Employing the Law of Large Numbers in Bottom-Up Forecasting

 

It is utterly implausible that a mathematical formula should make the future known to us, and those who think it can would once have believed in witchcraft. – Jakob Bernoulli (1655-1705)

forest through the trees

This is a topic I’ve touched on numerous times in the past, but I’ve never really taken the time to tackle the subject comprehensively.

Before diving in, I just want to make clear that I’m going to stay in my lane: the frame of reference for this entire piece is around forecasting sales at the point of consumption in retail.

In that context, here are some truths that I consider to be self evident:

  1. Consumers buy specific items in specific stores at specific times. Therefore, in order to plan the retail supply chain from consumer demand back, forecasts are needed by item by store.
  2. Any retailer has a large enough percentage of intermittent demand streams at item/store level (e.g. fewer than 1 sale per week) that they can’t simply be ignored in the forecasting process.
  3. Any given item can have continuous demand in some locations and intermittent demand in other locations.
  4. “Intermittent” doesn’t mean the same thing as “random”. An intermittent demand stream could very well have a distinct pattern that is not visible to the naked eye (nor to most forecast algorithms that were designed to work with continuous demands).
  5. Because of points 1 to 4 above, the Law of Large Numbers needs to be employed to see any patterns that exist in intermittent demand streams.

On this basis, it seems to be a foregone conclusion that the only way to forecast at item/store is by employing a top-down approach (i.e. aggregate sales history to some higher level(s) than item/store so that a pattern emerges, calculate an independent forecast at that level, then push down the results proportionally to the item/stores that participated in the original aggregation of history).

So now the question becomes: How do you pick the right aggregation level for forecasting?

This recent (and conveniently titled) article from Institute of Business Forecasting by Eric Wilson called How Do You Pick the Right Aggregation Level for Forecasting? captures the considerations and drawbacks quite nicely and provides an excellent framework to discuss the problem in a retail context.

A key excerpt from that article is below (I recommend that you read the whole thing – it’s very succinct and captures the essence about how to think about this problem in a short few paragraphs):


When To Go High Or Low?

Despite all the potential attributes, levels of aggregation, and combinations of them, historically the debate has been condensed down to only two options, top down and bottom up.

The top-down approach uses an aggregate of the data at the highest level to develop a summary forecast, which is then allocated to individual items on the basis of their historical relativity to the aggregate. This can be any generated forecast as a ratio of their contribution to the sum of the aggregate or on history which is in essence a naïve forecast.

More aggregated data is inherently less noisy than low-level data because noise cancels itself out in the process of aggregation. But while forecasting only at higher levels may be easier and provides less error, it can degrade forecast quality because patterns in low level data may be lost. High level works best when behavior of low-level items is highly correlated and the relationship between them is stable. Low level tends to work best when behavior of the data series is very different from each other (i.e. independent) and the method you use is good at picking up these patterns.

The major challenge is that the required level of aggregation to get meaningful statistical information may not match the precision required by the business. You may also find that the requirements of the business may not need a level of granularity (i.e. Customer for production purposes) but certain customers may behave differently, or input is at the item/customer or lower level. More often than not it is a combination of these and you need multiple levels of aggregation and multiple levels of inputs along with varying degrees of noise and signals.


These are the two most important points:

  • “High level works best when behavior of low-level items is highly correlated and the relationship between them is stable.”
  • “Low level tends to work best when behavior of the data series is very different from each other (i.e. independent) and the method you use is good at picking up these patterns.”

Now, here’s the conundrum in retail:

  • The behaviour of low level items is very often NOT highly correlated, making forecasting at higher levels a dubious proposition.
  • Most popular forecasting methods only work well with continuous demand history data, which can often be scarce at item/store level (i.e. they’re not “good at picking up these patterns”).

My understanding of this issue was firmly cemented about 19 years ago when I was involved in a supply chain planning simulation for beer sales at 8 convenience stores in the greater Montreal area. During that exercise, we discovered that 7 of those 8 stores had a sales pattern that one would expect for beer consumption in Canada (repeated over 2 full years): strong sales during the summer months, lower sales in the cooler months and a spike around the holidays. The actual data is long gone, but for those 7 stores, it looked something like this:

The 8th store had a somewhat different pattern.

And by “somewhat different”, I mean exactly the opposite:

Remember, these stores were all located within about 30 kilometres of each other, so they all experienced generally the same weather and temperature at the same time. We fretted over this problem for awhile, thinking that it might be an issue with the data. We even went so far as to call the owner of the 8 store chain to ask him what might be going on.

In an exasperated tone that is typical of many French Canadians, he impatiently told us that of course that particular store has slower beer sales in the summer… because it is located in the middle of 3 downtown university campuses: fewer students in the summer months = a decrease in sales for beer during that time for that particular store.

