Every so many years, a new supply chain terminology takes the front page and dominates the conversation in magazines and conferences. In the last decade or two we’ve seen JIT (Just in Time), TQM (Total Quality Management), 6 Sigma, S&OP (Sales & Operations Planning), Lean, Agile, Demand-Driven Supply Chains…to name just a few.
Often it is for good reasons that these concepts draw attention, as they have proven to provide value in certain companies or industries. What frequently happens next is that the marketing machines of commercial expert analysts, research institutes, consultancies, and IT vendors run overtime to capitalize on them. Usually the result is that the supply-chain concept is generalised as a solution for everything, accompanied by a simplified interpretation of the complexity involved in implementation. Risks and costs of implementation are hardly ever mentioned, only the benefits. On top of this, a sense of urgency and fear of being left behind is created, often pressuring companies to decide rapidly and adopt the new concept.
Big data has been one of the concepts to dominate the supply-chain scene for a while. There is no denying that we’re getting more connected, we consume more and more data, and we produce exponentially more data, but like any concept, big data is not a holy grail for every supply chain. In a recent article in Foresight, Shaun Snapp describes how Blueridge, a supply chain planning software vendor, argues that big data will be used to switch forecasting from the product and more towards the customer. This might be true in some cases, but in many supply-chain cases it will not.
Some industries are more supply driven rather than demand driven. We’ve recently seen the oil industry creating a glut in oil, whilst knowing that oil demand is hardly increasing. Similarly, in the mining industry, production often continues to pay off fixed asset costs, even when demand stagnates or drops. While these industries might use big data for exploration or other purposes, they are unlikely to use it for customer forecasts. Supply-driven industries focus less on the customer and so hardly require a customer-driven forecast. On top of this, in some industries, the ability to respond to customer demand is limited. In the agricultural industry, connected technology that produces masses of data is used for pest control. However, a tomato plant will produce tomatoes for nine months, regardless of customer demand. This limits the need for big data or customer based forecasting. The value of big data also depends upon where in the supply chain a business operates and what the influence is on the end-to-end value chain as well as the final customer. If a business has limited influence on the these fronts, customer data will not add value to its forecasts. A contract manufacturer that supplies a car or a phone manufacturer who is dependent on a yearly tender doesn’t need big data.
THE CUSTOMER-ORDER DECOUPLING POINT
To understand the potential of big data in forecasting for a supply chain, let’s start with a basic understanding of the customer order decoupling point CODP. This is the point in the value chain where the product is linked to a customer order.
In an Engineer-to-Order (ETO) environment, a company will work with the customer to design and make a product—for example, luxury yachts or specialized machinery. Supply chains with this type of CODP have long order lead times and only a few customers.
In a Make-to-Order (MTO) environment, the customer product is made from raw materials, parts, and components. The commercial airline industry operates in an MTO environment. And although some jet engines produce 10GB of data each second and apply artificial intelligence to optimize fuel consumption, there is no need for big data in customer forecasting. ETO and MTO environments have few customers. This is similar for companies that sell services. SpaceX, that delivers commercial payloads to space, has only a few customers. Big data might be useful for these companies in other areas, but not to forecast their customer demands.
In a Make-to-Stock (MTS) environment, a manufacturer will produce physical products to be held by a wholesaler or retailer, who then sells them to the final customer. The manufacturer might support 100,000 customer ship-to locations like DCs, retail outlets, hotels, hospitals, or restaurants. There is value in understanding this customer demand through point-of-sale (POS) information and including this in your forecast. Using POS data isn’t new in forecasting. Over a decade ago, I included POS information from retail stores four times a day in the production forecast for a meat manufacturer. What’s a more recent development is to apply demand sensing—automatic algorithm changes to short-term forecasts—to every shipto and product combination. To apply demand sensing to 100,000 ship-tos that all hold 1,000 SKUs, an algorithm needs to work through 100 million combinations. It is questionable to call this scale of information big data.
