Your AI Strategy Is a Data-Modeling Strategy

October 19, 2026

Summary

Many enterprise AI pilots fail to deliver results, and the data foundation underneath them is usually an overlooked reason why. In this piece, Forward Financing’s VP of Analytics and Data Science explains why AI exposes every gap in a company’s data foundation. He also shares how years of disciplined data modeling now power Forward’s machine learning models and conversational AI agents, and what leaders can do to get their own foundations ready.

***

The companies pulling ahead with AI aren’t the ones with the best models. They’re the ones with the cleanest data foundations.

Businesses are spending more on AI than ever, and a lot of it isn’t paying off yet. Gartner predicted that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025 — the actual number came in higher, with the average share of generative AI pilots scrapped before reaching production closer to 50%. Gartner has since forecasted that more than 40% of agentic AI projects will be canceled by 2027.

Ask why, and you’ll hear a handful of explanations: workflow integration, misallocated budgets, weak governance, runaway costs. All of them are real. But every one of them runs through the same bottleneck: can the system trust what it’s reading. That’s the layer most leadership teams have treated as an afterthought, and it’s the one this piece is about.

I lead the data team at Forward Financing, and I’ve come to believe many leadership teams underweight the data side of AI. It’s both a technology problem and a data problem. The winners won’t be the companies that picked the best model or the flashiest agent platform. They’ll be the ones that also did the patient, unglamorous work of organizing their data long before AI was on the agenda.

AI Shows You What Your Foundation Is Made Of

For years, messy data was survivable. Maybe a dashboard had a slightly off definition. Maybe two teams reported “revenue” three different ways. Maybe the logic behind a key metric lived only in one analyst’s head. Everyone knew about these quirks and worked around them, because someone smart was in the loop to catch the problem.

AI takes that person out of the loop. Put a language model on top of your data, and every ambiguity you learned to live with becomes a liability. The model will write a query that is syntactically perfect but wrong for your business, then hand you a confident, plausible-looking number that is simply incorrect. Garbage in, confident garbage out, at machine speed and scale.

The numbers should get every executive’s attention. Industry benchmarks have repeatedly found the same pattern: a language model querying raw, unmodeled database schemas typically lands somewhere in the 40-60% accuracy range. Ground that same model in a governed semantic layer, where metrics, tables, and columns are clearly defined, and accuracy climbs into the 85-95%+ range. The model doesn’t change. Only the foundation does.

It’s no surprise that data quality tops nearly every survey of AI adoption barriers — Gartner has found that 63% of organizations either lack, or aren’t sure they have, the data practices AI requires.

The Foundation Took Years of Deliberate Work

The work that makes AI succeed is work you should have been doing anyway, and it predates the hype by years. At Forward, disciplined data modeling has been a first-order priority and never an afterthought. We consolidated fragmented sources into a single, governed warehouse, then modeled that data into clean, layered, well-documented datasets. The result is a true single source of truth. A metric like conversion rate is defined exactly, and every report, model, and agent draws from that same definition.

We also put real effort into documentation, and we kept people accountable for definitions. AI can help you generate docs, but a person has to own what each metric means. Auto-generating definitions from raw tables just bakes in the ambiguity you were trying to remove.

That foundation is what lets us build with confidence today. The clean, governed layer that powers our machine learning models, the ones behind smarter, faster funding decisions for small businesses, also powers the conversational AI agents our employees use to get insights from our data using natural language. We never needed a separate “AI data strategy.” We needed a strong data strategy, and AI became one more thing we could safely build on top of it. The discipline compounds: better data leads to better models, and better models lead to better decisions.

What This Means for Leaders

Fund the foundation, not just the demo. A working demo proves the technology exists. It doesn’t prove your data can support it in production. Budget for the modeling and governance underneath, because that’s where many failures start.

Insist on a single source of truth. If two teams can produce two different numbers for the same metric today, your AI will produce a third. Define datasets and metrics once, and have every tool consume them.

Keep people accountable for meaning. AI can scale your experts’ knowledge, but it can’t decide what your business terms mean. Someone has to own each definition.

Treat data as a long-term asset, not a project. The teams succeeding with AI didn’t sprint to it. They invested in clean, modeled data for years, so they were ready when the moment arrived.

The Unglamorous Truth

Most of the AI conversation is about intelligence: smarter models, more capable agents. The real differentiator is whether that intelligence has solid ground to stand on. The companies pulling ahead made their data trustworthy years before AI made it a requirement.

So before you ask what AI can do for your business, ask whether your data foundation is ready to be trusted by something that won’t second-guess it. If it isn’t, fixing that is your AI strategy, and today is a good day to start.

Sources and Further Reading

About the Author

Mike Powers

VP, Analytics & Data Science