FileMaker AI in the Real World

Introduction to AI in FileMaker

Building an automated invoice processing app to scan, and code invoice line items using OCR and AI in FileMaker

For the last few years, I’ve enjoyed the opportunity to work as the primary developer on the custom FileMaker app for Blueprint Capital REIT, a real estate lending and development firm based in Seattle, WA. Although their app uses and integrates with a variety of tools and services, my head is usually living exclusively in FileMaker-land, extending the data model, building new layouts, and scripting workflows. So when the opportunity comes to integrate with something new outside of FileMaker, I get excited. And maybe a little nervous. And much more so (both excitement and nerves) for the prospect of adding some Artificial Intelligence to their app!

I, like many software developers, can often hold a (healthy?) skepticism of AI.

Toyota adds more operators to its “driverless” vehicle after it crashes into visually impaired athlete at Paralympic Games. https://t.co/nVFpWMIDbm

— hypervisible dot pdf (@hypervisible) August 31, 2021

How Tesla’s autopilot system interprets traffic lights on the back of a moving truck. I guess this scenario was probably not part of the system’s training data. A good illustration of how it will likely be impossible to reach full driving autonomy just by recording “more data”. https://t.co/eZEhxhpXBC

— Max Little (@MaxALittle) June 3, 2021

I used GPT-3 to generate new Disney villains.

I, too, rebel against the mandatory sasquatch dress code.https://t.co/n6nvZn50Jz pic.twitter.com/WDh73g3fv6

— Janelle Shane (@JanelleCShane) April 30, 2021

But as Blueprint’s accounting team took on more customers, they faced some challenges that improving the data-entry workflow alone couldn’t resolve. They estimated it cost $8 to process a single invoice; multiply that by an average of 1,200 (and growing) per month, and invoice processing quickly becomes an expensive task. Perhaps even more costly were the long hours and late nights the team put in each month. To make a significant impact, we would need to automate more of the process with the help of AI. (more…)