How to Use AI in Business: Start With Clean Data, Not Fancy Tools

A while back, I worked with a business that was excited to jump into AI for advanced analytics. They wanted dashboards and reports that predicted trends and surfaced opportunities to drive real value for their customers. On the surface, it sounds like a great move. But there was a big problem they faced with one of their critical data sets: their item master was full of duplicated items and errors (incorrectly setup items). To complicate their situation even more, they didn’t have a process defined to create or maintain items within their master so as new items would be added, their problem compounded.

They were facing the age old “garbage in, garbage out” dilemma. AI doesn’t magically fix problems with your data, in this case it would simply deliver inaccurate, and confusing insights even faster than before. The problem wasn’t that they wanted to implement an AI solution, the problem was their data wasn’t ready for it.

Prefer listening? Watch this week’s Solo Session where I expand on the topic.

Signs Your Data Might Be a Problem

Most people know their data is messy, and certainly “no perfect.” But it’s easy to ignore because you can still “make it work.” The problems is, that mess get amplified the moment you try to use it for reporting, forecasting, or AI. Here are a few red flags I see all the time:

• Reports don’t match. Sales, operations, and finance all pull numbers on the same thing, and every report says something different.
• Too much “cleanup.” Your team spends hours massaging the data before it can be used for insights.
• Duplicate or missing records. You’ve got the same item listed three different ways, or you can’t even find the right customer number when you need it.
• No oversight or governance. Nobody owns the process for creating and maintaining critical data masters like customers, items, etc.
• Customer complaints. Wrong addresses, repeated emails, incorrect invoices all point to bad data.
• No one trusts the numbers. If leadership second guesses every report, you don’t have a leadership problem, you’ve got a data problem.

Practical Steps to Get Your Data in Shape

You don’t need to overhaul all of your systems overnight… and in most cases that’s impossible anyway. Instead, start small and focus on building better habits that can at least get you on track moving forward. Here’s how I like to start:

1. Standardize

Define a process that allows for oversight and maintenance for critical master data sets. You can create templates that require consistent formats for key data like items, customers, vendors, etc.

2. Assign Ownership

Clean data doesn’t happen on it’s own. Make it someone’s responsibility (or better yet, a team’s) to manage and maintain it.

3. Centralize

The goal here is to create a single source of truth for your data. Depending on your organization infrastructure and system architecture this may be broken up across multiple sources of truth for specific subsets of the business but many times this is overcomplicated due to existing internal practices.

4. Validate

Here we can proactively check as well as start cleaning up historical transactions. Depending on how dirty your data is, a cutoff date may need to be established where by you have confidence that your data is accurate on a go forward basis.

5. Maintain

Think of this as your cycle of improvement where you can identify, prioritize, and implement new ideas to maintain your data integrity and governance.

Where AI Fits In

Once your foundation is solid, then AI can actually help you leverage your data. But you’ll need more than clean data to have a successful AI adoption. Every successful AI initiative I’ve seen start with:

• Clean, reliable data
• A clear definition of the intended outcome

Said differently, you have to be able to answer: what problem are we solving here? If you can’t answer that in plain English, AI won’t help you, it will be one more shiny object that you’re chasing.

Final Thoughts

AI is an incredible tool but one that relies on clean data. Are records duplicated? Do teams use different formats? Is there a clear process for maintaining your data masters? If you answered no, then AI won’t save you, it will only amplify the mess.

Start small. Pick one dataset that’s causing constant headaches and run it through the process above. Once your foundation is strong and you know exactly what outcomes you’re aiming for, then AI will actually deliver on it’s promise.

That’s it for today.

See you all again next week!

Dave