If we had visited every one of those 8 stores before we started the analysis (we didn’t), we may have indeed noticed the proximity of university campuses to one particular store. Would we have pieced together the cause/effect relationship to beer sales? My guess is probably not. Yet the whole story was right there in the sales data itself, as plain as the nose on your face.

We happened upon this quirk after studying a couple dozen SKUs across 8 locations. A decent sized retailer can sell tens of thousands of SKUs across hundreds or thousands of locations. With millions of item/store combinations, how many other quirky criteria like that could be lurking beneath the surface and driving the sales pattern for any particular item at any particular location?

My primary conclusion from that exercise was that aggregating sales across store locations is definitely NOT a good idea.

So in terms of figuring out the right level of aggregation, that just leaves us with the item dimension – stay at store level, but aggregate across categories of similar items. But in order for this to be a good option for the top level, we now have another problem: “behavior of low-level items is highly correlated and the relationship between them is stable“.

That second part becomes a real issue when it comes to trying to aggregate across items. Retailers live every day on the front line of changing consumer sentiment and behaviour. As a consequence of that, it is very uncommon to see a stable assortment of items in every store year in and year out.

Let’s say that a category currently has 10 similar items in it. After an assortment review, it’s decided that 2 of those items will be leaving the category and 4 new products will be introduced into the category. This change is planned to be executed in 3 months’ time. This is a very simple variation of a common scenario in retail.

Now think about what that means with regard to managing the aggregated sales history for the top level (category/store):

  • The item/store sales history currently includes 2 items that will be leaving the assortment. But you can’t simply exclude those 2 items from the history aggregation, because this would understate the category/store forecast for the next 3 months, during which time those 2 items will still be selling.
  • The item/store level sales history currently does not include the 4 new items that will be entering the assortment. But you can’t simply add surrogate history for the 4 new items into the aggregation, because this would overstate the category/store forecast for next 3 months before those items are officially launched.

In this scenario, how would one go about setting up the category/store forecast in such a way that:

  1. It accounts for the specific items participating in the aggregation at different future times (before, during and after the anticipated assortment change)?
  2. The category/store forecast is being pushed down to the correct items at different future times (before, during and after the anticipated assortment change)?

And this is a fairly simple example. What if the assortment changes above are being rolled out to different stores at different times (e.g. a test market launch followed by a staged rollout)? What if not every store is carrying the full 10 SKU assortment today? What if not every store will be carrying the full 12 SKU assortment in the future?

The complexity of trying to deal with this in a top-down structure can be nauseating.

So it seems that we find ourselves in a bit of a pickle here:

  1. The top-down approach is unworkable in retail because the behaviour between locations for the same item are not correlated (beer in Montreal stores) and the relationships among items for the same location are not stable (constantly changing assortments).
  2. In order for the bottom-up approach to work, there needs to be some way of finding patterns in intermittent data. It’s a self-evident truth that the only way to do this is by aggregating.

So the Law of Large Numbers is still needed to solve this problem, but in a retail setting, there is no “right level” of aggregation above item/store at which to develop reliable independent top level forecasts that are also manageable.

Maybe we haven’t been thinking about this problem in the right way.

This is where Darryl Landvater comes in. He’s a long time colleague and mentor of mine best known as a “manufacturing guy” (he’s the author of World Class Production and Inventory Management, as well as co-author of The MRP II Standard System), but in reality he’s actually a “planning guy”.

A number of years ago, Darryl recognized the inherent flaws with using a top-down approach to apply patterns to intermittent demand streams and broke the problem down into two discrete parts:

  1. What is the height of the curve (i.e. rate of sale)?
  2. What is the shape of the curve (i.e. selling profile)?

His contention was that it’s not necessary to use aggregation to calculate completely independent sales forecasts (i.e. height + shape) to achieve this. Instead, what’s needed is to aggregate to calculate selling profiles to be used in cases where the discrete demand history for an item at a store is insufficient to determine one. We’re still using the Law of Large Numbers, but only to solve for the specific problem inherent in slow selling demands – finding the shape of the curve.

It’s called Profile Based Forecasting and here’s a very simplified explanation of how it works:

  1. Calculate an annual forecast quantity for each independent item/store based on sales history from the last 52+ weeks (at least 104 weeks of rolling history is ideal). For example, if an item in a store sold 25 units 2 years ago and 30 units over the most current 52 weeks, then the total forecast for the upcoming 52 weeks might be around 36 units with a calculated trend applied.
  2. Spread the annual forecast into individual time periods as follows:
    • If the item/store has a sufficiently high rate of sale that a pattern can be discerned from its own unique sales history (for example, at least 70 units per year), then calculate the selling pattern from only that history and multiply it through the item/store’s selling rate.
    • If the item/store’s rate of sale is below the “fast enough to use its own history” threshold, then calculate a sales pattern using a category of similar items at the same store and multiply those percentages through the independently calculated item/store annual forecast.

