Pick one job for AI. Here's how to choose it.

Most businesses pick the wrong first AI project. They choose the one that sounds smartest in a meeting, or the one a vendor suggested, or the one that seems most futuristic. Then six months later it's half-built, nobody uses it, and they've spent money they didn't get back.

The real answer is simpler: pick the job that hurts the most right now.

What "hurts" actually means

Hurt doesn't mean catastrophic. It means something your team does over and over, the same way every time, and it costs you time or money or both.

A dental clinic in Kelowna spends 90 minutes every morning entering patient notes from handwritten forms into their software. That's hurt. A distributor in Calgary has three people who spend two hours a day matching purchase orders to invoices by hand. That's hurt. A plumber in Richmond quotes jobs over the phone and half his quotes never become jobs because customers forget the details. That's hurt.

Hurt is also measurable. You can say "this takes us four hours a week" or "we lose 15% of jobs because of this step" or "this task costs us about $800 a month in labour." If you can't measure it, you can't know if AI actually helped.

The one-job rule

Don't try to automate your entire workflow on the first try. Don't try to automate ten things at once. Pick one.

One job is small enough that if something goes wrong, you can still run your business. One job is simple enough that you'll actually finish building it. One job is real enough that your team will use it instead of going back to the old way.

A 12-person software company in Gastown was drowning in customer support emails. They wanted to build an AI system that would handle support, automate billing, predict churn, and flag upsell opportunities all at once. We told them to start with one: answering the same five questions that came in every day. That one job saved them 6 hours a week. Six months later, they built the next one.

An accounting firm in Burnaby wanted AI to review tax filings, organize deductions, and flag compliance issues. We started with one: extracting key numbers from client tax documents and filing them into their template. That one job took 45 minutes per client down to 10 minutes. Now they know what they want to build next.

How to spot a good first job

A good first job has four things:

It's repetitive. Your team does it the same way every time. There are rules they follow, even if those rules live in someone's head. If the job is different every time, AI will struggle and so will you.

It's measurable. You can count how many times it happens, how long it takes, or how much it costs. "We do this about 20 times a week" or "this costs us roughly $1,200 a month in labour" or "we spend 3 hours on this every Tuesday."

It's not the whole business. If this one job fails, your business keeps running. You don't stake everything on it working perfectly on day one.

Someone on your team already knows it's a problem. Don't pick something that sounds good in theory. Pick something someone actually complains about or mentions in a meeting.

A property manager in Surrey was spending 8 hours a week answering the same tenant questions about lease terms, parking, and maintenance requests. That's a good first job. A dental clinic in Kelowna wanted AI to predict which patients would cancel their appointments. Maybe, but first: do they have three years of cancellation data? Is cancellation actually costing them money right now, or is it just an idea? That's a second-job idea, not a first-job idea.

What to skip

Don't pick a job just because it sounds impressive. Don't pick something because a vendor suggested it. Don't pick something because your competitor might be doing it.

Don't pick a job that requires perfect accuracy on the first day. AI gets better over time, but if the first version breaks something important, your team will stop using it and go back to the old way. Then you've spent money for nothing.

Don't pick a job that only one person understands. If the person who knows how to do it leaves, you're stuck. Pick something your team does together or something you can document.

Don't pick a job that depends on data you don't have yet. "We'll build AI to predict revenue next year" sounds good until you realize you need three years of clean historical data. Start with something you can build with what you have.

The real cost

A good first AI project costs between $8,000 and $25,000 and takes 4 to 8 weeks. It saves between 3 and 8 hours a week in labour, or it prevents mistakes that were costing you money.

Some jobs are simpler and cheaper. Some are more complex. The range is real.

That's not cheap. But it's also not a bet-the-company decision. It's a real test. You'll learn what AI can and can't do for your business. You'll learn what your team actually needs. And if it works, you'll know exactly what to build next.

Where to start

• Spend 30 minutes this week writing down three tasks your team does every week that feel repetitive or costly. Include how long each one takes and how often it happens. Don't overthink it.

• Ask the person who does each task: "If this one thing took half the time, what would you do with the extra hours?" Their answer tells you what actually matters.

• Pick the one task that costs the most time or causes the most frustration. That's your first job. Write down why in one sentence.