06 March 2017

Forecasts are always wrong. So what?

By Xavier Perrin (xperrin@xp-consulting.fr)

We are teaching that "forecasts are always wrong"… and this is true! But, some people take the opportunity of this truth to assert that forecasting demand is useless.
For some time, I also hear or read more and more frequently that forecasting is useless because we are now living in a VUCA world (VUCA stands for Volatility, Uncertainty, Complexity, and Ambiguity).
This kind of reasoning is wrong because it misses a key aspect of the forecasting process.

Looking at the sailing boat on the left picture, conditions look cool. Calm seas, sunny and just right windy weather. Weather looks quite predictable in the hours ahead. There is no major issue, just a calm and nice cruise. Thus, each mate can relax and enjoy the sun. Now, if we look at the picture on the right, things look totally differently. It's also a sailing boat, but this one is a racing yacht. No doubt that this crew is willing to win. Conditions look not so nice. Seas are rough, wind gusts can hit the sheets at any time and waves are also unpredictable. Local conditions are unpredictable in the very-short range and make maneuvers critical and difficult. In such conditions, the crew must be totally focused and, above all, fully synchronized: the slightest mistake during a maneuver could capsize the boat.

Overcoming unpredictable conditions is easier when the crew and the boat are well trained and prepared, and, moreover, when the crew is acting synchronously. Winning a race is at first a question of planning! The crew doesn't know which conditions it will be facing during a particular stage of the race. But they are trained to act in every condition and, when the time comes, each mate knows what s/he should do in coordination with her/his team mates. Paradoxically, in the case of the boat on the left picture, where conditions are more predictable, the need for coordination is lower, and planning is easier: if weather forecasts are bad, the (wise) decision which is generally taken is not to leave the wharf!

One primary goal of forecasting is obviously to elaborate probabilities about future demand. But it is also the weakest point of forecasting because "forecasts are always wrong!" Also, demand forecasting feeds the planning process (S&OP and MPS levels of the planning system) which consists in taking decisions about demand and supply, and in allocating resources for reaching performance objectives. Planning firstly necessitates to define hypothesis regarding demand to develop plans for having all stakeholders acting consistently and synchronously. Without a formal expression of these hypothesis (a formal and unique set of forecasts), each manager who have to anticipate decisions – for example a purchasing manager negotiating with a supplier, or a production manager setting out a recruitment plan – will consider her/his own hypothesis, and there is little chance that they are consistent. The result could be having hired operators who cannot work due to lack of raw material, or to create useless inventories of raw material that cannot be processed due to lack of operators.

Thus, we need forecasts for planning, more specifically for synchronizing decisions. And, obviously, the closest to reality they are, the better it is.

However, we have to consider the value of striving to reduce forecast errors. In the case of regatta, we do not strive to forecast the strength of gusts or the height of waves with the better accuracy. Rather, we prepare the crew and the boat to overcome likely values with a reasonable margin of error. But the greatest effort is for preparing consistently the boat and the crew (For example, not overloading the boat with sheets that will likely not be used), and, above all, training and practicing for rapid, fluid and synchronized maneuvers.

In the same way, in the VUCA world, striving to reduce forecast errors could be a wasteful effort. Rather, our efforts should be directed towards creating the agility which allows to survive in the VUCA world. But agility isn't a feature that comes spontaneously when people become aware of its necessity! Creating an agile enterprise firstly requires creating teams embracing a broad range of skills, with multi-cultural open-minded people. In an agile enterprise, processes are agile. Agile processes are lean in that their lead time have been reduced by eliminating wastes. Moreover, managers and teams have got the skills for reconfiguring processes quickly for adapting to changing conditions. Such features cannot arise overnight. It necessitates clear vision of leaders, anticipation, appropriate resources… and time. And when time is required for doing something, planning is necessary. And planning necessitates shared hypothesis about what will happen. In other words, it necessitates forecasting!
Ce raisonnement est erroné car il occulte un des aspects du processus de prévision.Forecasts are always wrong. So what?

5 comments on “Forecasts are always wrong. So what?”