There is far more to it than that, but the separation of “height of the curve” from “shape of the curve” as described above is the critical design element that forms the foundation of the approach.

Think about what that means:

  1. If an item/store’s rate of sale is sufficient to calculate its own independent sales profile at that level, then it will do so.
  2. If the rate of sale is too low to discern a pattern, then the shape being applied to the independent item/store’s rate of sale is derived by looking at similar items in the category within the same store. Because the profiles are calculated from similar products and only represent the weekly percentages through which to multiply the independent rate of sale, they don’t need to be recalculated very often and are generally immune to the “ins and outs” of specific products in the category. It’s just a shape, remember.
  3. All forecasting is purely bottom-up. Every item at every store can have its own independent forecast with a realistic selling pattern and there are no forecasts to be calculated or managed above the item/store level.
  4. The same forecast method can be used for every item at every store. The only difference between fast and slow selling items is how the selling profile is determined. As the selling rate trends up or down over time, the appropriate selling profile will be automatically applied based on a comparison to the threshold. This makes the approach very “low touch” – demand planners can easily oversee several hundred thousand item/store combinations by managing only exceptions.

With realistic, properly shaped forecasts for every item/store enabled without any aggregate level modelling, it’s now possible to do top-down stuff that makes sense, such as applying promotional lifts or overrides for an item across a group of stores and applying the result proportionally based on each store’s individual height and shape for those specific weeks, rather than using a naive “flat line” method.

Simple. Intuitive. Practical. Consistent. Manageable. Proven.

Noise is expensive

Noise

Did you know that the iHome alarm clock, common in many hotels, shows a small PM when the time is after 12 noon?  You wonder how many people fail to note the tiny ‘pm’ isn’t showing when they set the alarm, and miss their planned wake up.  Seems a little complicated and unnecessary, wouldn’t you agree?

Did you also know that most microwaves also depict AM or PM? If you need the clock in the microwave to tell you whether it’s morning or night, somethings a tad wrong.

More data/information isn’t always better. In fact, in many cases, it’s a costly distraction or even provides the opportunity to get the important stuff wrong.

Contrary to current thinking, data isn’t free.

Unnecessary data is actually expensive.

If you’re like me, then your life is being subjected to lots of data and noise…unneeded and unwanted information that just confuses and adds complication.

Just think about shopping now for a moment.  In a recent and instructive study sponsored by Oracle (see below), the disconnect between noise and what consumers really want is startling:

  1. 95% of consumers don’t want to talk or engage with a robot
  2. 86% have no desire for other shiny new technologies like AI or virtual reality
  3. 48% of consumers say that these new technologies will have ZERO impact on whether they visit a store and even worse, only 14% said these things might influence them in their purchasing decisions

From the consumers view what this is telling us, and especially supply chain technology firms, we don’t seem to understand what’s noise and what’s actually relevant. I’d argue we’ve got big time noise issues in supply chain planning, especially when it relates to retail.

I’m talking about forecasting consumer sales at a retail store/webstore or point of consumption.  If you understand retail and analyze actual sales you’ll discover something startling:

  1. 50%+ of product/store sales are less than 20 per year, or about 1 every 2-3 weeks.

Many of the leading supply chain planning companies believe that the answer to forecasting and planning at store level is more data and more variables…in many cases, more noise. You’ll hear many of them proclaim that their solution takes hundreds of variables into account, simultaneously processing hundreds of millions of calculations to arrive at a forecast.  A forecast, apparently, that is cloaked in beauty.

As an example, consider the weather.  According to these companies not only can they forecast the weather, they can also determine the impact the weather forecast has on each store/item forecast.

Now, since you live in the real world with me, here’s a question for you:  How often is the weather forecast (from the weather network that employs weather specialists and very sophisticated weather models) right?  Half the time?  Less?  And that’s just trying to predict the next few days, let alone a long term forecast.  Seems like noise, wouldn’t you agree?

Now, don’t get me wrong.  I’m not saying the weather does not impact sales, especially for specific products.  It does.  What I’m saying is that people claiming to predict it with any degree of accuracy are really just adding noise to the forecast.

Weather.  Facebook posts.  Tweets.  The price of tea in China.  All noise, when trying to forecast sales by product at the retail store.

All this “information” needs to be sourced.  Needs to be processed and interpreted somehow.  And it complicates things for people as it’s difficult to understand how all these variables impact the forecast.

Let’s contrast that with a recent retail implementation of Flowcasting.

Our most recent retail implementation of Flowcasting factors none of these variables into the forecast and resulting plans.  No weather forecasts, social media posts, or sentiment data is factored in at all.

