The product was built on a mental model that users were very familiar with but resulted in an experience that felt inefficient, draining, and unforgiving. Imagine a large real estate company that manages thousands of tenants. Traditionally, one of its employees would be responsible for reading each Lease Agreement and entering their data into the company's system (e.g., name of tenants, lease start date, etc.) Now, this new AI application makes the entire process much more straightforward. It processes all these documents automatically using Optical Character Recognition (OCR) technology and Machine Learning algorithms. Humans only need to review the work of the AI at the end of the process. The application was initially designed to allow humans to review one document at a time, as users did before this product existed. And this approach resonated well with users because it felt familiar.

However, we had the opportunity to invent a new different mental model in this case. Humans review one Lease Agreement at a time because they can read one sheet of paper at a time. But software doesn't have this limitation. Software can look at a stack of Lease Agreements and immediately see all the tenant names across all documents. What if the application was built so users could review documents the same way? Our testing revealed that this new mental model allowed advanced users to review documents much more efficiently, even though they had to think of their work differently.

In this example, the mental model users understood most intuitively wasn't the most efficient approach. Testing the new mental model with users proved the value of the approach, which was added to the experience's vision for future changes.