The Fallacy of the Single Version of the Truth

This is a re-publication. First published Sep 30, 2013 on Supply Chain Insights community, now offline.

In recent years I have heard a lot of companies striving for a single demand forecast across all departments and functions to serve as a so-called single version of the truth. I am of the opinion that this goal, however lofty, is both unattainable and undesirable. Or as Lora puts it: "unrealistic and misguided". [removed dead link to a Lora Cecere article on Supply Chain Insights community]

It wasn't too long ago that she too openly professed the single number forecast. I can still clearly recall listening in on a webinar and for the first time (on this topic) hearing her brazenly go against the mainstream opinion. I did not feel merely supported in what I had believed all along; I felt proud for her that she took a stand. It was not a popular opinion to have at the time, and may still not be. But like so many things, once you realize something is fundamentally flawed, you cannot support it anymore. It is the brave and opinionated ones among us that take it a step further and make it a point to expose the truth, even if it means contradicting ourselves on earlier beliefs. Chapeau Lora!

While I believe that the single number forecast is not the answer, I also do not endorse departments working in isolated silos. The problems caused by the latter have been a primary driver for the quest of the former. If we were black-and-white thinkers this would constitute a catch-22. I believe the answer is in the middle somewhere; a shade of gray, and if the single version of the truth is the white end of the spectrum, the answer would be a light shade of gray, but not quite white.

The reason the single number forecast is unattainable is because every department not only has different objectives (causing different biases) but they all have different baselines to which they apply their efforts. The biases are important, they cause a natural friction that no executive mandate can remove. Operations will never accept the sales forecast because it is always positively biased, usually severely so. Conversely, sales will never accept the operational forecast since it is constrained and not a true representation of demand. The budget is frequently not grounded in any supply chain reality, rather framed in terms of ambitious targets and based on politics and sandbagging by various departments trying to secure their portion of it. Each of these has a clear purpose, and admittedly in most companies, many of these processes are not working optimally. But even if they are driven appropriately the biases will not completely disappear, and the different baselines will always remain.

Many articles and discussions on this topic explain that you have at least two types of forecast: constrained and unconstrained; at least two types of unit of measure: volume and monetary; and at least a couple of different time horizons. With regards to unit of measure: the budget will always be in monetary terms, the operational plans will always be in volumetric terms. Marketing and sales plans could be in one or the other, but the ultimate demand forecast is typically in volumetric terms in order to drive inventory and supply. With regards to time horizons: each of the forecasts typically applies to different time ranges, as well as be required in different time granularities. Budgets may be for a rolling 2 years in quarters, marketing plans for a half year in months, sales forecasts for a year in weeks, operational plans for 6 weeks in days. Even if a financial plan would be the most accurate forecast to use, it would be useless to drive supply since it is in the wrong unit of measure and not granular enough in both time and product dimensions. With regards to constrained forecasts - i.e. those forecasts where practical limitations constrain the unconstrained forecast that aims to predict the customers' pure demand - there are more than one versions. For example there may be vastly different lead times at which capacity, raw material purchases, and manufacturing needs to be planned. For example, 6 months, 3 months and 2 weeks respectively. At each of the time ranges a different forecast may be the most accurate, and thus the best choice to base such decisions on, i.e the best baseline forecast.

What some practitioners and analysts offer is that what may be the baseline for one purpose should in fact be the target for another purpose. I fully agree with that assessment. This is a key reason why a single version of the truth is not desirable.

One process that would greatly benefit from this view of the different forecasts is Sales & Operations Planning (S&OP). Rather than trying to make the single consensus forecast also the single version of the truth that all parties from sales and operations would use as their baseline going forward, make it just the agreed upon plan for a certain limited time frame, that all parties are held accountable for. For each party that consensus forecast now becomes the target that each tries to make their own plans adhere to. The key note at this stage is that consensus does not guarantee adherence. Just because we want to, does not mean we can. But because we have two versions - the consensus target and the plan that we intend to execute - we can always identify the gaps. And again, this applies to all parties individually, so there will be multiple plans and forecasts.

At each S&OP meeting both parties bring their plans as the starting point and the differences become discussion starters about risks and opportunities. Note that in this approach we do NOT base this cycle's meeting on the the prior cycle's output consensus plan. Since each of the parties' plans accommodate the prior S&OP cycle's consensus plan as far as achievable, they are now more representative than the prior consensus. The accountability of adherence to the prior cycle can be kept outside of the meeting, since that triggers discussions about blame, which is highly non-productive. This is impossible to do if the prior consensus serves as the single version of the truth. Naturally, adherence to the consensus and accountability for that should be monitored and managed, but it should not be in the same meeting, nor necessarily with the same audience.

This same concept ideally needs to be applied to all forecasting and planning processes. The concept is that two versions are better than one version. With two versions you can compare what you have with where you need to be, and can determine a course of action to get there or to some common ground. With a single version of the truth you lose that capability. The only question you can then ask is why did we not achieve the single version, and who can we blame for that?