The final decoupling point is to sell from stock. This is the closest point where the order is linked to the customer, who is also the final customer. Traditionally, this is the brick-and-mortar retailer. The largest retailers collect significant amounts of data from their POS and customer-loyalty cards. This data collection can be across tens of millions of customers; however, data collection is still restricted to the geographical network, the number of retail outlets, and the customer behaviour within the retail outlet. The online retailers of today don’t have these restrictions anymore. Amazon has over 300 million active accounts. Account holders can be anywhere in the world; every smartphone or laptop is a retail outlet, and online consumer behaviour can be followed whenever the consumer is online.
Also in the last decade traditional MTS manufacturers started to apply other distribution models. Nike once produced running shoes with an MTS model, supplying retailers. Now it sells online direct to customers from stock and also gives customers an Assemble-to-Order (ATO) option to design their own shoes. So, the traditional MTS manufacturers that started online stores are now online retailers. If you’re an online retailer that’s planning to grow to dozens of millions of customers, big data seems to get more relevant to understand your customer behaviour and include it in your forecast. But where big data really starts to play a role is beyond the point of purchase, after the customer has bought a product or service. Once we enter the customer’s daily life in real time and start connecting different products and services across multiple industries, the available data explodes.
BEYOND THE POINT OF PURCHASE
With over a billion customers, online services such as Google and Facebook are tracking their customers’ every move to forecast the content the customers want to see or places they want to go. Facebook uses AI to curate your content based on your historic behaviour. Google’s Moves mobile-phone app tracks your daily movements and makes suggestions on where you might go next. In a similar way, the retailer Amazon has entered the customer’s living room with its popular voice-controlled “smart speaker” Echo. With Echo, Amazon will create an understanding of your grocery shopping list, the music and radio stations you listen to, the time you come home and switch on your lights, and your voice-directed Internet searches, amongst many other things. Amazon can combine all this information with online purchase history of their products and services and it has patented anticipatory shipping, a system that forecasts and delivers products before the customer places an order. Applying anticipatory shipping to a 300-million-and-still-growing customer base sounds like big data territory. Through the creation of partnerships, traditional manufacturers and retailers can also go beyond the point of purchase to gather consumer information. Besides selling running shoes, Nike partners with Apple to sell sport watches that measure all types of real-time consumer health data. If Nike has access to this data, it can now connect online shoe-purchase behaviour with an understanding of workout regimens, sleeping patterns, heart rate, and kilometres run for their customers. Nike also partners to sell watches with TomTom, a Dutch map provider and consumer-product manufacturer. Besides producing sport watches, TomTom provides the maps for Apple’s 600 million iPhone users and for many major car producers. If these three companies, who are already in partnership, were to share all their customer data, they could in fact know their customers’ every move—or very nearly—and use this to predict their behaviour. Although privacy and data ownership play a significant role, when manufacturers and retailers across industries team up to gather information beyond the point of purchase across multiple products and services, data availability to forecast customer behaviour explodes.
THE BIG-DATA DECISION
Big data seems to be most relevant for demand-driven businesses with a significant customer base, and even more for businesses that can gather customer information beyond the point of purchase and enter the daily lives of the customer. Even then, the investment in big data seems more a holistic, strategic business decision, rather than a supply-chain or narrow forecasting decision. The analysts and researchers that push big data should provide more context in what industry and supply-chain situation big data will most likely function and should describe the risks and benefits. Before adopting a new concept in their value chain, it’s the supply-chain executives’ responsibility to ask themselves several questions, including:
- Does big data create efficiency in our operation?
- Does it deliver value to our customers?
- Does it give us a competitive advantage?
- What is the risk of implementation versus the return on investment?
Investing in big data is a major decision that should not be taken lightly, especially on the advice of analysts and researchers who have an interest in selling big-data solutions.
This article was first published in Foresight. Image credit http://bigdataanalyticsnews.com/