AI Return on Investment (ROI)

Building a compelling business case that wins executive support for a new investment is one of the hardest responsibilities of a successful business innovator.   Once the investment is made, an even more difficult task proceeds the case, proving there was an actual, attributable ROI.

AI is full of boundless hype.  It’s difficult for innovation leaders to make a case for applying it especially when it’s a relatively new technology in the commercial sector, and there’s a lot of fear and doubt abound.  Likewise successful AI is often a function of the depth and quality data fueling it.  Even when AI is successfully employed to solve an important problem, few innovators take the time to “prove” it.  This is rarely in issue in a technology company, but the lack of proof can torpedo AI momentum in more traditional businesses.

Early on in our company’s journey we had an opportunity to work on a feasibility study project partnering with the finance department of a large business services company.  The company had over a hundred thousand small business customers that made up a substantial portion of their revenue.  Our main contact made a compelling case for the initial investment that funded our study.  And, if it proved successful, our AI model would be operationalized.

The case can be distilled down to something like this:

Problem: “A significant portion of our small business customers do not pay their invoices on time which substantially, negatively affects our recognized revenue.  By investing in AI, we will identify and deploy optimal interventions that will significantly increase the number of customers who pay us on-time and, thus, increase our recognized revenue.”

This was a bold case for a company who hadn’t deployed AI for any operational or financial reasons up to this point in time — this was pre-chatGPT and pre-pandemic times.  Therefore, it was important for our main contact and his team to demonstrate that deploying AI to help solve this problem would generate a positive ROI.

Prior to embarking on this bold project, the people and process for the company’s small business revenue collection looked something like this:

On average, small business customers that didn’t pay their invoice on time, settled 68 days later.  Our feasibility study goal was to demonstrate a reduction in settlement time so that the business would capture its revenue as close to the due date as possible.

The first part of the project entailed compiling and analyzing all their historical accounts payable data for small business customers.  The data was high quality with significant depth and breadth (over 25 fields for each invoice record).  We then used this data to train a machine learning model that predicted when a given customer would pay their invoice.  Customers that were likely to pay late were segmented into groups based on the risk of them not paying on time (based on the model’s output) and the amount of their upcoming invoice.

This information was then used to fuel a series of people and process improvement experiments that entailed changes like:

• Send an automated email to the customer X days before the invoice is due
• Have a Collections Rep write and send a personalized email to customer Y days before the invoice is due
• Have a Collections Rep contact the customer over a phone call Z days before the invoice is due

The challenge was to find the most affordable action(s) that would work most effectively to decrease the settlement time for each customer segment.  The higher the invoice amount and risk, the more manual the intervention needed.

By coupling the output of the model with people and process improvements, the company was able to show a significant reduction in the average settlement days from 68 days down to 40 days.  This was a 40% improvement resulting in an estimated 14% increase in monthly recognized revenue, a very large amount given the number of small business customers that paid late.  The investments made to train the machine learning model and experiment and change processes were significantly less than the captured revenue making this a very positive ROI for the company.

In conclusion, the successful implementation of AI in this real-life example demonstrates the critical importance of understanding and baselining a current state process, training a machine learning model to predict something that may optimize the process, and then experimenting on process changes using the model output to find the optimal outcome.  This iterative approach allowed the company to identify and execute interventions that would significantly increase the number of customers who paid on time, leading to a substantial increase in recognized revenue when operationalized.

This case study underscores the importance of human-machine collaboration in process improvements powered by AI.  By combining the insights from the machine learning model with the expertise and judgment of human collection representatives, the company was able to design and implement interventions that were both effective and cost-efficient. This successful collaboration resulted in a tangible and measurable improvement in the company's financial performance, demonstrating the potential of AI to drive positive business outcomes when coupled with human expertise.

About Synaptiq

Synaptiq is an AI and data science consultancy based in Portland, Oregon. We collaborate with our clients to develop human-centered products and solutions. We uphold a strong commitment to ethics and innovation. 

Contact us if you have a problem to solve, a process to refine, or a question to ask.

You can learn more about our story through our past projects, blog, or podcast.