"You need clean data before you can use AI" is a myth.

In many of my prospect and client calls, I keep hearing this misconception that you need to have all of your data completely cleaned, deduped, tagged and organized before your organization can start to use AI.

This myth is perpetuated by people with poor AI strategy, and poor implementations. I know it is tempting to build an agent, give it access to all your systems and data, and let it go wild. Not only is this an operational nightmare and liability, people believe that this is the primary form of AI deployments.

Of course if you give an AI agent unrestricted access to all your company data, with no instructions, guidelines or guardrails, you are going to get poor AI performance. How is it supposed to know what data to use, and what is considered the correct source of truth?

Don’t treat AI any differently than you would a new employee. You wouldn’t (or shouldn’t) dump every task onto a new employee at once. You give them tasks overtime after proper instructions and training.

Now use that same philosophy with building agents. Start with a very specific workflow. Choose a task where you are able to provide clean data sources, with clear success criteria.

One you have that workflow working, add a second one, and a third. Build up overtime. This also gives you and your employees time to prep data sources for future workflows.