A lone figure standing at a cliff edge looking across a glowing chasm at a luminous structure on the other side
April 14, 2026 · Brandon Dietz · 9 min read · Thesis

Why implementation is king.

Here's the thing that's been bugging me for months: basically everyone I talk to knows about AI. They've used ChatGPT at least once. Some of them have even paid for it. But almost none of them have actually changed how their business runs. And that gap — between "I know AI exists" and "AI is doing the thing I used to do every Tuesday morning" — is the entire game.

I'm not being dramatic. The numbers are kind of insane when you actually look at them.

The stat that gets me every time

McKinsey's 2025 State of AI report says 78% of companies now use AI in some form. That sounds huge. Mission accomplished, right? Except then you dig in and only 5.5% of them are actually getting real business value — defined as more than a 5% EBIT impact attributable to AI and "significant" reported value. Five point five percent. Out of seventy-eight.

(Quick note: I had to double-check this one because I thought for sure McKinsey was measuring EBITA. They're not. It's EBIT — earnings before interest and taxes, no amortization adjustment. So the bar is already LOWER than I was assuming, and companies still can't clear it. That makes the number worse, not better.)

The rest are what McKinsey (kind of brutally) calls "pilot purgatory." They've run a pilot. They've had a meeting about it. They've bought a seat on somebody's enterprise plan. But the actual day-to-day operations of the business haven't moved. 62% of organizations are experimenting with AI agents. Only 23% are scaling them. So more than half have tried, less than a quarter have succeeded. That's the gap I keep talking about, except now it's got a number on it.

And in Canada, it's even more lopsided

Here's where it gets uncomfortable if you're Canadian. Statistics Canada's Q2 2025 analysis found that only 12.2% of Canadian businesses used AI in the previous 12 months. That's up from 6.1% the year before — so yes, adoption is doubling year over year. But it's still barely more than one business in ten, and that's counting ANY use, not just serious operational integration.

The breakdown by industry is pretty revealing. Information and cultural industries lead at 35.6%. Professional services are at 31.7%. Finance and insurance are at 30.6%. But trades? Retail? Home services? Construction? They're mostly nowhere. Which is funny because those are exactly the industries where AI could save the most time per dollar. The people who need it most are the ones who've adopted it least.

The Canadian Federation of Independent Business (CFIB) has a parallel report that's even more direct: 92% of small businesses in Canada use digital tools of some kind, but only 10% have fully integrated them across operations. Ten percent. Most of the rest are in what I'd call half-integrated purgatory — they have the tool, they don't have the workflow.

Here's the good news side of the CFIB data though: Canadian SMEs that actually use generative AI report saving an average of 1.08 hours per day. That's over five hours a week. Twenty-five hours a month. Three hundred hours a year of reclaimed time per person. And 55% of businesses that invest in digital tools see returns within the first two years, with an average 29% boost in productivity in year one. For every dollar spent on tech, the average Canadian SME sees $1.60 back.

The returns are real. The adoption is not.

A luminous bridge made of glowing amber and teal threads being woven across a dark chasm
The bridge is built mid-span. You don't have to see the far side to start laying threads.

So why is it so hard to cross?

I've thought about this a lot, and honestly I think the reason is pretty boring. It's not that AI is too hard. It's that implementing anything — AI, software, workflows, whatever — is a long, messy list of small decisions that nobody wants to make. Which model. Which API. Which prompt. What happens when it fails. Where do the outputs go. Who reviews them. What happens when the vendor changes pricing next quarter. What happens when your forms change. What happens when you need to hand it off to someone new.

Each of those is answerable. But stacked together, they add up to a full-time job. And most small businesses do not have a full-time job to spare. They have an owner who's already running at 110%, a couple of employees who are also running at 110%, and a to-do list that gets longer every Monday. The idea that they're going to carve out a quarter to properly implement an AI automation strategy is, frankly, a joke. It's not gonna happen. And I say that as someone who's tried to make it happen for myself multiple times.

The CFIB data actually backs this up: the #1 and #2 barriers to AI adoption for Canadian SMEs are "lack of understanding of AI's benefits" (62%) and "lack of in-house resources" (60%). Those aren't technical barriers. Those are implementation-time barriers. Somebody has to sit down and do the work, and the business doesn't have the somebody.

That's why most "AI for small business" content is basically useless. It's pitched at a business owner who has time to go learn. Most don't. What they need is somebody who's already done the learning, already made the mistakes, and is willing to sit down with their actual systems and wire it up.

Which is kind of what I want to do

So I picked a side. Obsidian AI Labs does not do AI awareness consulting. I'm not gonna walk into a room and explain that LLMs exist. You already know. Everybody knows. If you want that, there's a podcast for it.

What we do is implementation. Two ways, because different people want different things:

Option one: show you how. If your team wants to own the infrastructure long-term — and some of them should — we walk through the setup, leave videos and docs behind, and make sure at least one person on your staff can maintain and extend the thing when we're gone. The goal with this path is that you don't need us in 6 months. You use us to get over the initial hump, and then you're running.

Option two: do it for you. If you'd rather stay focused on your actual business — selling, serving customers, running the team — and have somebody else own the plumbing, we can do that too. We build it, we deploy it, we maintain it, we migrate it when the underlying tools change (and they WILL change, trust me). You pay a reasonable monthly fee and you stop thinking about it.

Neither option is better. They're answers to different questions. A five-person trades company probably does not want to become an AI tooling company on the side. A professional services firm with a sharp ops person might want to absorb the capability and keep iterating. Both are fine. The answer depends on what your team actually wants to be good at, and that's a conversation, not a pitch.

The thing nobody talks about

There's a second part of this that I keep trying to explain, and it's honestly harder to sell than the implementation piece. It's what happens to your brain after the menial layer actually goes away.

For most of us, our whole working life we've been operating with this implicit math: every hour of thinking costs us two hours of execution to pay for it. You want to try an idea? Cool, now you also have to do the work to test it. You want to try three ideas? Now you're doing six hours of work. And so people stop having ideas, because the cost of each idea is so high. Thinking gets expensive. You default to what worked last time.

When the execution layer stops eating your week, that math breaks. You have more ideas than you can implement. You have to pick, which is uncomfortable because we're not used to the ideas being the bottleneck. You end up thinking about thinking. Which sounds pretentious as hell until you actually experience it and realize it's just... more fun. More of your time gets spent on the part that's actually interesting.

Stop with the control. Let AI take over the menial things — humans rise to thought leadership and innovation. If we can 10× our output, we can hire more people, not less. It changes from a knowledge-based economy to an ideas-based economy — you're only limited to the idea that comes to mind.

Sidenote: one of the things that worries people about AI is they think it's gonna take their job. And I get it. I'm a cybersecurity guy by day, I've seen how automation eats roles. But I genuinely think the math goes the other way for most SMBs. If I can make your operation 10× more productive, you don't need 10× fewer people. You can take on 10× more work. You can hire more people, not less. The ceiling on small businesses has always been execution capacity. AI takes the lid off, and what you do with the room is up to you.

Where we start

Okay, tying it all back. If any of this resonated, and you want to stop being in pilot purgatory and actually implement something, email me at info@obsidianailabs.ca. No sales call, no obligation, no consulting hours. Just tell me what's eating your week and we'll figure out if there's something worth building.

Sometimes the answer is "yeah, that's an obvious AI fit, let's do it." Sometimes the answer is "honestly you should just fix your existing spreadsheet, AI isn't the right tool for this yet." Either way, I'd rather tell you straight than sell you a contract you don't need.

Implementation is king. Everything else is just talking about AI.

Sources

← Back to all posts