AI Won't Fix a Weak Data Foundation

Everyone is talking about AI right now. Boards and CEOs want to see movement, and CIOs are under pressure to show something concrete. 

We recently acquired Layout.dev, an AI application-building platform that’s designed to accelerate how organizations create applications and agentic workflows. For me, this isn’t about reacting to what’s happening right now. It’s about where things are going. Here’s how I think about it.

Business users don’t just want another dashboard. They want to connect to the systems they use every day, SAP, Workday, and ServiceNow, and get answers without waiting months for IT projects and pipelines.

For years, the interface moved from writing SQL to drag-and-drop. The next step is obvious. You type what you need, and the system builds it. You don’t click through screens. You say, “Connect these sources. Show me this. Add a filter. Build this for me.” That’s the direction.

That’s why Layout.dev matters. It changes how you interact with the platform.

But this is also where companies get into trouble.

AI works well when you’re drafting documents or generating code, because a human reviews the output. It breaks down when you try to run the core of the business on it.

If you can’t trust the system you use to close your books, how can you trust an automated system to act on your data?

Closing the books is the simplest test. Can you produce numbers you would report to the government? If the answer is no, adding AI won’t fix it.

AI is an accelerator. It makes you faster. But AI is trusted only if it works on trusted data. If the foundation isn’t solid, you’re just building faster on top of a problem.

If anything, the acquisition reflects a conviction I’ve held for years: AI will not fix a weak data foundation. It will only expose it. 

Right now, companies move quickly to build demos and connect a model to some data. In two weeks, they can show something that at first glance looks impressive. That’s not the same as running a business on it. 

When we talk about enterprise AI, it’s worth asking a simple question. Say that you’ve invested millions in your data platform Snowflake, Databricks, Fabric, whatever it is. These are smart systems built by smart engineers. But can you use any of it to close your books? 

Closing the books requires numbers that reconcile. Numbers you would submit to the IRS. If your current system can’t reliably do that, then adding AI on top won’t magically fix the underlying problem. 
‍

“Good Enough?” 

The first problem is the quality of the data. 

Most enterprise data environments weren’t designed to operate directly on detailed, highly relational source data at scale. The engines underneath were built decades ago. ETL pipelines exist because those engines can’t handle the raw complexity of enterprise systems directly.  So, the data gets simplified just to make the engine run. Once you do that, you’ve already lost part of the context. 

When you reshape and aggregate data, you lose detail and eventually start to see mismatches. Sometimes they’re small and might be acceptable. But “good enough” won’t cut it in areas that don’t tolerate approximation like finance, supply chain, or banking. 

That’s why many AI initiatives look promising at the surface but then stall when they reach the core of the business. Writing emails, drafting documents, and generating code — those are productive uses. A human still reviews the output, and the error tolerance is manageable. 

But when you are reconciling invoices, matching contracts, validating currency rates, or recognizing revenue, the answers must be precise. If the underlying data is incomplete or simplified, the model will produce an answer anyway. If something is missing, it still gives you an answer. That’s a problem in finance. 
‍

Building a Foundation 

The second problem is control. Large language models are generalists. They won’t understand your company’s specific financial rules or contract clauses. If you expose enterprise data to a model without constraining it, you’re relying on probability where determinism is required. 

 Our view at Incorta has always been that the foundation should come first. We operate directly on live enterprise data using Direct Data Mapping™. Instead of rebuilding everything into aggregates before it becomes usable, we work with the data as it exists in systems of record, such as SAP, Oracle, Workday, and others, preserving detail and relationships. 

When we apply AI, we don’t just hand the question to an LLM and expect it to guess the right tables. Incorta first narrows the data down to the exact tables and columns. Then we ask the model to work inside that. That way, the model is operating within the real structure of your enterprise data, not inventing context. 

This is exactly why Layout.dev matters. 

For decades, the interface evolved from SQL to drag-and-drop dashboards. Now the interface is shifting again. Instead of building reports manually, users want to describe what they need to connect these systems, monitor these invoices, flag anomalies, trigger actions, and have the system build it. 

Layout.dev gives us that interaction model. It allows business users and developers to define workflows and agents directly on top of live enterprise data. 

But here’s the key: the interface can change without compromising the foundation. The easier it becomes to build an agent, the more important it is that the data it acts on is complete and governed.  

Look at something simple like invoices. Today, validating an invoice means chasing it across systems, invoices, contracts, rate tables, and currency rules, and manually reconciling the pieces. If you have hundreds of them, it takes a lot of time. An agent can automate that. But if the underlying data doesn’t reconcile across systems, you’re automating error at scale. 

AI is a multiplier. It multiplies precision, or it multiplies error. If the foundation is accurate and trusted, AI can increase speed and efficiency. If the foundation is fragmented or simplified beyond recognition, AI will expose those weaknesses more quickly. 

So before making AI mandates across the enterprise, ask yourself whether your current data platform produces numbers you would confidently report to the government. If not, that is the place to focus first. 

AI will absolutely change how we interact with systems. That’s why we invested in Layout.dev.  

To learn more, read our official press release here.