Fix the Data Before the AI: Why Your Data Foundation Decides Your Transformation
Most AI conversations I'm in start with the model: which tool, which vendor, which agent. Very few start with the question that decides whether any of it works: is our data in a state where a machine can rely on it?
I've seen first-hand how bad data governance makes it harder for a business to adopt technology. It's rarely the shiny new tool that fails. It's everything underneath it.
Three planning systems, one company
At a big tech company, I saw three different finance planning systems in use across different segments of the business.
That meant three support teams maintaining three systems that did the same job. It also meant there was no easy way to consolidate all the segments in one place.
The details make it worse. One of those planning systems doesn't even have actuals from the company's SAP accounting system loaded. So the business unit using that tool can't do actuals-vs-forecast analysis within the tool. They do their reporting in Excel.
Another business unit also reports in Excel, through the OLAP system. It works, but it's very slow and inefficient compared with having all the data aggregated in a database like Snowflake and building automated dashboards on top.
The result is a lot of inefficiency and duplicated work across business units. The same numbers are pulled, mapped, reconciled and reformatted by different people, in different tools, every cycle.
Now try to put AI on top of that
Ask an AI assistant to explain why a segment is off forecast. Which system does it read from? One of them doesn't have the actuals. Another depends on Excel files sitting on someone's laptop. Whose definitions does it use?
Every question an AI tool tries to answer inherits the fragmentation underneath it. You don't get one intelligent planning assistant. You get three partial ones, or one that is confidently wrong.
That's why the transformation needed to get this company ready to deploy AI is a massive undertaking. It isn't about picking a model. It's about untangling all the multi-system dependencies and all the manual Excel mappings used across the organization. That work has to happen before AI can add real value, and it's far bigger than the AI project itself.
How companies end up here: the workaround trap
None of this comes from bad intent. It comes from people trying to get work done.
What I see again and again is employees building short-term workarounds to make something work fast, instead of fixing the process for the long run. A spreadsheet that bridges two systems. A manual mapping because the master data doesn't line up. A new tool for one team because the shared one doesn't quite fit.
Each workaround is rational on its own. Added up, they become the operating model: undocumented, owned by nobody, and invisible until you try to automate across them. AI doesn't fix this. It amplifies it. An automated workflow built on a workaround makes the workaround permanent and harder to see.
It's much easier to set things up properly in the first place. But that's often not what happens when a company becomes very successful very quickly, or evolves over time by prioritizing quick wins over long-term scalable solutions. Growth rewards speed, and the bill for that speed shows up later, usually right when you want to transform.
This isn't just my experience
The research points the same way.
Gartner reports that 63% of organizations either do not have, or are unsure if they have, the right data management practices for AI. It also predicts that through 2026, organizations will abandon 60% of AI projects unsupported by AI-ready data. (Gartner)
In Cloudera's State of Enterprise AI 2026 report, a vendor-commissioned survey of 1,500 IT leaders fielded by Wakefield Research, 49% said that if they could restart their AI journey they would strengthen governance from the beginning, and 46% would invest in data quality earlier. (Cloudera)
Treat survey numbers as directional. But the direction is clear: the organizations furthest into AI wish they'd fixed the foundation first.
What a good data foundation actually means
It doesn't mean a multi-year data programme before anyone touches AI. Gartner makes the point that there is no way to make data AI-ready in general or in advance, because readiness depends on the use case (Gartner).
In practice, a good foundation for the processes you want to transform comes down to four things:
One source of truth per core process: one planning system, one set of actuals, one definition of revenue.
Data in one place: aggregated in a central database, not reassembled in Excel every month.
Clear ownership and shared definitions: every important dataset has a named business owner, and key terms mean the same thing in every segment.
Workarounds surfaced, not hidden: you know where the manual mappings and side spreadsheets are, and you have a plan to retire or formalize them.
The leader's data foundation checklist
Before you approve your next AI or digital initiative, ask your team these seven questions about the process it touches:
Where does the data for this process live, and how many systems hold a version of it?
Which one is the source of truth, and does everyone agree?
Are actuals and plans in the same place, or does someone stitch them together in Excel?
Who owns this data? Name a person, not a department.
What manual mappings and workarounds exist today, and who maintains them?
How many teams support the systems behind this process, and could that number be one?
If we automated this tomorrow, what would break first?
If your team can't answer the first three quickly, start there, not with a vendor demo.
Where to start
Pick one high-value process you want AI to help with. Run the checklist on it. Consolidate or formalize what's underneath it, then put AI on top. You'll move slower for a few weeks and much faster after that, and you'll build the habit of fixing processes instead of patching them.
Technology adoption is mostly a data and process problem dressed up as a technology problem. Fix the foundation, and the AI part gets a lot easier.
If you want to run this checklist with your leadership team, that's exactly what we do in our AI leadership workshops. Get in touch.
Comments