or a messy onboarding can cost you the whole client. In my practice, a messy onboarding is one of the main reasons a relationship starts badly and a client doesn't stick around.

Process (example)How oftenHow longHours a weekCost when wrong
Weekly client reports6 a week1.5 hours9Medium
Proposals2 a week3 hours6High
Status update emails12 a week15 minutes3Medium
Client onboarding2 a month6 hoursabout 3High
Annoying work is not the same as costly workWe all have tasks we hate doing, but they're not always the ones costing us the most. The most costly processes are often quiet: your team keeps doing them so nothing breaks, and nobody wants to take responsibility for changing them.

Be honest with your numbers. If you use AI to help you build this list, it can't see how much time you spend on anything. Those numbers come from you, so don't undersell or oversell them.

Step 3: Make it work by hand first

Before you hand any work to AI, make sure the process works properly by hand. I use an exercise with my clients for this.

Imagine it's the 1970s. You don't have any technology, you barely have a computer in the office. If someone on your team followed your instructions, could they do this work by hand, over and over, and always get the same good result?

If not, the process isn't complete yet. Usually it's one of three things:

  1. The instructions aren't complete, or the steps aren't in the right order.
  2. People do the same job in different ways.
  3. Nobody owns the final review and approval.

In my experience, almost every company has at least one of these. And if you put AI on top of a process like that, you get the same messy results, just on a bigger scale and faster.

3 in 4 vs 1 in 4companies getting real profit from AI that redesigned how the work gets done, against the restMcKinsey, The State of AI, August 2026, 1,719 respondents

This isn't only my view. In McKinsey's 2026 State of AI survey, nearly three in four of the companies getting real profit from AI had redesigned how their work gets done. Among everyone else, it was about one in four.

You can use AI to write the instructions too. What I like to do is a screen recording while I do the work, talking through what I'm doing the whole time. Then I give the recording or its transcript to my AI and ask it to build a step-by-step process from it. We did this exercise with the paralegals at the law firm, and it also showed us gaps in the process that we fixed as part of the same project.

Try this prompt

Here is the transcript of me doing this task while explaining what I'm doing. Turn it into numbered steps: for each step, what goes in, what happens, what comes out and who does it. Then list the gaps you see: missing steps, places where the right choice isn't clear, and anything with no owner.

Step 4: Split it into steps and decide what AI does

Every process has steps. For the law firm's paralegals they were:

1Open the website
2Search by keywords
3Compare with existing results
4Compare with the client's request
5Write the report

Every step repeated, in the same order. Before you hand a whole process to AI, check each step with two questions:

  1. Does it repeat the same way every time?
  2. If AI makes a mistake, would you spot it quickly?

If both answers are yes, AI can do that step, and a person checks it. Run it, test it, and move on. My own rule on top of that: any step with a high-stakes result keeps a person who approves it before AI moves to the next step.

Why step by step, and not the whole process at once? Because AI is great at some things and bad at others, and if you left a gap in the process, doing it all in one go won't show you where. Testing step by step lets you see every result AI gives you along the way.

Also, AI doing 100% of a process doesn't mean you get the most out of it. I still do a lot of things by hand, because it's faster than explaining to AI what it needs to do.

At the law firm, AI helps so much because the research is repetitive, technical work, and the paralegals can check the report quickly and spot any mistakes.

Start very simpleBefore any automation, start with one chat. I give Claude the purpose, the instructions and the process, and tell it to do the work. I run the same chat for a week or two, correcting it as I go. Once it gets it right, I turn the instructions into a skill.

Step 5: Track it and keep improving

Before you switch the AI part on, write down how long the process takes right now. At the law firm it was three hours per research. Then run the AI part for about a week and compare the same numbers.

In my experience it usually takes two or three weeks to get a new process working well. For a new process, I check:

Week 1Almost every day
Weeks 2 and 3Once a week
After thatEvery few months, or when something changes

Whenever I find an improvement, I add it, check it for a few days and repeat the cycle. This matters even more with AI, because the models get updated all the time. Even if AI handles 100% of a process, it can still get better.

That's why the five steps work as a loop, a flywheel: once one process is running well, you go back to your list from step 2 and start again with the next one.

The five steps at a glance

StepWhat you doWhat you end up with
1. ListGo through one week: calendar, email, task toolEvery process that repeats
2. ScoreHow often ร— how long, plus the cost when it goes wrongOne process to start with
3. By handThe 1970s test, written steps, one ownerA process that works without AI
4. SplitTwo questions per step, a person on high-stakes stepsThe steps AI does and the ones people keep
5. TrackStarting numbers, then daily, weekly, every few monthsA process that keeps improving

Don't try to run all your processes through this at once. Go one process at a time, especially if you're new to using AI in your company. It takes time for you and your team to adjust.

You don't have to do all of this by hand. The free AI Consultant takes your business through the same five steps. Paste it into ChatGPT or Claude and it does the work with you.
Get the free AI Consultant

If you want to turn your instructions into a saved skill, watch Build the Perfect Claude Skills in 2026. If your problem is too many tools, watch I Found $1M in Wasted Software Across 100 Companies.

Two more ways I can help

Not sure which process in your business to start with? In a free call, I'll show you what I'd fix first. Book your free call โ†—

Tired of opening five tools to find one answer? Binee is an AI business brain that plugs on top of the tools you already use, so the answers come to you. Try Binee free โ†—

Get the next guide
in your inbox

Every new guide comes with a video and a free tool. Join the newsletter to get them all.