Business forecasting 101

04

Difficult forecasts

Spare parts forecasting

When equipment inevitably breaks down, the right spare parts must be on-hand. Problem is, forecasting demand for these parts is complex and difficult, far more so than for ordinary production components. One spare part may be critical to zero-downtime, but rarely needed. Another may be expensive, carried for a long period, and prone to obsolescence. These and other factors can stymie spare-parts demand planning.

Regardless of the difficulty, achieving the optimal size and configuration of spare-parts inventory is critical. Too little and you run the risk of costly or even fatal operational delays. Too much and you tie up cash in spare-parts safety stock — carrying costs kill too.

The most accurate forecasting methods for spare parts are those able to handle intermittent demand, which is characterized by randomness and long periods of zero demand. Indeed, spare-part usage often vacillates between erratic and zero. There are many ways to get around this lack of obvious recurring patterns. One popular method is Croston’s intermittent demand model. A more advanced approach is Monte Carlo simulation, which uses probability theory to model the time between orders, as well as order sizes. Other approaches include increasing the forecast time scales from daily to weekly to monthly and so on, to reveal more stable and predictable patterns.

In all cases, optimizing spare parts inventory should take into account a number of key factors — Is the part expensive? Is it critical? What is your storage capacity? What is the lead time for purchase/manufacturing of this part? Which parts should be ordered on demand, versus kept in stock?

New product forecasting

New product forecasting is one of the greatest challenges that forecasters face. With limited historical data, domain knowledge becomes vital. The following describes methods forecasters can use to augment their own judgment.

All of the following methods are available in Vanguard Predictive Planning.

Comparable forecasting
Used when there is a new product that is similar to another established product. The comparable product’s historical patterns and trends help shape the new product’s forecast. Comparable Forecasting, however, does not capture new product launch effects.

In Vanguard Predictive Planning: Users can select comparable products by any attribute, such as product family, SKU code, region, and more.

Spread curve
A spread curve takes the pattern of a demand factor (e.g. seasonality) and applies it to the forecast of a new product.

In Vanguard Predictive Planning: A user with domain expertise can select the likely demand effects that will impact a new product and apply them to the forecast. Effects can include seasonality, decay, launch time, life cycle, region, and more. Vanguard works with a client’s domain experts to discover, test, and isolate the effects of key demand factors.

Supersession
Used when a new product is likely to cannibalize or replace an existing product or set of products. Supersession is especially useful for technology products or certain seasonal items that have multiple iterations.

Adoption modeling
Used to capture new product launch effects. New products garner increasing sales at first. Sales then slow as products mature, moving from the early adopter phase toward market saturation. Adoption modeling is typically used in long-range forecasting, or in cases where product-life-cycle effects are prominent. It does not capture effects such as seasonality or cannibalization.

Combination of methods– Unique to Vanguard Predictive Planning
Vanguard Predictive Planning can combine all the above new product forecasting methods so that a single new product forecast can finally include new product adoption curves, comparable products, demand effects via spread curves, and cannibalization effects.

Using comparables forecasting

You can sometimes improve a forecast by basing it on the historical data of other similar items. This technique, called comparables-based forecasting, is especially useful in cases where you have a limited amount of data or where the data is very erratic. In both of these cases, it can be difficult to discern patterns and trends in items individually while it is much easier to discern patterns and trends if each item is treated as a member of a group of similar items.

Subscribe to our blog

Never miss industry news, technical information, or thought leadership articles.


Request demo
Get a quote
Chat now!