SAP Infuses AI Into Costco’s Baked Goods Business

Home > Research > SAP Infuses AI Into Costco’s Baked Goods Business

SAP Infuses AI Into Costco’s Baked Goods Business

Traditionally driven by a culture of tribal knowledge bolstered by the long tenure of its employees, Costco determined it could benefit from a more precise and data-driven approach to improve inventory accuracy, demand forecasting, and reduction of product waste.

The Problem

Costco’s SVP Jeff Lyons initially asked his CIO for an introduction to “Mr. Watson” of IBM according to Vinnie Mirchandani’s account in SAP Nation 3.0. Costco's CIO redirected the request to SAP as Costco is a current SAP ERP client. Mr. Lyons is seeking to solve a problem within Costco's most profitable business that includes its prepared foods offerings (e.g. rotisserie chicken, beef, organic foods, wine, and prepared meals). Who doesn’t grab a $4.99 rotisserie chicken when visiting Costco?

The problem in need of improvement is the waste that occurs as an inherent characteristic of perishable food items. Prepared foods have a short shelf life, and when inventory levels exceed demand, food, packaging, and labor are wasted. This is a pretty straightforward problem. Additionally, Costco had over seven years' worth of consumer sales data enhanced by a robust forecasting methodology and an impressive depth of tribal knowledge from long-tenured employees. With a median tenure of 4.8 years for all employees, it is not uncommon to see employees with 10-25 years of experience.

Still, Costco inherently realized it was leaving money on the table. How could Costco put that treasure trove of data to good use? As Mr. Lyons states, "As a company, we are driven by what is called the six rights of merchandising, with the top three being the right item, the right quality, at the right price." Costco was not accurate enough in these areas and saw an opportunity for improvement!

Costco is seeking to leverage AI/ML as a potential catalyst to improve its efficiency and accuracy. Costco also sees the value in bringing in external data to overlay atop its internal data. For example, the ability to overlay weather pattern data on top of retail data to discern patterns in how weather events impact consumption by various regions across markets. Or to associate how often customers purchase gas after visiting Costco.

The Solution

Costco invested in an engagement with SAP to embark on a pilot project to see if AI/ML technologies could drive meaningful improvement in the problem areas noted above. After working through a design thinking phase to more clearly identify and define the problems across multiple product lines, Costco moved forward with an ML pilot focusing on the bakery department in one of its northern California locations. The bakery department is an especially target-rich environment as damage and waste are higher relative to many other departments.

To summarize, the bakery employees would enter data on waste, spoilage, and inventory on-hand into a tablet from which the data was used by SAP's ML technology to generate the day's production run schedule. Mr. Lyons provides a telling story that emerged from this process.

In his words, “We told him only ignore the model if it’s totally wrong or it’s obvious that it’s 100% off. One day, the model called for two batches of croissants. He (sic bakery manager) had only planned on making one, but he decided to follow the model’s guidance. He made two batches – that is, 360-some-odd units. The morning after, they only showed nine units left. So two batches was exactly the right thing to do. He wouldn’t have done that on his own.”

So not only was waste being reduced in this example, but revenues were being increased as well. Now, this is also an example where a bakery manager with deep experience was still able to improve top and bottom line results materially. This is also encouraging for Costco as even the most tenured employee will eventually get promoted, retire, or leave the company, so the ability to standardize on an algorithmic model will enable new employees to ramp up productively in a short period of time.

This pilot also incorporated the “time and motion” studies for the bakery processes into the model. This data is used to optimize production operations and provide improved job scheduling for the food processing aspects of the operation.

The Results

Caution: results feedback was generalized and lacking in hard numbers, likely due to the need to protect competitive advantages realized during the pilot. Costco was able to implement the pilot project in three months and has achieved material reductions in labor hours and product waste. This was best characterized by calculating that if the same pilot solution were applied to all stores in the same region, and attained similar results, the annual savings could amount to $8 million.

Results were encouraging enough to yield the second stage of investment that will see the solution implemented at an additional 500 bakeries in the US. Perhaps even more encouraging are the ancillary opportunities that Costco has identified as being able to benefit from applying the same process and SAP ML capabilities. One example is the food court (legendary for the $1.50 hot dog combo), which represents a more complex food distribution process.

Recommendations

Ensure the ROI matches your investment and risk profile for your AI use case. We are missing some critical pieces of information from this case study: cost of SAP software solution, SAP resources, and Costco's internal time and resources. While it is common practice to make a "bet" on the future ROI for this type of project, it is imperative to have a budget and tight management of project costs to ensure that ongoing funding yields results.
A laser-like focus and clarity of mission are required to maximize the chances of success. A deep understanding of your business processes is a critical success factor. Costco clearly understands its perishable food business and the specific areas ripe for optimization. This type of clarity enables a clear communication to vendor prospects to which they can respond with a higher degree of confidence and methods for innovative technology application. A lack of clarity is the equivalent of a financial black hole.
Verify, validate, and repeat. Don’t confuse initial results with a correct IV&V exercise. Tread softly and methodically as you scale up. Initial results may change based on small modifications to the use-case inputs and have effects that ripple across various business processes and environments.

Bottom Line

As organizations seek to do more with AI/ML technologies, look at your incumbent vendors to identify where your respective interests may intersect. This offers the potential to raise the vendor relationship to a more productive, win-win level. Finally, these productive partnerships often carry additional discounts and benefits across the full array of services acquired by the customer, especially if you are designated a "strategic partner."