Omnichannel Retailers Leverage Optimization for Order-Promising

Historically, omnichannel software solutions were not capable of optimization. This meant that retailers were not maximizing their profitability. That is changing.

Omnichannel refers to the ability of a consumer to get products fulfilled to them in various ways. A customer can of course buy goods in a store. They can also order online and have goods shipped to their home. Or they can order online and pick up the product at the store. There are several other fulfillment and product return options as well.

An omnichannel order management system helps companies meet growing customer expectations. For example, Amy Tennent, a senior director of product management for Manhattan Active Omni at Manhattan Associates, says that in Manhattan’s newer omnichannel implementations retailers want detailed delivery and pick-up information to be displayed on the product details page of a retailer’s e-commerce site rather than after a shopping cart is filled. Providing this availability information earlier leads to fewer cart abandonments and less consumer frustration.

An omnichannel order management system allows an organization to execute order fulfillment across all channels and fulfillment flow paths. Historically, an OOM has access to all the company’s inventory. The system allocated which orders would get inventory from a given location to fulfill an order.

Rules were used to make these decisions. For an online order that needs to be delivered within 24 hours, for example, the rule might state that the goods should be shipped to the consumer from the closest location. These rules can be put into a hierarchy. For example, look to fulfill from the closest store location to the consumer first, but if not all the ordered items can be fulfilled from this location, then fulfill from this the e-commerce warehouse. The rules hierarchies in these systems can get quite complex.

The problem with the rules that retailers use for omnichannel fulfillment is that they leave money on the table. The hierarchy of rules is too simple to fully optimize profitability.

Now, however, a couple of software companies, Manhattan Associates and Blue Yonder, have introduced optimization capabilities into their omnichannel solutions. These more advanced solutions leverage network inventory visibility and demand forecasting models and constraint logic.

Constraint logic involves the physical capabilities of a supply chain. If an order is to be fulfilled from a warehouse, for example, the Manhattan Active Omni solution can be configured so it understands whether that warehouse has the necessary labor to fulfill the order by the cutoff time. Or if a shirt needs to be monogrammed, the solution understands that it will take an extra 8 hours to do that and only one of their warehouses can do monogramming.

It is no accident that the two OOM providers that can do omnichannel optimization also provide supply chain planning solutions. Supply chain planning solutions have made use of optimization algorithms for decades.

Optimization for Order Promising

When Blue Yonder or Manhattan Associates looks to optimize order promising, they are talking about both providing an accurate promise date to a consumer as well as the optimal location, based on maximizing profitability, from which to fulfill an order.

Blue Yonder has a microservice they call Commits and Fulfillment Optimization. One would assume when it comes to maximizing profitability that if a customer wants to order five items for shipment to their house, and the retailer only has two of those items at the store closest to the consumer, but all five items are located in a more distant warehouse, then it would almost certainly make sense to ship from the more distant warehouse. This is not always the case. As a fashion product approaches the end of its life, for example, demand models can be used to calculate how much to mark that item down to get it off the shelf and when those markdowns should begin. When you combine markdown optimization machine learning with order fulfillment, the optimal answer may be to sell the last two items at a store and incur extra shipping costs because the extra shipping costs are less than what the markdown would soon be.

Manhattan Associates also optimizes order promising based on demand forecasting in a similar manner. However, Ms. Tennant points out that the way different brands think about leveraging inventory in their network varies significantly. “So, it’s key that the algorithms that we are exposing within the Active Omni product have functionality and machine learning in the context of different types of businesses.”

Machine Learning and Accurate Promising

Retailers don’t want to over-promise and under-deliver. So, the parameters they use in making their promises are conservative. The pickup time by carrier, transit times, and estimates about how many items a warehouse can pick per shift, are conservative to ensure that promises can be kept. But there is a cost to this. It can lead a retailer to promise something will be delivered in four days when it really could have reliably been delivered in two. This can lead customers to abandon their carts and go elsewhere. Among a sample of Manhattan Associates’ omni retail customers, 85% of the time the orders could have been accurately promised 1.5 to 2 days earlier.

Manhattan Associates is using machine learning to accurately parametrize the order promise. By looking at the historical data on an ongoing basis on what is achieved – by carrier, by lane, by mode, by pick rate at a fulfillment location, by day of the week or season of the year, and so forth – the system can be autoconfigured to make accurate promises for earlier deliveries.

Blue Yonder’s Srinivas Pujari – the corporate vice president of product management for Blue Yonder’s Commerce solutions – explained that one of the problems with making an accurate promise is that the store’s supposed inventory count is not accurate. Most retailers don’t use scanners to keep their in-store inventory accurate. Many retailers, as a result, have inventory accuracy levels on the store shelf of 90% or less. In contrast, in warehouses where scanners are used, the inventory accuracy can be greater than 99.9%. If a retailer promises a customer that they can deliver an e-commerce item based on the mistaken belief that the inventory is in stock at a store, that retailer will end up with unhappy customers.

But even if a retailer does use in-store scanning to improve their stores’ inventory accuracy, the accuracy will still not be as good as in a warehouse. At a store, the store inventory system uses the point-of-sale system to decrement units from inventory as they are sold. However, the inventory system does not know whether items are in a consumer’s shopping cart. This leads to a situation where the store system thinks it has inventory it really does not. Theft is also more theft at a store rather than a warehouse.

Retailers deal with this issue by not making all their in-store inventory available for sale for omnichannel flow paths like buy online/ship from store or pick in store/pick up at the curb. If the store inventory system believes that they have 20 bottles of two-liter Diet Coke, they may only say that 12 of those units are available on their website.

But what are the rules that should be used to determine the correct inventory promising number to deal with inventory accuracy issues? Ideally, this is a machine-learning problem. Maybe between 8 and 10 are displayed online in the morning, when the store inventory system says that 20 units are in stock. Perhaps, at the end of the day, only 6 should be assumed to be available when the system is saying 13 are in stock.

Manhattan Associates is not as bullish on using machine learning to solve the inventory accuracy problem at the store. “If you recall, ‘safety stock’ was an issue in warehouses years ago, too,” Ms. Tennant said, “but tech and operations fixed that. Stores are plagued with more chaos than a distribution center, but the digital demand for store inventory means they, too, must turn to tech that can track changes in real-time. The use of RFID for cycle counting can greatly improve inventory accuracy.”

Manhattan’s Ms. Tennant remarked that “over the last 20 years, the footprint of order management has expanded and been redefined over and over again.” The inclusion of optimization and machine learning in omnichannel order management systems, however, are among the most interesting and impactful changes to this solution set.