Only 5% of enterprises say their data is ready for AI. I think the real number might be even lower. Most organizations are about to learn what I spent 15 years teaching companies in the Business Intelligence (BI) era: The problems you ignore don’t disappear. They wait.
In business, perfection is impossible and there isn’t time to chase everything. Problems tend to get addressed only when they directly threaten the flow of money. I learned the hard way running a data and BI consultancy, because our work had a way of forcing organizations’ unaddressed sins to the surface.
Miscategorized entries in a product catalog, for instance, don’t prevent customers from buying products off the shelf and so they can sit undetected for years. But when you start tracking sales performance by category over time, that formerly harmless data quality issue becomes critically important. Ditto for imprecisely tracked inventories, scattered duplicate customer records, and time zone mismatches across systems. These problems aren’t noticed until something new, like BI, or now AI, brings them into focus.
That was the lesson of the BI era. It’s about to become a front-and-center crisis in the AI era.
Scale makes debt worse, not better
Here’s what leaders need to understand: Throwing more computers and engineers at artificial intelligence projects doesn’t solve underlying data friction. It amplifies it. I’m already seeing AI projects collide with the same below-the-radar data quality problems that derailed BI initiatives, except the stakes are higher and the pace is faster.
And in the AI era, “data quality” itself has gotten bigger. AI demands more of your data than BI ever did.
Historically, data quality problems meant inaccurate data sitting in a database—wrong product category, wrong inventory level. Incorrect data of that flavor results in incorrect decisions, whether it’s a human reading a misleading dashboard or an AI agent querying the data directly. This “original flavor” of data quality is more important than ever, and in my experience, there’s still much work to be done here.
But there are new flavors of data quality that BI could sometimes afford to ignore. AI cannot.
The decoder ring your AI agents need
The first is what the industry is now calling “semantic models,” a term you’ll hear constantly in the coming year. Semantic models are essentially decoder rings for your data. They capture things like: How do we calculate the long-term value of a customer? Yes, there’s a formula but the semantic model also has to know precisely where to find the data inputs for that formula, tracking a customer across e-commerce platforms, brick-and-mortar point-of-sale systems, and every table and field in between.
Semantic models were valuable in BI, but you could get by without them. Many prominent BI platforms, including Tableau and Qlik, lacked credible support for them. BI developers served as a kind of semantic shock absorber, translating data into meaning on demand.
An AI agent can’t wait days for a developer to do that translation. And it can’t be trusted to invent its own definitions. This is why Snowflake, Salesforce/Tableau, Databricks, and AWS recently banded together to create an open standard for semantic models—Open Semantic Interchange. Their sudden reversal, after more than a decade of downplaying semantic models, tells you everything you need to know about how essential they’ve become.
Knowledge now needs the same care as data
The second new flavor, certified knowledge, has no real BI equivalent.
Your company doesn’t run on structured data alone. It runs on institutional knowledge: what industry you’re in, who your ideal customer is, what your brand voice sounds like, how you position against competitors, and even things as specific as “the agreement Sales and Marketing reached last week.” AI agents need that same context to do useful work.
That knowledge exists in your business today. But it’s scattered across slide decks, internal wikis, executives’ heads, and folders on laptops. And those locations often contain conflicting versions of what should be a single, certified truth.
Putting AI agents to work means consolidating that knowledge into certified locations: one official version per topic, kept current, with a short list of designated owners.
The good news: the talent is already in your building
Leaders surveying this landscape often assume they need to hire their way out of it. They don’t.
Power BI developers have been building semantic models for years. They just didn’t know they were doing the prep work for AI. Data professionals of every stripe have the skills to solve the knowledge problem, too, as soon as they recognize that institutional knowledge is just another form of data. These employees have been an asset all along. The AI era is the moment it pays off.
The companies that win the next decade won’t be the ones that spent the most on models and infrastructure. They’ll be the ones who did the unsexy, essential work of getting their data house in order before the AI agents arrived and the debts came due.