Where should your company start with AI?

The order we give clients. One paid AI plan and a one-page policy, then one task your team dislikes, then your data. Plus the starting points we talk owners out of.

Start by putting everyone on one paid AI plan with a one-page policy, then automate one task your team already dislikes. Connect your systems after that, once people trust the tools. We tell clients to crawl, walk, then run, and we talk a lot of owners out of rebuilding their whole tech stack as step one.

Crawl with one plan and one page

The first step is setting the rules, and it's less work than it sounds. Pick one AI platform for the company and put everyone on a paid team plan. Open the data settings and check what happens to what your staff type in. Write one page that says what's allowed, what isn't, and what people should expect.

Some of your staff already pay for their own accounts and worry about losing what they've taught them. Julian's fix is a handoff prompt. Each person asks their current assistant to summarize their projects, context, and ongoing threads, then loads that summary into the new shared workspace.

Send a short survey at the same time. Ask how people use AI today, how they'd like to, and what concerns them. You'll learn who your keen users are and where the fear is.

Walk with one task

There are two ways to get the first win, and the right one depends on who needs convincing.

With training first, everyone learns AI on their own work, and the best ideas for automation come from the staff afterward. We go this way when the team is new to AI or the owner is wary of technology. Taylor once took an owner's order spreadsheet, which covered five locations, and had AI consolidate it on the call in two minutes. She said it was work she paid someone to do every month.

With automation first, you take one monotonous task and automate it so a skeptical leader can see a result. For a company whose managers built a reporting roll-up by hand every Friday, we proposed starting there. A food brand came to us with seven workflows in mind. We suggested starting with sample shipping alone, because it was tightly scoped and the return would be easy to see.

On either path, watch the person who does the task before anyone builds. Taylor asks where they move their mouse, how long each step takes, and which software they open. Skip that and you build from a spec, then find out a colleague does the job a little differently.

Keep a person in the loop at the start. We aim for 80 percent automated with a human reviewing the output, and AI drafts while a person sends.

Run once your data is connected

AI is only as good as what it can read. Julian's picture of AI without context is a new hire locked in a closet with the lights off, with instructions slid under the door one page at a time. He also calls unstructured data a library with no catalog.

The run stage is connecting your systems into one data layer that AI can read, then building agents and dashboards on top. A home services client of ours had a decade of quotes, invoices, and job photos spread across five systems. Those now sit in one live dataset, and the quoting work is being built on it.

You don't need to feed it everything. One firm asked how we'd handle 50,000 emails and terabytes of files. Our answer was to start with active projects only.

The three stages

StageWhat you doMove on when
CrawlOne paid plan, checked settings, a one-page policy, training on each person's own workStaff use it weekly without being asked
WalkOne or two tasks automated, with a person reviewing the outputThe first one saves real hours and people trust it
RunSystems connected into one data layer, with agents and dashboards built on topEach new build can reuse data that's already connected

Starting points we talk owners out of

The first is the full overhaul. Owners arrive with a vision of replacing the whole tech stack, and Taylor's answer is that you have to crawl before you can go anywhere. Start with one department that's ready and use its early win to show the others.

Agents are the second. Julian tells buyers that agents are often overhyped, and that the first value usually comes from plain if-this-then-that automation. An agent is a set of instructions tied to your business context. Without the data layer it starts from scratch every time.

The third is tidying up first and adding AI later. One owner told Julian she'd prefer to sort out her systems before thinking about AI. His answer was that a company already replacing its systems is at the right moment to build AI into the new setup.

Give it an owner and champions

Name one person who owns AI at your company. Then pick champions. In a 10-week program at a Canadian restaurant group, we took two or three people from each department and trained them weekly. They rated their own AI skills at 2.5 out of 5 at the start and 4.5 at the end, and they went back to teach their departments.

What waiting costs

We give buyers three costs. Staff increasingly want employers who will train them on AI, and they'll look elsewhere for it. Without an approved tool, people paste company files into free ones. Smaller firms that adopt AI can also outperform larger firms that don't, so six months behind today is a wider gap than six months behind was a year ago.

Your first week

  1. Check the data settings on every AI tool in use. It takes two minutes each.
  2. Choose one platform and move everyone onto a paid team plan.
  3. Write the one-page policy and hand it to every person.
  4. Survey the team on how they use AI and what worries them.
  5. List the five tasks people complain about most and pick one.

If you'd like a second set of eyes on that list, book a free audit and we'll help you choose the first task.

More guides.

Find out what we'd build for you.

The free audit is a 30-minute call. We ask how your team works today and tell you where we'd start.

Book a free audit ↗︎