Constant hype about AI. Relentless demand to produce results. Not enough clarity about exactly how to get there. These AI-related pressures come up every day when I’m talking with OnX clients across Canada.
They also showed up in primary research we conducted earlier this year with some of our government clients. In our survey, 94% of public sector leaders told us they support AI adoption, but only 6% have high-quality, well-integrated data.
These clients aren’t unique. What I see across the board are enterprises and governments pouring money into AI but neglecting a critical foundation for success.
It’s time to get serious about building it.
What houses teach us about AI readiness
Experienced homeowners know to brace for challenges when modernizing or expanding a property. A project may uncover structural, electrical, and/or plumbing issues, and any good contractor will advise resolving those issues before moving ahead with other upgrades.
It’s not so different with AI. Many leaders are getting unwelcome surprises about how much behind-the-scenes work is required to get their technology house in order. At the top of that list is preparing and protecting data.
Too often, though, data goes into an AI project unlabeled, ungoverned, and/or scattered across platforms. By now, we know that AI is nothing if not confident. Even with bad data, it’ll serve up answers. You might be able to explain away the first wrong one. The second raises some eyebrows. By the third, the team stops relying on the tool. What was supposed to transform the business becomes a sore subject, and the economics can be significant.
We saw this firsthand in a multi-billion-dollar Canadian enterprise that spent millions to deploy AI across its operations. When nearly every initiative failed, the leadership team overseeing the effort decided to step back from AI altogether. They realized they were trying to innovate on top of a shaky foundation.
Three high-priority renovations
When you buy a house, you take responsibility for managing and improving it. You can’t simply plug a smart panel into outdated wiring or ignore structural issues when building an addition. The same is true with AI. You can’t tack it on to your existing systems and data and expect it to work.
From my perspective, there are three critical investments for ensuring a strong data foundation.
1. Label your data
Think about how your own streaming service learns what you like. Every show you watch gets tagged, and those tags stack until the recommendations feel almost intuitive.
Your organization’s data works the same way but with higher stakes. Client records, transaction history, and service tickets aren’t useful to an AI system until they’ve been tagged with enough precision that the system can identify what matters.
Here’s a specific use case. By labelling your receivables properly (think: client, terms, and payment behaviour), you can easily ask some valuable questions. Which clients are consistently paying 15 days late? What would a 1% to 2% late-payment penalty return across all of them? The answers are already in your ledger, but without labels, AI systems can’t find them.
Without question, this is the most complicated of the three. It’s also the one I would prioritize. Get the labelling right, and everything downstream gets easier. But if you get it wrong, no amount of governance or migration can make up for it.
2. Govern your data
Labelling tells you what something is. Governance tells you who’s allowed to see it and what they can do with it. Not every piece of data belongs in front of every tool, employee, or AI model. Some of it’s so sensitive that it shouldn’t leave the building; some of it’s safe to expose broadly. Deliberately establishing those lines is hard but essential work.
When governance is missing, an AI system will grab as much as it can. A single compromised tool could tap into your most sensitive data, creating an exposure that goes beyond security. When you act on badly governed data, you’ll also inherit legal and compliance consequences with every decision that follows.
3. Migrate your data
Even properly labelled, properly governed data can’t do its job when it’s living on the wrong platform, split across disconnected systems, or trapped somewhere AI tools can’t access. This is like plumbing: unglamorous and easy to defer but at the root of many AI stumbles.
Labelling, governance, and migration are the code you build to before anything else goes into an AI tool. Consider them non-negotiable prerequisites: renovations you must make before you get to make the more exciting and innovative upgrades.
Building on OnX’s foundation
At OnX, we’ve spent 43 years earning the trust of Canadian organizations across every level of government, banking, insurance, and retail. That history matters because readiness work isn’t something you can shortcut with a generic playbook. That’s especially true in complex, highly regulated sectors. It requires a thoughtful examination and deep understanding of your data and processes.
That’s the foundation we bring to every engagement: national expertise, delivered with local accountability, industry knowledge that reflects how your business runs, and a partnership model built to last well beyond the end of a single engagement.
I’ve said before that trust is the hardest thing to earn and the easiest thing to lose. That’s precisely why we won’t push an organization into AI work before its data is ready for it. It may not be the flashy conversation, but it’s the right one.
Your foundation must come first.
Where to start
You may not have shaped your current environment, but it’s where you’re living and what will make or break your performance with AI-driven innovations. Fortunately, the fix is more within reach than you might think.
At OnX, we prefer to have that discussion with you now rather than after something’s gone wrong. We’re here to help you stop losing sleep over AI and start getting real value out of what you’re building. Reach out to kick off that conversation.