Before you build AI, Fix the foundation: Why data chaos will break your vision

It started on a Tuesday.

The CFO had just walked out of a board meeting, fired up.

“We need to be smarter about AI,” he said. “The board expects progress this quarter. Let’s put together a pilot. Use some of that customer data. You know, build something fast.”

It wasn’t the first time the idea had come up.

The data team had done the slide decks before. They’d even built a slick "demo chatbot" that could “analyze” customer feedback. It looked great in the room.

But nothing ever shipped. Nothing stuck.

Why?

Because everyone working with data knew:

To build it for real, would mean rebuilding the company’s entire data foundation and platform from scratch.

And this one we have? It’s a spaghetti monster — 20 years of patchwork, system by system, team by team — built without a clear data strategy in sight.

The First Signs of Trouble

Two dashboards. Same KPI. Different numbers.

An executive asks:

“Wait, how many customers did we acquire last quarter, I see different answers?”

Silence.... Then a 30-minute explanation begins. Filters. Definitions. Overlapping sources.

And, of course, the familiar:

“Well, it depends on how you count.”

Confidence? Gone.

And before anyone dares say the letters “A” and “I” again, someone from data team mutters:

“We should probably fix the data first…”

Why? Because without clean, connected, contextualized data, none of the shiny stuff works.

What you thought AI needed vs. What it actually requires

Here’s the thing: Most executives still believe AI is something you can buy your way into.

Hire a “data person.”Plug in ChatGPT. Voilà = transformation.

But no model, no matter how advanced, and no flashy tech stack will save you from a chaotic data ecosystem.

If your business logic lives in scattered spreadsheets…If no one agrees on what an “active customer” is…If your systems barely talk to each other…

Then your AI initiative isn’t a transformation.

It’s a science project. A shiny demo. One that never scales.

What the data team knows (Especially in bigger companies with many systems and complex products)

From the outside, it looks simple. Just get the numbers. Build the model, Analysis, Dashboard. Ship it, and nobody cares how you got to those results.

But inside? This is what really happens when someone asks a “basic” business question (exaggerated a bit for the purpose):

• "How many customers do we have?" CRM calls it Cust_ID. ERP calls it CustNo. One field is empty half the time. Data people fixed that after 2 days but now, which do we mean, are we counting in those that are in the system but not active or do we want active, but who can answer what active is? Do we have an integration to that new system, heard we migrated some customers there last week without anyone notifying data engineering and analysis team.

  
  

• "Let’s merge it with sales data." Except the sales team uses product names, and the warehouse logs org. codes. But wait product team has an excel that translates them so we can know what is what (if its updated because "Jessica" is on vacation this month).

  
  

• "Clean it up?" Sure, if someone can manually fix typos, duplicates, missing values, and naming clashes from three tools no longer in use.

  
  

• "Just apply the business logic." Which one? Gross revenue? Net? Do we subtract refunds? Cancelled subscriptions? System X is not built like that so we can not actually see that metric so we need to calculate it. Wait, its calculated already directly in BI system in one dashboard, but that guy quit 3 years ago so, hmm how was he thinking. Lets not even talk about endless joints in SQL in your data warehouse that only 2 guys in the company know something about (if they have documented it).

  
  

• "Feed it into the AI model." Okay. Weeks later. After chasing definitions, formats, and 10 back-and-forths with stakeholders. Hope this is right now.

  
  

• "Monitor for changes." Worked for a whole week until developer somewhere renames a field. Again! And it all breaks.

  
  

This is not about SQL. It’s not about dashboards. Its not about your analysts or engineers or even efficiency of your IT.

Data ecosystem is not a system or an application, its an organism that changes and feeds on rigorous structure. Feed it and it will grow!

It’s about Context, Communication, Structure, Ownership, Good data contracts between data producers and data consumers. Rigorous definition documentation and central calculation of things. Its about changing millions of details in your business, processes and system flows.

It’s about companies going back to drawing table and building the right data foundation from scratch peace by peace, data package by data package, pipeline by pipeline.

This is the real work. And it’s invisible — until it’s not.

The best companies have understood this, others have not

• Take Commonwealth Bank of Australia. They didn’t build 2,000 AI models by luck. They started by building a unified data platform — one that processes 157 billion clean, monitored data points daily. Then came the AI couple of years later.

  
  

• Or John Deere.They didn’t just install smart tractors. They created closed-loop feedback systems from field sensors to cloud analytics platform. That’s how they cut herbicide use by 59%.

  
  

• And Air Canada? Their chatbot invented a refund policy. The company, not the AI, was held accountable in court. Why? Broken data foundation and no good data ecosystem behind the AI.
  

  

  Mature companies are investing in DATA first!

The quiet truth most leaders miss

Real AI transformation isn’t loud. It’s not a flashy demo or viral LinkedIn post.

It’s silent, foundational, and invisible — until it’s not.

It starts with the data. The hard, boring, cleaning, fixing, agreeing, structure-intensive part.

Because if your Power BI dashboards show different numbers today…What do you think your AI agent will do with the same mess?

As a C - level executive, Before you say “Let’s Do AI,” Do this first:

• Learn about your data ecosystem

• Ask what will it take to fix it, and do not be surprised or angry, if the answer is millions and years. Try to truly understand and see how you can help speed it up. Data people need your support here and don´t judge them, trust me its not their fault!

• Ask if your systems are monitored and if they share information

• Ask what happens to data when a source system changes something or when product/commercial people magically change business logic in that random meeting.

• Ask who owns the business logic, not just the code. Your managers maybe know, but do they agree with each other, and is that retrievable in your systems?

• Ask if your data team has the competence to transform into tech that is actually needed forward, even after you fix the data.

• Ask if IT security is ready, if compliance is ready, if there is any governance in place which will be needed.

  
  

These are not tech/data questions. They are transformation questions. Strategy & Culture questions. Leadership questions. Investment questions. Compliance questions.

Final Thought

Your AI ambitions are valid. Your vision is important. But if you want the kind of transformation that sticks, not just another pilot:

Fix the data foundation first. Make sure that your data ecosystem provides good, clean, structured and validated data sets that feed your AI in a good way.

That’s how you go from playing with AI...…to winning with it.

//Adnan