  1. Great analogy Xavier!

    Two more comments:
    1. Forecasts are wrong for sure, but the key point is measuring by how much. Safety stock, extra capacity or flexibility agreements with suppliers will be sized differently depending on the forecast accuracy regularly achieved.

    2. A good forecast process brings together sales, marketing, Planning staff to review assumptions and share market intelligence. This is as important as the numbers. Giving up forecasting because of inaccuracy would mean losing this 'soft' information.

  2. Exactly Xavier
    I agree also with Hervé.
    If forecasters tell me that the forecast is 100, I cannot do anything with this, since I know that it's wrong!
    But if they tell me 100 plus or minus 20, or even 100 plus or minus 100, I will be able to decide a level of production, a level of stock, according to the required service level.

  3. Hello Xavier,
    Thanks for the discussion.
    The different levels in the planning process, S&Op, MPS where a probability of projected demand will help decouple and synchronize resources, supply and demand call for a different granularity and time bucket of the forecasts.
    Forecasting at S&Op level pave the way to make decision in time at the medium, long term horizon.
    Statistically, as the forecast is a probability, it is better to calculate the forecast directly from the data available at each level ; i.e. the family level of forecast should ideally not be a sum of end item figures, but calculated directly at family level.
    There is no safety stock table against family level forecast, but rather 2 figures forecast.
    MPS end item level has been developped to decouple variability of demand from the MRP order generation, in order to dampen nervousness of the system.
    End item forecast becomes therefore the primary mechanism to deploy the planning, plus forecast error related safety stock.
    An alternative to forecast driven mps is to place inventories at strategic intermediate points along the supply chain, depending on the targeted Customer lead time, having these buffers dynamically adjusted against the average daily consumption, real or anticipated ( e,g, including promos, forecast ...).
    This is ddmrp methodology.
    The placement of buffers protects too from supply side variations.
    Buffers can be reviewed daily with the use of average daily consumption ; forecast driven mps are often reviewed once a month.
    Globally that just says that anticipations, forecasts are still used, for S&OP, for adjusting average consumption, but projected end item demand level is no more the primary mechanism of order generation.

    1. Thank you, Guy, for your relevant comment!
      Indeed, you describe how we should manage forecast inaccuracies with the help of DDMRP. I would like to reword it from another point of view. It is first necessary to well understand the value stream. To do this, Value-Stream-Mapping is a powerful tool, helping identifying bottleneck operations and imbalances between current cycle times and takt time. Buffers, which buffer operations against variations, are here called supermarkets. They are replenished with the help of kanban loops, which are updated according to changes of the average daily usage. This should not be done too frequently for avoiding nervousness and instability. Kanban loops also integrate buffers for compensating demand fluctuations. Moreover, still for creating stability, heijunka consists in leveling load and mix at the pacemaker (the process step which pulls all preceding steps).

  4. Bonjour,
    je choisis d'écrire en français.
    Les prévisions sont effectivement de plus en plus souvent fausses et trop variables; cela vient souvent d'esprits trop bourrins et trop limités dans le temps, voir d'un manque de compréhension et de conviction (vs expériences) de l'importance du PIC (S&OP). Le PIC, bien mené, permet de bien piloter l'entreprise sur tous ses axes (humains, financiers, industriels et achats, innovations, marketing, ventes-clients,...).
    Avec la mise en place du Lean, démarche constructive et qui lance du progrès et de la dynamique lorsqu'elle est bien menée, Carol Ptack nous a fait part d'un très bon mode d'adaptation MRP2 et Lean: le DDMRP & DDS&OP, qui permettent de toujours bien piloter l'entreprise sur les bases du PIC et de bien faire concrètement à l'heure et de qualité sur les bases du Lean. Ce fonctionnement permet de s'adapter aux variabilités et de les piloter de mieux en mieux (en interne entreprise et en externe-fournisseurs) pour assurer la satisfaction des clients et la vie-survie et même le développement, de l'entreprise.

    Have a good and dynamic way :-).

    Geneviève Gonin-Lanès,
    Expert en Lean et Supply-Chain management.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.