Tecknocity
All posts
AI implementation 7 min readNov 17, 2026

How to Implement AI in a Business: Why the Order Matters More Than the Tool

Most AI projects stall because they start with a tool. The order we use with clients: find the costly process, fix it by hand, give AI the right steps, track it.

Prefer video? This guide is also available as a walkthrough. Watch it on YouTube.

On this page

When a company asks us to help them with AI, the first question is usually which tool they should buy. We've heard it in companies of every size, and we've worked with more than 300 of them over 16 years. It's the wrong first question, because by the time a tool is chosen, nobody has decided which part of the business it's supposed to change.

So we start from the other end, with one process. Below is the order we use, the belief that usually gets in the way at each step, and what we see across clients.

The order behind every AI project we run

1. Find it

List every process that repeats and score them, so you start with the one that costs the most.

2. Fix it

Make that process work by hand: written steps, one way of doing it, one owner.

3. Split it

Decide step by step what AI does, what it helps with, and what stays with a person.

4. Track it

Measure it before and after, and keep improving it. Then start again with the next process.

Step 1: Find the process that costs the most

"Our problem is sales. That's where AI should go."

In our experience, most of the hours sit in the back end: delivering the client work, reporting, admin. The front end gets the attention, but the time and the mistakes are usually further back.

We build the list by going through one normal week with the team, from their calendars, inboxes and task tools, and keep anything that happens more than twice. Then we score each process on three things: how often it happens, how long it takes each time, and what it costs when it goes wrong. How often times how long gives the hours a week. What it costs when it goes wrong decides what moves up: a late delivery or a messy onboarding can cost a whole client.

Step 2: Make it work by hand before AI touches it

"It's messy now, but AI will clean it up."

In practice it does the opposite: put AI on a process with gaps and you get the same mistakes, faster and more often.

We test every process with one question: if it were the 1970s and there were no software, could someone follow the written instructions and get the same good result every time? Most processes fail at least one of three checks. The instructions are incomplete, people do the same job in different ways, or nobody owns the final review.

3 in 4 vs 1 in 4

companies getting real profit from AI that redesigned their workflows, against the rest

McKinsey, The State of AI, August 2026, 1,719 respondents

McKinsey's 2026 State of AI survey points the same way. Nearly three in four of the companies getting real profit from AI had redesigned how the work gets done, against about one in four of everyone else.

How we fix it

Record it

The person who does the work records their screen while doing it and talks through each step.

Write it

AI turns the recording into numbered steps: what goes in, what happens, what comes out, who does it.

Close the gaps

We mark missing steps, unclear decisions and missing owners, and fix them with the team.

Name the owner

One person reviews and approves the result.

At the law firm, we did this with the paralegals before any AI was involved. It showed us gaps in the research process that we fixed as part of the same project.

Step 3: Give AI only the steps it does well

"If AI can do the whole thing, let it."

AI doing 100% of a process rarely gives the best result. A Harvard and BCG study of 758 consultants, published in Organization Science in 2026, found that on tasks inside what AI does well, people using it did more work, faster, at about 32% higher quality. On tasks outside it, they were 19 percentage points less likely to get the right answer.

So we go step by step and ask two questions. Does the step repeat the same way every time? If AI gets it wrong, would someone spot it quickly? Both yes means AI does it and a person checks. Any step with a high-stakes result keeps a person who approves it before the next step runs.

Open the website

AI

Search by keywords

AI

Compare with existing results

AI

Compare with the client's request

AI

Write the report

AI drafts, a paralegal double-checks

That's the law firm's research process. Research that took about 3 hours per case now takes about 10 minutes. The paralegals used to spend about 40% of their time on it. No one lost their job: the firm got about three days of paralegal time back every week and started taking on more clients.

We start every AI setup as simply as possible: one chat with the purpose, the instructions and the written steps, run for a week or two and corrected as we go. Only once it works do we save it as a reusable skill or build automation around it.

Step 4: Track it and keep improving

"Once it's set up, it's done."

Before the AI part goes live, we write down the starting numbers: time per run, how often it goes wrong, anything else the owner cares about. Then we compare every week. A new process usually needs two or three weeks to work well.

Week 1

A quick check almost every day.

Weeks 2 and 3

A weekly check against the starting numbers.

After that

A check every few months, or when something changes: a new team member, a new client type, a new AI model.

Then

Back to the list from step 1, and the next process.

AI models change constantly, so even a process that runs well can get better. That's why we treat the four steps as a loop that keeps running.

What you can do yourself, and when to get help

You can do this yourself

  • List and score your processes in an afternoon
  • Run the 1970s test on the top one
  • Record yourself doing it and let AI write the steps
  • Run the AI part in one chat for two weeks

When it's worth getting help

  • Nobody can explain the process the same way twice
  • The data for one process sits in five or more tools
  • Several high-stakes steps need approvals and an audit trail
  • You want several processes running at once

If you're new to this, do one process at a time. It takes time for you and your team to get used to working this way. Our free AI Consultant takes you through the same steps for your own business: get it here.

Not sure which process in your business to start with? In a free call, we'll look at your week with you and show you what we'd fix first.

Book your free call

One more thing

🧠 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

From reading to doing

Sounds like your business?

Every post here comes from real client work. Bring your situation to a free 30-minute consultation and we will tell you honestly what it would take, and whether it is worth it.

Book a Free Consultation