2020 has been quite a year for all of humanity. We’ll likely be recounting stories about the pandemic in the same way that we talk about World War II, September 11, and other life-altering world events.
Arguably, the retail industry has faced some of the most significant pandemic-driven impacts in the shortest amount of time. Before the pandemic, e-commerce accounted for 13% of the retail sales in the United States. Given store closures and social-distancing measures, we can safely say that percentage has now more than doubled. For example, Stripe, an e-commerce payments platform, went from handling $1 billion in payments last year to more than $10 billion in transactions in the first six months of 2020. A potential 20x growth!
In many cases, consumer shopping habits may have changed for good. Take for an example how my own family buys groceries. We have gone from shopping solely in-store (Trader Joe’s, farmers market, and Whole Foods) to getting groceries delivered to us each week from Whole Foods (owned by Amazon) and Costco (delivered by Instacart). I’ve noticed that (after a rocky start) the quality of service provided has improved significantly over the course of this year: from having to wait up to a week for a delivery slot for Whole Foods to having food delivered to my doorstep in less than two hours. The question my family—and many others—is now asking is this: Will we ever go back to the grocery store? Is it worth it to lose two hours of precious weekend time to shop in a physical store?
Another consideration for retailers is the Amazon factor. Retailers and brands have to figure out how to coexist with and thrive in an ecosystem dominated by Amazon. It is possible! Let’s take the example of one of my favorite coffee brands: Equator. The company—which began in 1995 as a small operation out of Mill Valley, California—is one of the most popular gourmet coffee brands on Amazon. Recently, I switched from buying the product on Amazon to buying directly from their webstore. When I got my first shipment, I was pleasantly surprised to discover that the coffee had been roasted only the day before. Ah, the joy of a fresh roast!
Do you see how Equator coffee is being very shrewd about its e-commerce strategy? They’re fully present on Amazon, with a complete store front, but they also give customers who buy directly from them something extra. The consumer can choose between the convenience of Amazon or something special on the product/service side by buying directly from Equator’s website. I call this having a bimodal channel strategy.
This rapidly developing e-commerce environment poses many challenges that are likely causing supply chain leaders’ heads to spin and are keeping them up at night. In many cases, artificial intelligence (AI) solutions can help them navigate those new challenges. We like to think of AI as a catchall term to capture the idea of solving problems with algorithms and data. The algorithms could be from deep learning, machine learning, operations research, or another approach relevant to the problem. Let’s consider some of those challenges and potential solutions below.
Time to redesign your supply chain
Your current supply chain is probably optimized for a pre-COVID world—both in terms of the kind of products and services you offer and how you deliver them (for example, through a physical store).
You now need to rethink the smartest way to restructure your supply chain to fit the new reality. For example, with current e-commerce volumes, what delivery terms will you offer: same day, next day, or two-plus days? If it’s same day or next day, you’ll likely need to set up a network of dark stores or local warehouses—but how many and where?
There may also be other structural questions: What level of service will you offer for which product assortment? Can you segment customers based on ideas like customer lifetime value? Will you fulfill from stores? If so, how will you change your store merchandising and replenishment strategy?
These are all classic supply chain design questions that need to be addressed when structuring your supply chain to support this manifold increase in e-commerce volumes. Commercially available supply chain design software can help with the answers, particularly when used in combination with models to analyze customer lifetime value, product affinity (which products are likely to be ordered together), and other factors.
In response to an increase in online orders, one retail client recently accelerated the implementation of its e-commerce strategy, doubling the number of e-commerce fulfillment centers (FCs) from three to six. To determine the optimal locations for these three nodes, the team relied upon supply chain design software. In addition, they used tailored AI models to predict product affinities and an optimization model to determine stocking strategies. As a result, the team was able to configure the company’s stocking strategy to maximize the percentage of shipments sent from the closest node to the customer while also minimizing the total number of split shipments.
A new approach to capacity planning
If you have retail fulfillment centers (FCs), I am guessing you are constantly running into capacity issues. This could be due to seasonal spikes in demand, marketing promotions, a surge in inbound volumes during certain times of the week/season, temporary labor capacity constraints, or some other unforeseen reason.
AI can really help here. Prediction models built using deep learning techniques can help you understand the right volumes (units and orders) to process per day. (These models will need access to data around such things as your website activity, customer loyalty, historic transactions, and promotional activity.) Meanwhile optimization models can help match demand with the supply of capacity available in the system in an efficient manner.
FC managers often find these combinations of prediction and optimization models to be an upgrade over their current planning capability, which is typically a spreadsheet-based system or a workflow-based planning software provided as part of their enterprise resource planning (ERP) stack. The new generation AI-based planning solutions are very effective in alleviating order backlogs and helping set the right service-promise expectations with consumers. Additionally, these models can also help the business shape demand through controlled promotions, digital marketing, and more.
