Start with the smallest AI product that proves your idea

The graveyard of AI projects is full of ambitious first versions. A property manager in Surrey spent three months building a tool that would handle tenant screening, lease analysis, and maintenance scheduling all at once. Six months in, it did none of them well. A dental clinic in Kelowna tried to automate their entire patient intake and follow-up flow in one go. After two months and forty thousand dollars, they had something that confused staff and annoyed patients.

The pattern is always the same: too many problems at once, too many edge cases, too much time before anyone outside the building sees it.

The right approach is the opposite. Pick one small, specific problem that a real customer faces every week. Something that costs them time or money. Build a tool that solves only that problem, get it working with a real customer, measure whether it actually helps, then decide what to build next.

Why narrow is faster

A narrow problem is easier to solve because it has fewer moving parts. A plumber in Richmond spends two hours a week writing follow-up text messages to customers after jobs are done. That's one problem. A tool that reads the job details from his system and writes the message saves him 90 minutes a week. It's measurable. It's real. He can tell you whether it works.

If you try to build a tool that also schedules callbacks, flags upsells, updates the CRM, and generates invoices, you have five different problems with five different success criteria. One of them will be hard. You'll spend weeks on it. The whole project stalls.

Narrow also means you can build and test faster. A tool that does one thing can be live with a real customer in two to three weeks. You'll know within a month whether the idea is worth expanding. A broad tool takes two to four months before you even know if the core idea works.

What to measure

Before you build anything, define what success looks like in numbers, not feelings.

For the plumber: 90 minutes saved per week, measured by tracking how long messages take before and after.

For an accounting firm in Burnaby that wants to automate invoice filing: 15 to 20 invoices filed per hour by the tool, versus 8 to 10 by hand. Time saved per week. Error rate before and after.

For a distributor in Calgary whose sales team spends time pulling order history for each customer call: time from "customer calls" to "information ready", measured in minutes. Before: 8 minutes. After: 2 minutes.

Pick a metric that matters to your customer's wallet or calendar. Hours saved per week. Invoices processed per day. Calls answered faster. Errors caught before they cost money. Then measure it for real, with a real customer, doing real work.

Most AI projects fail because people guess at the value instead of measuring it. They assume a tool will save time and never check. Three months later, they've built something no one uses.

How to keep ownership

This matters because you'll want to expand the tool later, and you can't do that if someone else owns the code.

There are three ways to build:

Template platforms. You plug in your data, check some boxes, and get a tool. Fast to start. You don't own the result. If the platform changes pricing or shuts down, you're stuck. If you want to customize it later, you can't.

Off-the-shelf software. You buy a tool that thousands of other businesses use. It's standard, not custom. If it doesn't fit your exact workflow, you change your workflow. You don't own it.

Custom tools built by a partner. A person or team builds a tool made for your specific problem. You own the code and the data. You can change it, expand it, or move it whenever you want. It takes longer than a template, but you control the result.

For a small, narrow first project, custom is the right choice. You're testing an idea, not betting the company. A custom tool that solves one problem takes 10 to 20 hours of work, not months. It costs less than you think. And you own it completely.

Where to start

• Pick one problem and one customer. Find a real person at your company or a friendly customer who spends at least two hours a week on a specific, repetitive task. That's your problem. Get their permission to build a tool just for them.

• Write down the before and after. What does the work look like today. How long does it take. What does it look like after the tool works. What time or money does it save. Be specific: "Sarah spends 90 minutes every Thursday writing follow-up texts" beats "we need better follow-up."

• Talk to a builder this week. Find someone who builds custom tools for small teams, not someone selling a platform. Tell them the one problem, the one customer, and the metric you want to measure. Ask how long it would take to build a first version and what it would cost. You should hear two to four weeks and a number you can live with.