How to choose your first AI project

Most businesses pick their first AI project the wrong way. They start with the most exciting idea, or the one a vendor demoed, and then look for a place to use it. A better approach starts with the work your team already does, and picks the parts that are boring, frequent and easy to check.

This guide walks through a simple method that works for a five person shop and for a department inside a large company.

Step 1: List the repetitive work

Spend one week writing down the tasks that repeat. Ask each person on the team to keep a short list. You are looking for work that follows a pattern, even a loose one.

Good candidates usually sound like this:

• Answering the same customer questions by email, phone or message.
• Copying information from one tool into another, such as from Gmail into HubSpot or from Shopify into a spreadsheet.
• Writing the weekly report, the meeting summary or the handoff note.
• Chasing people for missing information, approvals or receipts.
• Sorting, tagging or routing incoming requests.

Do not filter yet. A long, messy list is fine. The scoring step does the sorting.

Step 2: Score each task on four things

Give each task a simple score from 1 to 5 on four questions. Keep it rough. You are comparing tasks with each other, not building a business case.

Hours. How many staff hours does this take in a typical month, across everyone who does it? A task that takes ten minutes but happens forty times a week matters more than one long task each quarter.

Risk. What happens if the output is wrong? A draft internal summary with a small mistake is easy to fix. A wrong price sent to a customer is not. Lower risk scores higher here.

Data access. Does the information the task needs already live in tools you use, like Google Workspace, Microsoft 365, Slack, Salesforce or QuickBooks? If someone has to dig through paper files or ask three colleagues, score it lower.

Approval needs. How many people need to sign off before the output is used? One person checking a draft is ideal. A task that needs legal, finance and a manager to approve each time will move slowly.

Add the four scores. The tasks at the top of your list are your candidates.

Step 3: Pick two or three wins

Choose two or three tasks from the top of the list. Not one, because a single project can stall for reasons that have nothing to do with AI. Not ten, because nobody can watch ten new workflows at once.

A good starting set usually has:

• One task that saves a clear number of hours for one person or role.
• One task that removes a handoff between two tools or two teams.
• Nothing that sends anything to a customer without a person approving it first.

Write down, in one sentence each, what done looks like. For example: “The weekly sales summary is drafted in Notion every Monday morning and the sales lead edits and sends it.”

Step 4: Set a baseline before you change anything

This is the step most teams skip, and it is the one that tells you if the project worked.

For each chosen task, record how it runs today:

• How many times it happens in a week.
• Roughly how long each one takes.
• Who does it.
• How often it goes wrong or needs rework.

Estimates are fine. Ask the people who do the work. Their answer will be more accurate than a guess from a manager.

Step 5: Start in draft mode and measure weekly

Run each new workflow in draft and approval mode first. The AI prepares the work and a person reviews it before it goes anywhere. This keeps risk low while the team learns what the AI gets right and where it needs help.

Each week, compare against your baseline:

• How many times did the workflow run?
• How long did review take compared with doing it by hand?
• How many drafts needed real changes?
• What went wrong, and was it fixed?

Turn this into an estimate of staff hours returned. Keep the estimate conservative. If a task used to take 30 minutes and now takes 10 minutes to review, count 20 minutes, not 30.

After four to six weeks you will know which projects are worth keeping, which need changes, and which to drop. Then go back to your list and pick the next two or three.

How this changes by business size

The method is the same at every size. What changes is the scope.

Local businesses such as wineries, shops and studios usually find their first wins in customer questions, bookings and orders spread across tools like Square, Toast, Shopify, Commerce7 or WineDirect. One owner or manager can run the whole scoring exercise in an afternoon. A single well chosen task can return an estimated few hours a week.

Growing companies of 20 to 500 people often have plenty of AI seats but patchy use. Here the best first projects tend to be reports, handoffs and admin that get copied by hand between tools like HubSpot, Asana, Jira or Slack. Run the listing step team by team, then score across the whole company.

Enterprise teams should start inside one department, not across the whole business. Pick a department with a clear owner and a willing team. Bring security and change management in at the scoring stage, because approval needs often decide which tasks are realistic in the first 90 days.

Where to go next

If you want a quick read on where your business stands, take the free AI score. It takes about three minutes and points to the areas worth scoring first.

You can also see how this looks for local businesses, growing companies and enterprise teams, or browse worked examples on the use cases page.