Case Study: FileMaker Machine Learning Capabilities

Executive Summary

As a low-code development tool, FileMaker is as effective a solution for unconventional business workflows as it is for standard processes, like a project management system or a QuickBooks integration. We can even leverage FileMaker machine learning capabilities to automate talent recruitment.

Reading and categorizing job descriptions is challenging work. It’s very monotonous, and it takes a long time for a person to become an expert. But this is an ideal use case for custom software. See how we can simulate algorithms and teach FileMaker to tag content automatically.

The Scenario

A small but innovative talent recruitment company leveraged custom business software to help customers place targeted employment ads across various contextually relevant media outlets. The technology is intended to automate mundane but essential tasks that people don’t necessarily like or want to do. But it’s also one of their core competencies.

In theory, the operation is relatively straightforward:

1. The customer posts a free-form text block describing the job on their website.
2. The technology scrapes the post from the customer’s website.
3. The content is reviewed and categorized before it is distributed and promoted on applicable employment and trade properties.
4. The company provides proof of performance to customers for invoice reconciliation.

A human was responsible for processing the job post. But, it was a tedious process that required a lot of time for someone to get good at and become an expert. Usually, the person would eventually move on to a more fulfilling role, so someone else would have to step into the role, thereby restarting the time-investment clock.

So, this type of task is ideal for a computer to perform. Of course, it would require a good amount of programming and machine learning to accomplish the workflow, but nothing a team of professional FileMaker developers couldn’t handle.

The Challenges 

• Store a lot of text-based data and integrate it with a FileMaker back office system.
• Digitize a monotonous and evolving process.
• Distribute customer ads to disparate employment platforms and third-party sites via application programming interface (API) functionality.
• Replace an expensive and hard-to-configure web data extraction tool.

The Solution

The first thing The Support Group did was rebuild the system with FileMaker as the authoritative source. Given the expanse of the company’s data, we decided to synchronize the records with external databases to make it more efficient for FileMaker to host the data.

Next, we developed a homegrown Python-based tool to scrape data, specifically job postings, from customer websites. The off-the-shelf application that initially handled this task was expensive and difficult to configure. The new program featured a user-friendly interface and significantly streamlined the extraction process.

Then it came time to automate the categorization process. The goal was to minimize as much human interaction as possible. So, we manipulated FileMaker in such a way that it would rely on machine learning to analyze and categorize job postings. Basically, we fed the FileMaker app different job postings and “taught” it how to process keyphrases and assign appropriate tags to the ads. The teaching is done with a combination of the Python programming language and SQL database technology.

The system distributes the ads to internal job boards and third-party media outlets depending on the tag values, so the ads must be tagged correctly. We use APIs to automate this process, but each platform has its own API requirements and a different way of tagging text. As a result, we created a what-you-see-is-what-you-get (WYSIWYG) tool to accommodate the various API requirements. So even non-technical users can configure the API connections. In addition, the mechanism is flexible enough to talk to different APIs, like JSON and REST API.

The application is a constant work in progress, particularly the machine learning aspect. Job titles, roles, functions, etc., change frequently. Furthermore, customers use their own writing styles and formats for the descriptions, so the software has to be able to parse content regardless of its presentation. Therefore, it continually learns and adapts to unfamiliar and specialized words. Humans spot-check the workflows from time to time to optimize efficacy. And a human intervenes if the application cannot categorize the job post.

Some technologies used 

• Python 
• FileMaker
• MySQL and Postgres

The Results

•  The technology’s accuracy and reliability make it an indispensable part of the operation. 
• The company’s customers appreciate the simple, autonomous, and transparent advertising service.
• The company saved millions on technology expenses and human resources costs.
  • We optimized the tech stack.
  • The company redirected 80% of its staff to sales positions where they focus on selling the service instead of working on it, resulting in increased employee retention and job satisfaction.

Curious how automation or machine learning could fit into your FileMaker system? Connect with our team to start the conversation.