The secret to getting good results from these AI-based solutions is twofold. First you need to have a rich trove of data. The more data you can provide to make the system more intelligent, the smarter the model predictions are. Second, you need to use modern AI algorithms that can identify hidden patterns in the data and leverage them in the predictions.
A large Fortune 500 retail company found its e-commerce business was growing at a pace of 40+% year-over-year. The company was reluctant, however, to take the capital-intensive step of adding fulfillment capacity to its supply chain. Instead, the CEO wanted to explore whether a AI-enabled software solution could alleviate the problem. A tailored AI solution was built to help the company predict if it was going to run into capacity issues, whether due to a seasonal surge in demand, a promotion, or a shortage of labor. An optimization model then followed up these predictions with a suggested action plan, such as hiring additional labor, shaping demand, or deactivating certain promotions.
Once built and implemented, the business found the solution to be so dependable that it built its entire integrated business planning process for e-commerce around it. In addition, the operational efficiencies gained through the solution have allowed the business to postpone building a new fulfillment center by at least two years.
Root cause analysis of failures
No matter how well you run your e-commerce business, the sheer volume of daily orders processed inevitably means you will face some number of failed orders each day.1 This could be true for one of several reasons. Maybe you received a disproportionate number of orders close to the cut-off time, or maybe too many high-value orders got stuck in the fraud check process, or perhaps a disproportionate number of them required split-shipments. There’s a whole host of triggers.
When orders fail, you want to avoid a “blame game” between the various operational teams. The hard thing is that when orders fail, there is a waterfall effect that makes it very difficult to understand what really caused the failure based on simple data analysis.
This is where a prediction model that is trained to detect the root cause of these order failures can be very helpful. These powerful models can elegantly and efficiently inform you why an order failed, providing more granularity than any manual approach could on its own. Using this method can help put your business on a path to continuous operational delivery improvement. Additionally, the next level of evolution for these models is to have them tell us which orders are likely to be delayed before the failure takes place. This information can be used to expedite orders, inform the customer about the delay in advance, or determine possible workarounds.
One of the largest apparel and athleisure companies in the world has seen significant value in using AI solutions to help with detecting delivery failures. The company has seen benefits both in terms of improving consumer satisfaction scores and creating operational efficiencies. By leveraging this solution, supply chain managers can consistently uncover the true root causes of e-commerce failures and even predict them before they become a problem.
Smart inventory cleansing
E-commerce has this uncanny ability to proliferate your product portfolio—mostly because the business is no longer constrained by physical store shelf space. However, while this proliferation may be tempting, it is not healthy. You will end up holding a lot of inventory in your fulfillment centers, tying up both your working capital and precious capacity. It is therefore important to cleanse your system of “nonproductive inventory.”
With AI, instead of just using past data to make these inventory-cleansing decisions, you can build predictive models. Once these models are trained2 and back-tested,3, they can help you confidently decide which products you can continue to store (and where) and which you need to liquidate through your regular liquidation channels.
A fashion retail company recently faced an unexpected abundance of inventory due to COVID-19–related store closings and new product shipments not having anywhere to go. Instead of relying on human intelligence, which could be both biased and hard to scale, the chief supply chain officer asked the team to build a machine-learning model to make these decisions.
His decision proved to be the right one. The machine-learning model was relatively easy to build, as almost all the data needed was readily available. The model also was scalable, and (once the stakeholders understood that it was good at predicting which products were most likely to be unproductive) there was wide-scale adoption of the solution. The business is now committed to enhancing the solution with additional features (such as bringing in prediction around downstream liquidation revenue) and is expanding the scope of adoption to other divisions.
The pandemic has hastened the adoption of e-commerce across the world in an unprecedented manner. The challenge, of course, is getting the business comfortable with the rapid change of pace that we are all experiencing. At the same time, this change also presents us with a huge opportunity to make our businesses more AI savvy. E-commerce is inherently a more digital process, which creates data: the fuel for AI systems. E-commerce also demands that the business be highly scalable, which is not feasible without a mindset to automate every process.
What we discussed in this article is just scratching the surface on using AI in your business. Observing the market and being open to new and powerful AI solutions can enable you to (a) be ready for longer-term e-commerce dominance, and (b) start using the technology to run smart e-commerce operations. My advice: Strap yourself in for an exciting ride!
1. A failed order is not just the order that you cannot fulfill due to lack of inventory, but also the order that the customer does not receive at the level of service you promised.
2. “Training” is a term used in machine learning to describe building a model specific to a certain problem and/or dataset(s).
3. “Back-testing,” or “testing,” is a term used in machine learning to describe the process of testing a trained model against past data to understand how good the model predictions are likely to be.
Ganesh Ramakrishna is a senior vice president at LLamasoft Inc.