The ladder before AI
Where to start with AI in your business (and where not to)
Want to use AI in your business but not sure where to start? Pick a workflow that stalls every week, measure it, and test the four simpler steps first.
To start with AI in your business, pick a workflow that stalls every week, write down how it works today and test the simplest step that solves it first. Sometimes that means configuring a tool you already pay for. Sometimes it means connecting two systems. AI comes in when the task involves unpredictable language and a person reviews the result. Below is how to make that choice without spending money to find out.
Monday, 8:40 a.m. Three WhatsApp messages asking the same thing, a spreadsheet nobody has updated since Friday and a system that doesn’t talk to either of them. You are the bridge between all of it. Then someone in your business owners’ group chat sends a video: “this AI agent does the work of three people.”
You want to buy it. I get it.
The most common mistake: starting with the tool
If you start with the tool, you end up hunting for a problem for it to solve. The request shows up as “I want a chatbot” or “I want an agent.” It almost never shows up as “every Friday we lose two hours checking orders by hand.”
The second sentence is worth far more than the first. It has a day, a task and a time. You can measure it, test it and know whether it got better.
Start with the last real problem
When someone asks me for an agent, my first question is about the past: what was the last request that got lost or delayed? The answer is usually very specific. A quote nobody followed up on. A document that only showed up on the day of the deadline. A client who asked for a status update three times.
That real event shows where information changes hands and gets lost. That is where to look first.
Three questions help you pick the workflow:
- Does it happen every week, with a volume you can count?
- Is someone in charge of it today?
- Can you measure how long it takes right now?
Three “yes” answers and you have a good candidate. If the owner is missing, the first step is to name one. No tool does that for you.
The four steps before AI
At Korai Labs we call this the AI-ization Ladder. For each workflow, we test from the simplest step to the most complex:
- Configure. Make better use of what you already have. A lot of scheduling software already sends automatic confirmations and nobody turned them on.
- Connect. Get two systems talking so nobody has to retype anything. The form submission goes straight into the CRM.
- Automate by rule. A clear rule runs the task. “If the document has been missing for three days, notify the person in charge.”
- AI with human review. A model reads free-form messages, understands the request and drafts the reply. A person reviews the exceptions.
AI sits on the fourth step for a practical reason. It is the piece that costs the most to maintain, needs the most supervision and is the only one that can get things wrong in creative ways. When a lower step solves the problem, the lower step wins.
John Gall, an American pediatrician who wrote about systems in the 1970s, observed that every complex system that works grew out of a simple system that worked. AI projects follow the same law. The ones that succeed usually start small and unglamorous.
How to know if it worked
Before you change anything, write down the baseline: how long the workflow takes, how many times a week it happens, how many errors show up. Without that number, any result is just an opinion.
After the test, compare the same period. Did it improve? You expand. Didn’t improve? You stop. Stopping is a good decision too, and it’s cheap when the test was small.
When it’s not time for AI yet
- When nobody can explain how the workflow works today.
- When the real problem is that nobody owns it.
- When your current tool solves it and just needs to be configured.
- When the task involves a legal, clinical or financial decision with no person reviewing it.
If you want a first map without paying for it, the AI-ization Diagnostic asks 10 questions about your operation and shows what level your business is at, including where AI shouldn’t come in yet.
Which workflow stalled in your business this week?
Frequently asked questions
Do I need an IT team to start using AI?
For the first step, no. Picking the workflow, measuring how it runs today and testing the simplest steps depends on the people who know the operation. Integrations and AI models need technical help, in-house or hired.
What is a good first AI project for a service business?
A repetitive workflow with weekly volume, one owner and a time you can measure. Confirmations, message triage and passing requests between systems are usually good candidates.
Is using ChatGPT the same as adopting AI in my business?
It helps with individual work. Adopting AI in a workflow means connecting the model to your company's data and systems, with review rules and one person in charge.
