Maximizing EBITDA with A.I.: Focus on the Outcome

by Jon Steele

Maximizing earnings before interest, taxes, depreciation, and amortization (EBITDA) can be achieved if retailers stop trying to understand their customers, and start focusing on the outcome.


The idea that a retailer must “understand” each customer and their relationship with a product in order to be profitable and competitive is antiquated. In fact, I would argue it is likely causing more harm than good for those stuck in a bygone era. As we’ve moved into a transformative digital era, retailers no longer need this deep understanding of what drives customer decisions. Instead they need to rely on machine learning (ML) to do it for them.

Five to ten years ago there was a more literal meaning to that notion of “understanding” your customers. Analysts would observe customers through purchase behaviors via point of sale (POS) or loyalty programs, quantitative market research studies, ethnographies and direct interviews, and many other techniques. The outcome of this activity would be distilled into segmentations (demographic, psychographic, lifestyle, etc.) or “path to purchase.” The end goal, although sometimes lost in the fervor of these exercises, was to affect business changes to drive higher sales, margin or market share. To do that effectively, this customer understanding would then need to be married with past actions of the retailer: e.g. What was on the front of the flyer last year, what stock keeping units (SKUs) performed well or failed when prices were reduced in-store, and so on.

To facilitate this process, an old and withering approach of dumping data reports from a patchwork of legacy systems is required and then a human being would be tasked with predicting the future. For example, they would need to predict the impact of putting a particular brand of cereal on sale so that mom would shop at your store instead of the competitors’ and maybe even grab ancillary products, like milk and peanut butter, while she is shopping. The depth of the sale price, the promotional mechanic, the SKU(s) chosen, the timing, and duration all require dozens of hours of high-cost human resources, analysis, and product supplier negotiations. This process then repeats itself in weekly, bi-weekly or four-weekly periods throughout the year.

The problem surrounding what products to promote and how to promote them, has just recently come into stark focus. According to Catalina, the money invested to support retailers inputting their goods on promotion is the second highest expense trailing behind the cost of goods sold. They also note that up to 70 percent of all U.S. retail sales are affected by one kind of promotion or another. The market research group IRI found that over 50 percent of sales in supermarkets were on promotion in the UK at a time. These promotions add up to trillions of dollars in retail investment per year - clearly a lot of money at stake for retailers to get their mass promotion planning right.

Retailers are investing trillions of dollars to change customer behavior in their favor. The majority of that investment is coming from thousands of humans trying to “understand” millions of other humans. This is absurd.

Today, technology has made customer understanding not only fanciful, but no longer required. In an age where data storage is practically limitless and cloud-compute power is widely available, the savvy retailer of today should be entirely focused on the “outcome” and not the “understanding.” The advantage of trying to affect millions of humans with promotion mechanics on millions of SKUs is that machine learning (ML) has a veritable wonderland of data to do that understanding for you. Additionally, ML can deploy that understanding automatically into the infrastructure you already have in place. Those human resources dedicated to understanding can now be redeployed in other areas of need, such as supplier partnerships, strategy development, or more efficient buyer teams.

Some will tell you integrating machine learning into your systems is easy - send promotionally optimizing robots into the business, and the job is done. Of course, that is not the case. There are many hurdles in order to achieve that, including consolidating data across silos, deploying automation and designing the right data science models. Another constraint and one of the most daunting factors to consider is what to do with the very expensive technology that is already in place. All enterprise retailers have a JDA, SAS, SAP, etc. system that buyers and analysts are comfortable with and use regularly. The problem is that many of these are not capable of expert system machine learning that has a retail-specific focus. Additionally, the machine learning or artificial intelligence capabilities they do have are not fully effective because their maximum computational power is not leveraging the computational flexibility afforded by the cloud. What this also means for the business is that algorithms of old are not evolving based on your business needs in an actively changing and highly competitive retail environment or, worse yet, the outdated data science methodologies remain stagnant in the absence of needing to make expensive upgrades.

A common challenge to integrating ML into your system is simply having the courage to embrace ML and AI solutions that deliver the “outcomes” needed. Those retailers with the fortitude to skip over understanding and move right to outcomes will be those who win. The culture of your organization must have the strength to start deploying ML in ways that test the effectiveness quickly and deploy rapidly. You need to be willing to do this in the face of countervailing traditions that you deal with internally. IT leadership may be wary of the cloud, buyers and analysts will want to trust their gut (stack ‘em high!), and enterprise technology suppliers will lock you in with their contracts and pat your head with promises of changes that are coming. But this doesn’t need to be your reality.

Automation and autonomous systems are taking over traditional human tasks like driverless taxis or self-serve checkouts. A great deal of money will be saved through these efficiencies. With expert systems in ML now emerging, the time has come for retailers to relegate select intellectual tasks, like their mass promotional planning efforts, to cloud-based intelligent decision automation delivering “outcomes” that drive profitability. The time is now and we all know that time is money.

Jon Steele is Vice President of Business Development at Rubikloud.