- •Measure, review, and improve: Most companies find that the single hardest measurement to agree upon is forecast accuracy. Let the people responsible for maintaining the forecast (generally sales and marketing) propose a consistent way to measure it, understanding that no approach will be perfect.
- The purpose of measurement is to track progress
over time. Is accuracy generally improving or not? If
not, why not, and in what areas? What products,
groups, markets, etc. are prone to the most inaccu
racy? Why? What can be done to improve forecasting
in these areas?
- Look at measuring forecast accuracy less as perfor
mance measurement and more as "environmental
benchmarking." Think of it as a thermometer telling
you when it's getting warmer or colder, so that you
know how to dress differently to accommodate the
environmental forces that you may not be able to
- Measure the accuracy at the level of detail which is
forecast. This may mean that the family forecast ac
curacy is an appropriate measurement for some prod
ucts, where line item or line item by region is
necessary for others.
Always measure the accuracy of the latest forecast
as adjusted in the previous month, even though this
may represent a change within the cumulative plan
ning lead time. If this is not measured, then there is
no good handle on how well the forecast update pro
cess is improving the numbers. Allowing forecast
changes within the cumulative lead time is a practi
cal necessity, since you will allow customers to order
within the cumulative lead time in any quantity, re
gardless of the forecast.
- In some cases it may be helpful to additionally mea
sure the accuracy of the forecasts as they existed at
cumulative lead time (for example, two or three
months prior to the month being measured).
- In other cases, it may be appropriate to additionally
measure forecast accuracy over broader periods of time
than just weeks or months: for instance, a rolling two-
or three-month forecast accuracy measurement that
dampens normal week-to-week or month-to-month
variations in the short term. These measurements may
be more effective in helping to identify bias and not
let normal variation be a distracter.
- The forecast accuracy measurement could be a per
centage of the actual sales vs. the forecast, or it could
be the standard deviation or average percentage varia
tion. It is strongly recommended that you consider
measuring the percentage of items that fall within
their predicted range of variation, thus identifying
how good a job has been done forecasting the items
within their "normal" variation.
- For instance, an item that typically has demand
variations of plus or minus 100 percent from month
to month will have accuracy measured based on
whether or not the actual demand fell outside that
100 percent "planned for" variation tolerance. Con
versely, normally consistent, "predictable" items with
historical variation of, for example, plus or minus 15
percent from month to month, will have their mea
surements reflect what percentage of these items fell
outside this predicted range.
- Enforce clear responsibility. Make sure it's clear
who's responsible for forecast accuracy. This doesn't
imply that every activity to support forecast updating
need be done by the same person. For instance, in many
companies master schedulers or planners may analyze
mix variation for product options or between products
in a family, and pass their observations and recommen
dations for forecast change to the responsible sales or
marketing person. The actual input of changes to the
data base or to factors that will affect a forecast genera
tion program may be done by people other than those
responsible for forecast accuracy.
It's critical to clarify who's responsible for the activities vs. who's responsible for the resulting accuracy.
The measurements should then be monitored, analyzed, and reported in a form where the responsible parties can explain the factors that they believe are causing the variation and propose changes that need to be made in the future.
- Recognize and reward improving forecast accuracy.
Don't "penalize" forecast inaccuracy, but pay attention
to it and ensure that the proper resources are devoted to
attempt to minimize it in the future. The focus should
always be on identifying the ultimate root cause of the
demand variation, even if that means pursuing informa
tion through multiple demand chain partners. It's only
by understanding the true cause of demand variation that
you can do a better job of predicting it in the future. The focus should be on the understanding of the causes, not just the publication of the accuracy numbers themselves.
• Document assumptions: Maintain a simple process
for documenting the underlying beliefs, assumptions,
or situations that lead to changing a forecast or, in some
cases, to leaving the forecast as is, given changing mar
ket factors or past demand variations. By reviewing these
"documented assumptions" on a monthly basis, a more
rigorous and consistent analysis of root causes is facili
tated. If the assumptions prove untrue, it may be time
to change the forecast numbers. If they come true, but
the demand patterns don't emerge as expected, this
should trigger consideration and analysis of how to bet
ter project the future numbers. By ensuring a routine
and consistent review of these assumptions, an institu
tionalized, simple, straightforward, and efficient process
can pursue the potential root causes of demand varia
tion in the future.
• Plan for inaccuracy: As described above, build a cul
ture, an acceptance, and an understanding of the reality
that forecasts will never be perfect. But to maintain cus
tomer service, this means constantly looking for ways
of accommodating these variations in the most cost and
time efficient manner. This should be the responsibil
ity of all supply chain partners and all functions within
the company, including sales and marketing, materials
management, planning, purchasing, manufacturing,
and even development and design personnel.
• Use Sales and Operations Planning to monitor per
formance: This is a consistent, monthly, institutional
ized performance review by a cross-functional and
multi-level management audience to ensure that atten
tion is being paid where variations occur, and that real
istic plans, budgets, targets, and approaches are being
pursued, with the reality of actual performance as a con
stant validator of all future plans.
Forecasting is an old problem made worse by the dynamic business environment of today. Ironically, however, the solution isn't new and improved "advanced forecasting" mathematical software tools. The solution is more enlightened and selective understanding, application, and use of a variety of traditional approaches developed over the last SO years. The really useful tools are not new. The key, however, is applying them appropriately across multiple organizations, to unify demand chain and supply chain partners. With a clear understanding of the causes of variation, techniques to accommodate the variation should be implemented based on sharing information, and better integrating and aligning data and business practice, across all partners.