Share this post:
I’ve always sought out challenging applied research problems and am particularly attracted to applications of technology in the social sector. That’s why I was thrilled to work on a supply chain-focused project with the non-profit organization St. John’s Bread & Life (SJBL) within IBM’s Science for Social Good program in 2017. Based in Brooklyn, SJBL is a private agency working to alleviate urban hunger and poverty by providing emergency food through a digital-choice pantry (serving 150-200 families a day) and a full commercial-grade kitchen plus dining hall (serving 700 hot meals per day), complemented by an array of community social services.
St. John’s Bread & Life operates in a beautifully refurbished former furniture factory in a Brooklyn neighborhood
Our team helped SJBL identify and bring to bear several supply chain performance metrics, which are standard in the for-profit world, but nascent in the non-profit space. Over the year-long project, I was able to apply my experience doing supply chain scenario modeling for wholesale distribution, including project that was a 2013 finalist for the INFORMS Franz Edelman award1. It was extraordinary to go from saving a gigantic pharmaceutical company over a billion dollars, to helping a small non-profit understand its own best practices2.
We were interested in applying data science to SJBL’s extensive digital history and making use of quasi-structured data. To better understand the organization and its inner workings, I volunteered across SJBL’s operations. I picked and packed food items in the food pantry, served food in their dining hall and helped prepare food in SJBL’s commercial-style kitchen. To fully understand the extent of the skills required by SJBL staff, I took the course and written exam to pass the New York City Department of Health’s Food Protection Certificate required of staff supervisors. The course helped me understand the planning, food storage, handling, and item usage decisions they make, which are often manual, but sometimes assisted by online historical data and patterns.
This direct observation and hands-on participation boosted our data science methodology, in terms of mining significant and meaningful patterns from more than four years of detailed historical data, comprised of millions of event records. After “normalization,” which placed the SJBL historical data into a more “regular” supply chain data modeling format (e.g., by introducing “stock keeping unit identifiers” and by organizing the event data in terms of demand, fill, supply order, and receipt time series), we were able to effectively model the SJBL food supply chain, both historically (i.e., a “trace simulation”) and conceptually (for “what if” analyses).
Examples of stock keeping unit (SKU) categories found from classification applied to over four years of SJBL’s food pantry event history
Using classification techniques, we were able to isolate four clear patterns of supply-and-demand: regular items, both from a demand and supply availability perspective; semi-regular items, which are available in interval spurts where they sometimes resemble regular items; sporadic and rare items, which show up every now and then, and are usually unplanned and sometimes large, commercial donations. A favorite example of the later are the duck eggs donated once in a large volume that were very popular.
Our study helped SJBL identify and quantify their supply chain performance metrics so they can share best practices with similar organizations.
For example, we discovered and proved through historical data that SJBL’s supply chain for both the food pantry and kitchen resembles a retail grocery store and restaurant operation, respectively.
Prototype functionality developed to demonstrate the data science, our cloud-based engines and API
The digital food pantry hums in its supply chain operations. In fact, from SJBL’s rich digital history, we were able to accurately measure its inventory turnover — a standard performance metric used by for-profits to measure efficiency — and found that SJBL rivals some of the best for-profit retail distributors and grocers.
SJBL’s service level attainment rates (i.e., fill rates for food items and delivered community services) are competitive, and their working capital, upon evaluation of their entire warehouse inventory over more than four years, is within 22 percent of “optimized” based on advanced decision models.
All of these findings are impressive, especially considering that SJBL’s operations management decisions are primarily made with data from a simple, yet effective digital fulfillment system.
Our team also created a cloud accessible version of the SJBL supply chain data model, as well as engines that illustrate the use of the data for performance analytics and “what if” analyses. The API access is intended to eventually allow other organizations to explore SJBL’s historical data and to further our findings.
St. John’s Bread & Life: Anthony Butler, Al Diefenbach, Candyce Mason, Marie Sorenson, and Carolyn Tweedy
IBM Research: Mary Helander, Lei Kuang, and Pietro Mazzoleni
1K. Katircioglu, R. Gooby, M. Helander, Y. Drissi, P. Chowdhary, M. Johnson, and T. Yonezawa (2014). “Supply Chain Scenario Modeler: A Holistic Executive Decision Support Solution.” Interfaces. 44(1): 85–104. DOI: 10.1287/inte.2013.0725
2M. E. Helander, L. Kuang, P. Mazzoleni, A. Butler, A. Diefenbach, C. Mason, M. Sorenson, and C. Tweedy (2018). “Findings From Modeling And Analyzing A Non-profit Organization’s Emergency Food And Community Social Service Operations.” INFORMS 2018 Business Analytics Conference, Baltimore, MD, April 15-17, 2018. http://meetings2.informs.org/wordpress/analytics2018/