What if you could 

• Reduce the time required to get new sales representatives selling effectively and closing deals 
• Sell complex configurable products through your e-commerce channel with less sales assistance 
• Leverage ML/AI to streamline sales and quote processing without the need for a data scientist 
• Strengthen customer loyalty and trust by intelligently recommending the products and configurations consistently fulfill the customer’s needs 

You can. Check out this blog series to learn how. 

Background

All the recent excitement about ChatGPT has brought to light the power and ease with which machine learning (ML) and artificial intelligence (AI) can be used in everyday life to do many interesting things (like help write a blog). We at SAP have known the power and value of ML/AI for quite some time and have been working to identify scenarios and use cases where ML/AI can be applied to solve business problems and drive significant business value. One area where ML/AI is particularly relevant is around product configuration in the manufacturing arena. This blog series will focus on how ML/AI can be used to complement traditional configurators to enhance overall experience for users. The series will start with a description of the business challenge, then describe how ML/AI can be applied to reimagine a business process and will culminate in a third part that will include an exciting announcement. I hope you enjoy this blog series.

What is the challenge facing manufacturers?

I can go onto any sales marketplace, select one of thousands of products (like a blender or toothpaste) buy it, and have it delivered to my house in a matter of hours. I can even go onto various car buying websites, select a car, buy it, find financing, and have it sitting in my driveway in a few days. However, if an enterprise buyer wants to buy a pump, robot, forklift, or any other industrial product, they typically must work with a sales representative for days or weeks to get a price and lead time quote. Why is that?

When buying simple products like a blender, toothpaste, or a simple configurable product like a laptop computer, it is easy for the buyer to determine a suitable product that meets their requirements. Also, the user can make the option selections to configure the product because there are typically only a few different options, and they are straightforward. And if the user picks the “wrong” products or configures “wrong” and the needs are not met, it’s usually quite easy to exchange. There is little risk if the buyer “gets it wrong”

However, for many manufacturers of more complex configurable products, the process is more challenging. First, making the linkage between the needs and the product is much more difficult to determine because the needs are typically much more complex, and the products or often very complex with different configurations, etc. The problem is exasperated, because the configuration model used in the product discovery, selection, and quotation process (a process that could be performed by a non-technical buyer) is sometimes the same model that is used for sales order processing and production (a processes that are typically performed by more technical oriented users trained t use IT systems).

Over the years, many manufacturers have been using configurators to streamline the product configuration process that is typically performed during sales order processing. The product configuration models used by these configurators are usually developed by technical, product engineering-oriented people that have deep product knowledge. These models are often defined by concrete physical or technical attributes of the product and therefore lend themselves to rules or arithmetic computations where outcome can be easily validated. Does it work within the engineering limits, can it be built? The goal of the configurator is typically to ensure that the resulting configuration is technically feasible and can be built. This is important especially in higher volume manufacturing because the objective is to reduce costs and customer lead time (even for market of one customized products) by allow customers’ orders to go from order creation, right into production with little or no human intervention (especially from engineers). Although the configurator will ensure technical feasibility, it does not necessarily guarantee that the configured product will meet the customer’s needs. This may sound like a trivial, semantic nuance, but rest assured it is not. Allow me to explain why.

Because the configuration model and underlying configurator is already available and will always result in a technically feasible configuration and can also generate a selling price it seems ideal suited to be used during the sales process. Additionally, it lends itself to faceted search (often used during product filtering in e-commerce and/or CPQ applications) because some of the characteristics can be used as filter criteria. However, the challenge is that many salespeople and certainly the customer, may not be familiar with the technical attributes that are identified product technically oriented product modelers. Therefore, a sale person or customer needs help identify or translate the technical attributes into the customer’s needs. As a result, the salespeople need to have deep technical knowledge of the products and features. This typically requires a significant level of training and/or expertise gathered over many years.

Also, a sales representative is typically involved in the process to reduce the chance that the wrong product and configuration is selected. This is necessary because the risk is greater if the buyer makes the wrong selection. At a minimum, you may not be able to return the product because it is customized to your needs (not to mention it may have cost hundreds of dollars to ship it), but worse, it could be dangerous to humans if you get it wrong. Therefore, traditionally, you had a well-trained sales rep (or dealer/partner) in the middle that translates the customers’ needs into the manufacturer’s product.

Let’s look at a simple example of a warehouse manager looking for a new forklift. The traditional buying experience is depicted below:

In this example, we have a warehouse manager in the market for a forklift. On the left side of the image above, the needs for the warehouse manager are listed. These attributes represent the way the warehouse manager will think about the job to be done. On the right we have the offerings from the manufacturer. Their offering includes numerous products that have technical characteristics that need to be selected to ensure the correct configuration which could include millions if not billions of different combinations.  

Needs are defined independently from the product (e.g., forklift). They are defined by the entity (or person) that has the needs (job to be done). In our example, the warehouse manager’s job to be done, is to move pallets in the warehouse. To do this job he/she has many considerations/needs including:  

• How heavy are the pallets, how big are they, if need to put/remove pallet from a rack, how high is the rack, how wider are the aisles that can move within 
• How many pallets typically need to be moved 
• What is the operating environment – dirty, noisy, flammable liquid, inside (temperature controlled) outside (rain, snow, salt air, freeze) 
• Does the warehouse manager have personal or corporate goals like reduce CO2, minimize energy usage, lead time, price, etc.?

These needs are typically completely independent from the solution (e.g., forklift). In fact, in some cases, these needs could be met without a forklift. If the pallets are small and not too heavy, and the racks are not too high and don’t need to move too many, a person (or people) could do it! 

It is up to the sale representative to bridge that gap. The sales representative would typically speak to the warehouse manager, understand the needs on the left, and then know the product portfolio recommend the best product and configuration to meet the customer’s needs. Even with a configurator, a sales rep. is often required because the characteristics in the model are often product feature oriented and, in many cases, very technical.  

 A challenge that we often hear from many manufacturers is that many of their experienced sales reps are leaving, retiring, etc. The problem is that it takes lots of time, training, and experience to get new sales reps up to speed (sometimes as long as 9 months to a year). Additionally, manufacturers can more easily find salespeople in their local markets, but when they try to expand to other markets throughout the world, it can be very challenging. As a result, manufacturers run the risk of having inexperienced sales reps that may recommend products and configurations that are not ideally suited for the customer’s needs. Also, because in many cases, there could be multiple products and configurations that could meet a customer’s need, there is often inconsistency in the recommendations, this negatively impacts supply chains planning.  

Where ML and AI can help, is by helping to bridge this gap between needs and the product and configuration. We can train an ML model based upon the historical quote, opportunity, order data to pick up the patterns and correlations between the need’s attributes and product and technical attributes. This would “codify” the experience of the salespeople by analyzing the historical data. Once the model is trained and deployed it can help the sales rep by acting like an intelligent assistant. And, for e-commerce scenarios, it could allow manufacturers to sell more complex configurable products through the e-commerce channel. 

Stay tuned for the next installment of this blog to learn how machine learning and artificial intelligence can enhance needs-based product selection and streamline sales and quote processing for and drive significant value for manufacturers.