For each of the two versions (baseline and target) there may be multiple choices. Each plan gets updated many times within a cycle, and may have input from multiple parties in that process. The rule should be to pick that version of the plan or forecast that is most up to date, can be presented in the unit of measure and granularity needed, and is most reliable within the time range that we are impacting. The production schedule may be considered highly reliable and is granular enough to serve as any baseline. But it may only be so for the next week, and unreliable 1 month out. If for a given planning purpose we are impacting the time range from 2 to 6 weeks out, it would be clear that the production schedule is not the ideal baseline. You need to pick the appropriate baseline to suit the purpose.

Sometimes you may encounter a situation in a planning process in which there are more than two versions. This is common in S&OP. It has a much higher risk of derailing than the two version scenario, but also has the potential for increased targeting of efforts... if properly managed. Only one version should serve as the target, the other versions would be different baselines and/or different scenarios to consider. For example, the budget could be the target, whilst the operations plan is the supply baseline forecast, whilst the demand plan is the demand baseline forecast, and the marketing plan serves up potential means to achieve the targets within the constraints set by the other plans. Again, this will need proper management, and guiding that a predetermined protocol or set of principles.

This approach has the ability to work very efficiently, but it is imperative that all planning cycles are closed loop cycles. Just like the operational plan will be targeting the current consensus forecast and become the input to the planning cycle that generates the next consensus forecast, other plans and forecasts need to do the same. If an S&OP consensus plan renders the current budget unfeasible, then the budget needs to either also be updated, or predefined rules applied on how to handle the delta until the next budget update cycle, or it needs to simply be excluded from the next S&OP cycles. Any plan that has not closed the loop since the prior cycle will inject garbage into the current cycle.

The above describes the approach and process on how to deal with multiple versions of the truth once those versions are accepted. One important factor I did not yet touch on, is that many times the reliability of the various baselines and targets are disputed. Stepping away from the hard requirement of a single version of the truth, that factor is greatly diminished but not fully removed. In a single version of the truth environment arguments around "my version is the real truth and yours is not" abound. In a multi-version environment we can agree to disagree, and focus on how to close the gap between the different versions by changing plans to apply supply chain actions or demand shaping actions. However, even in these circumstances, opposing parties may frequently dispute the accuracy or efficiency of the other party's forecasts or plans. Operations will continue to be conservative and sales will continue to be optimistic, and each will continue to discount the other for it. This will frequently derail the discussion on how to close the gap between baseline and target into a discussion about the baseline being further away from target than it needs to be under current conditions and that targets are far-fetched and need to come down to be realistic.

There are three important factors that need to be in perfect alignment to alleviate this conflict:

1. Incentives. Each department needs to be incented in a way that does not drive away from overall company strategy. For example, while it is perfectly natural to have sales reps get bonuses based on sales volume, there should be a sufficiently-sized barrier to prevent the natural positive bias that bonus program creates from making it into the sales forecasts. And sales reps should then also be responsible for preventing product shortages by not promoting or shifting promotions when supply is destined to be too little. In other words, not just any volume, but profitable and achievable volume should be their incentive. The?same concept applies to manufacturing, inventory, finance, etc.
2. Accountability. Very closely related to the incentives, each person should be accountable for that portion of the forecast or plan that he or she controls. It should not necessarily be part of the planning cycle itself, but forecast accuracy or even better forecast value-add (FVA) should be measured. It should be used to discount any potential reward if exceptionally bad, and it should be accounted for in the baseline or target forecasts. One could consider applying historic bias of a person to any adjustments made, and even ignoring his or her adjustments completely when a certain bias threshold is exceeded (this to prevent incremental gaming of the process).
3. Process and Technology. I group these two together, since the majority of their value can only be obtained when they are used in finely choreographed combination. Having the right processes in place is crucial. Having closed loops to form a coherent fabric of processes that build on each other, control and enrich each other is just as crucial. Technology is the enabler. Even when a part of a planning process occurs in a meeting, technology will have the role of enabling the baselines and targets to be reliably established going into the meeting, and technology can and should be an enabler to facilitate decision making during the meeting. The incentives, accountability, and historical accuracy can become a formal part of the overall process via technology. Any dispute over baseline or target forecast or plan can be dismissed by virtue of it being formally addressed elsewhere, and any inaccuracies as far as can be determined already being accounted for.

The above has described almost the opposite of a single version of the truth. It has described many different versions of the truth, but they are ideally all single versions of different truths. Truths that apply to different portions of the supply chain, for different planning time horizons, and at different stages in the various planning cycles. Truths that can be viewed as "supply perspective" vs "demand perspective" vs "financial perspective" and so forth, or as "as-is" versus multiple possible "to-be"s. It should be clear that these differences cannot realistically be covered by a single forecast, and that is indeed unattainable and undesirable to try to do so.

However, process and technology do provide another opportunity to get much closer to the single version of the truth, as close as it makes sense to do so. It is more of a common ground than a holistic single forecast, but it does not suffer from differences of opinion, nor from biases caused by incentives. It can be the base that spawns the various baselines that each planning process requires. I will discuss this in a next blog post.

Find all my articles by category here. Also listing outstanding articles by other authors.