A window into Comparable Forecasting
New product demand forecasting and planning is exceedingly difficult.
Short and fast-changing consumer demand cycles raise the specter of failure.
New products are usually comparable to previous ones. Applying the right shared attributes will improve forecasts.
When Samsung released the S6 Edge smartphone a couple of years back, the company was painfully unprepared to meet demand. Samsung planners had bet that the handset’s first-of-a-kind design (the sleek curved screen) would appeal to a niche audience, thus limiting sales. Boy were they wrong.
The S6 Edge was a hit for the masses, or should have been. Samsung had a hot product but a skimpy logistical roll-out that was overrun by unanticipated demand. This led to production shortages, shipment problems, lost sales, and probably lost market share.
Meanwhile, Samsung had overestimated sales of its regular S6 handset, a cheaper model whose overstock led to an ill-timed surplus of unsold handsets. Apple and a host of lower-cost producers in China and India were ready, and pounced with their own competitive releases.
Thus, Samsung’s S6 product releases became a textbook example of the difficult, but critically important role, of new product forecasting. The stakes are especially high in technology, consumer electronics, and fashion, where new products abound, and where sales can spike and fizzle in short demand cycles – sometimes just months or weeks long. This makes product forecasting, launch timing, and capacity planning exceptionally difficult. It raises to new heights the importance of dynamic and agile supply chain planning.
It’s important to note that across industries, some 80% of new products fail. Not surprisingly, one of the biggest contributors to that failure rate is unnecessarily inaccurate forecasting. While no system or organization gets forecasts 100% right, the best of the best get them consistently less wrong.
New products are inherently difficult to forecast because they have:
- No direct historical data to extrapolate
- Diverse or sometimes unknown buyer sets
- Harder-to-predict adoption rates and lifecycle spans
The promising, and the proven
In the consumer arena, many supply chain leaders are using real-time or near-real-time historical data as a demand sensing tool, synthesizing the latest sales and other data to gauge true demand. Vanguard’s forecasting & planning platform factors multiple point-of-sale data streams to improve consumer goods forecasts. Other, non-historical, data streams used in demand sensing include weather, economic indicators, and social media. These outside data can be used for additional regression analyses to better sense true demand, such as which new smart phone consumers actually want. In Samsung’s case, better stock-out or social data may have helped.
In any case, before we jump to promising new kinds of demand signals (and we will in subsequent posts), let’s start with some market-proven fundamentals in the realm of historical, time-series forecasting. As it relates to new products, the remainder of this post will deal with Comparable Forecasting, one of several key statistical methods for modeling and simulating new product demand.
About Comparable Forecasting
Again, the major problem with trying to forecast demand and adoption rates for new products is that there is no history to work with – at least no direct history. But often there are similar products with comparable attributes and sufficient historical data to serve as a proxy.
Comparable Forecasting applies the patterns and parameters of similar previous products to the forecasting of new products. You can use it for both initial and later-stage adoption rates. Obviously, it works best if your new product or service is substantially similar to an established, or previously-launched item – or better yet, set of items. With the right technology platform, you can select comparable past products by virtually any attribute, such as:
- Product code
- Product family
- Adoption profile
- Life cycle span
- Combinations of attributes
Some organizations are able to aggregate whole series of similar previous products with which to model forthcoming launches. We’ve seen firsthand the additional improvement that aggregated data sets can contribute to Comparable Forecasting. Independent research bears it out.
A recent study led by researchers from Northwestern and Penn State universities found that Dell’s comparable product data, when properly assembled, could raise forecast accuracy on new products by as much as 9%. That’s huge, especially given that new products account for more than a quarter of sales (on average) for retailers and consumer product makers. The potential returns are also huge: reduced inventory, stock-outs, and emergency shipments, plus higher sales and customer satisfaction.
In Dell’s case, researchers found that by clustering previous products with similar product life cycle (PLC) curves, they could model each new product according to the PLC cluster it most resembled. This improved new product forecast accuracy, as each cluster encompassed an aggregated series of product histories (a lot of data). They combined this with additional information for each new product, such as launch timing, sunset plans, and more to fine-tune individual demand forecasts.
Combining new product forecasting methods
As far as Comparable Forecasting goes, remember that this is but one of several important methods for new product forecasting. In fact, it’s probably most effective when combined with other statistical techniques. Indeed, our own IBP platform integrates multiple techniques in addition to Comparable Forecasting to maximize forecast accuracy, including:
- Spread Curves for new product adoption
- Supersession for cannibalization effects
We’ll discuss these techniques in forthcoming posts.
For now, you might conduct a review of the product portfolio – past, present, and future. Look for lessons in the sales data. Identify the products, product sets, or seasonal or geographic releases that are analogous to imminent or future new products. Also, review the forecasting & planning process. It hopefully combines at least some level of statistical modeling & simulation, workforce collaboration, and automation. Either way, it may be well worth looking into a best-of-breed solution that combines all three in a single, cloud platform.
Stay tuned for more to come on new product forecasting.
About Vanguard Software
Vanguard Software introduced its first product for decision support analysis in 1995. Today, companies across every major industry and more than 60 countries rely on Vanguard Software’s Integrated Business Planning (IBP), forecasting, and advanced analytic cloud platform. Vanguard Software is based in Cary, North Carolina.