Messy data is holding enterprises back from AI

In March 2024, I wrote that without good data, your generative AI system will be about as helpful as a warehouse fire. As I teach my generative AI architect students, AI, specifically generative AI, is ultimately a data-oriented problem. If your data game is weak, your AI solution will be too.

AI systems need gigabytes and gigabytes of clear and accurate data to be effective. They find patterns within data and respond to your requests of what those patterns likely mean and how to leverage the insights for strategic business purposes. Suppose the data lacks hygiene or accuracy or is dysfunctional? You’ll get incorrect inferences for your AI system or perhaps answers you won’t know are wrong until it’s too late.

Hey, at least AI is cheap to build and run… oh wait.

Nowhere near ready for AI

According to a recent Enterprise Strategy Group report, a survey of 800 IT decision-makers revealed that more than three in five organizations have notable gaps in their AI readiness, particularly in infrastructure and data ecosystems.

Organizations remain optimistic about AI’s long-term potential, but many still need to prepare for wide-scale adoption. More than half of AI decision-makers are concerned about their IT teams’ ability to keep pace with the rapid innovation driven by generative AI. Enterprises must enhance data processes and strengthen infrastructure, among numerous other tasks.

The measurement of success for generative AI projects varies significantly among enterprises. Approximately two in five organizations monitor progress based on qualitative impact analysis, the accuracy of AI responses, or user and process quantitative benefits. Slightly fewer organizations, about 38%, use cost savings as a primary success marker. As AI initiatives demand a larger share of enterprise budgets, the push for ROI is expected to intensify.

Overall, the report underscores that although enterprises are eager to harness the potential of generative AI, significant infrastructure and data management groundwork is required to realize its benefits and ensure sustainable, long-term success.

A CIO’s to-do list from hell

Most enterprises knew they had data issues long before AI started to impact the market in significant ways. Indeed, most have avoided AI and business intelligence investments due to their lack of confidence in their data. Nobody in the company completely understands where the data is and what it means. Silo leaders own and manage the data, so there is no single source of truth for things as simple as what a customer is and where customer data should come from. Redundancy is common in sales, production tracking, and other areas where the data is mismanaged.

How did things get this bad? Most enterprises spent years focused on new, shiny objects such as ERP and CRM systems, which contain important data, but it’s locked up in proprietary data stores. After ERP and CRM came data warehousing, distributed systems, data integration, and now cloud. Through it all, data has gotten more complex, distributed, and heterogeneous, with a lack of centralized control. Too many companies don’t understand the metadata and can’t trace data properly through the business processes. Also, acquisitions have driven some data redundancy; many enterprises still operate the older systems that came with the businesses they acquired. Now, we’re facing AI, where the meaning, structure, and truthfulness of data are not optional.

CIOs must fix their data before AI can provide any value. This may be too expensive or too risky for many. The inability to adopt AI due to data issues could end up killing some businesses as their competitors embrace AI as a critical force multiplier of innovation. We’ll end up with the haves and the have-nots.  

Closing the gaps

Attention enterprises: If your data isn’t ready, you should avoid AI. Many of the failed AI projects I’m called to opine on can be traced back to poor data ecosystems and an unwillingness to fix them. Some companies believe that AI will be able to fix the data for them, but that’s never the case.

My advice is that getting your data in a better state offers more benefits than just AI readiness, and it’s worth the time and money to do so. AI investment appears to go toward fixing past screwups, and many CIOs are more than happy to make it the next leader’s problem. You’ll have to ask your leadership and your board for money to fix years of neglect and lack of data strategy, with no direct business benefit that is easy to trace. Most CIOs won’t have this conversation.

My advice is to fix your data whether or not you’re looking to leverage AI. Perhaps AI can be the motivation to finally get your enterprise’s data in a much more functional state. I’m okay with that.