Before AI, Get Your Data House in Order

By Sara Hanks, President & CEO, NWIRC

By now, everyone has heard of AI. What manufacturers are really asking is: “What should we be doing to get started?” At NWIRC, we’re starting to work more directly with companies exploring AI and advanced analytics. But the first step isn’t spinning up a chatbot or investing in a flashy new platform. It’s way more foundational: getting your data in shape. That might not sound exciting, but it’s the difference between solving real problems and spinning your wheels.

Start by Identifying the Data That Matters

Every company has data, on machines, in spreadsheets, even scribbled on whiteboards, but not all of it is important. Start with the processes that drive your business forward. What drives cost? Where do delays happen? What makes customers happy—or not?

Then ask: what data is being generated in those processes? That’s your “important data.” And that’s where your attention should go.

I learned this the hard way during my time leading digital transformation at GE Transportation. We had powerful tools and eager teams, but the data was a mess. Part numbers were duplicated. The same supplier had five names in the system. Our reports told a different story depending on who ran them. We couldn’t automate anything until we fixed the basics. So we started small: defining a source of truth for major data, standardizing formats, and rebuilding trust in the numbers. Only then could we do the cool stuff.

Fix the Data Defects

Once you’ve honed in on your important data, it’s time to dig into quality. A quick data audit doesn’t require fancy software – just a curious mindset and a willingness to look under the hood. Watch for these common issues:

• Missing Fields: Are critical inputs left blank?
• Inconsistency: Are names, products, or machines labeled in multiple ways?
• Wrong Format: Mismatched units, inconsistent date formats, or symbols that confuse the system.
• Outliers: Numbers that clearly don’t make sense, such as cycle times of zero or machines that ran 18,000 hours.
• Duplicates: Similar records that aren’t quite the same, often caused by disconnected systems.

Fixing these issues pays off quickly. You’ll see clearer reports, fewer surprises, and a more confident team. You’ll spend less time explaining bad data and more time using the data in decision making. And best of all? You’ll be ready for what’s next.

When to Start Using AI

You don’t need perfect data to get started—but you do need clarity. Once you’ve cleaned up your most important data, it’s time to explore. But before you dive in, set a few ground rules:

• Who can use AI tools, and which tools?
• What types of tasks are appropriate?
• What data can and can’t be shared with public AI platforms?

These don’t need to be complex policies, just a starting point. Guardrails give your team confidence to try new tools without worrying about crossing a line. Then, keep it simple. Use AI to summarize customer feedback, assist with documentation, or automate repetitive admin work. Focus on tasks where small wins can free up time and improve accuracy.

What’s Next?

At NWIRC, we’re building assessments and tools to help manufacturers evaluate their data, clean it up, and make confident steps toward using AI. If you’re not sure where to begin, or whether your data is even usable, we’re ready to help. Because before you do something smart with AI, you need to be smart with your data.