None. Zip. Zilch.  Nada.  Heck, it’s so rudimentary that it doesn’t even use any artificial intelligence – I know, you’re aghast, right?

The secret sauce is an intuitive forecasting solution that produces integer forecasts over varying time periods (monthly, quarterly, semi-annually) and consumes these forecasts against actual sales. So, the forecasts and consumption could be considered like a probability.  Think of it like someone managing a retail store. They can say fairly confidently that “I know this product will sell one this month, I just don’t know what day”!

The solution also includes simple replenishment logic to ensure all dependent plans are sensible and ordering for slow selling products is based on your opinion on how probable you think a sale is likely in the short term (i.e., orders are only triggered for a slow selling item if the probability of making a sale is high).

In addition to the simple, intuitive system capabilities above, the process also employs and utilizes a different kind of intelligence – human.  Planners and category managers, since they are speaking the same language – sales – easily come to consensus for situations like promotions and new product introductions.  Once the system is updated then the solution automatically translates and communicates the impact of these events for all partners.

So, what are the results of using such a simple, intuitive process and solution?

The process is delivering world class results in terms of in-stock, inventory performance and costs.  Better results, from what I can tell, than what’s being promoted today by the more sophisticated solutions.  And, importantly, enormously simpler, for obscenely less cost.

Noise is expensive.

The secret for delivering world class performance (supply chain or otherwise) is deceptively simple…

Strip away the noise.

Unvarnished

It’s an altercation that’s stuck with me for decades.

Roughly twenty years ago I was leading a retail team that would eventually design what we now call Flowcasting. We were an eclectic team, full of passion and dedicated to designing and implementing something new, and much better.

After a particularly explosive team session – that saw tensions and ideas run hot – everyone went back to their workstations to let sleeping dogs lie. One business team member, who’d really gotten into it with one of the IT associates, could not contain his passion. He promptly walked over to the team member’s cubicle and said…

“Oh, one more thing…F**k You!!”

Like most of the team, I was a little startled. I went over and talked to the team member and we had a good chat about how inappropriate his actions were. Luckily the IT team member was one cool dude and he didn’t take offence to it – the event just rolled off his back. To his credit, the next day my team member formally apologized and all was forgiven.

Now, please don’t think I’m condoning this type of action. I’m not. However, as a student of business, change and innovation I’ve been actively learning and trying to understand what really seeds innovation and, in particular, what types of people seem to be able to make change happen.

And, during my research and studies, I keep coming back to this event. It’s evidence of what seems to be a key trait and characteristic of innovative teams and people. They are what many refer to as…

Unvarnished.

If I think back to that team from two decades ago, we were definitely unvarnished. We called a spade a spade. Had little to no respect to the company hierarchy and even less for the status quo. And, as a team, we were brutally honest with each other and everyone on the team felt very comfortable letting me know when I was full of shit – which was, and continues to be, often.

But that team moved, as Steve Jobs would say, mountains – not only designing what would later morph into Flowcasting, but implementing a significant portion of the concept and, as a result, changing the mental model of retail planning.

I had no idea at the time, but being unvarnished was the key trait we had. Franseca Gino has extensively studied what makes great teams and penned a brilliant book about her learnings, entitled “Rebel Talent”.

She dedicates consider time to unvarnishment and quotes extensively from Ed Catmull, famed leader of Pixar Animation Studios who’s worked brilliantly with another member of the unvarnished hall of fame – Steve Jobs.

According to Catmull, “a hallmark of creative cultures is that people feel free to share ideas, opinions and criticisms. When the group draws on the unvarnished perspectives of all its members, the collective knowledge and decision making benefits.”

According to Catmull, and others (including me), “Candor is the key to constructive collaboration”. The KEY to disruptive innovation.

Here’s another example to prove my point. When I was consulting at a national western Canadian retailer, our team was lucky to have an Executive Sponsor who was, as I now understand, unvarnished as well.

As the project unfolded I was amazed how he operated and the way he encouraged and responded to what I’d call dissent. Most leaders of teams absolutely abhor dissent – having been unfortunately schooled over time that company hierarchy was there for a reason and was the tie-breaker on decision making and direction setting.

Our Sponsor openly encouraged people to dissent with him and readily and openly changed his mind whenever required. I vividly remember a very tense and rough session around job design and rollout in which he was at loggerheads with the team, including me. When I think back, it was amazing to see how “safe” team members felt disagreeing with him – and, in this case, very passionately.

As it turned out, over the next few days, we continued the dialogue and he changed his opinion 180 degrees – eventually agreeing with his direct report.

Neuroscience refers to this as being able to work with “psychological safety” – which is a fancier way of saying people are free to be unvarnished. To say what they believe, why and to whom with no consequences whatsoever.
Without question, as I’ve been thinking and studying great teams and innovation I realize just how brilliant this Sponsor was and the environment he helped to foster.

How many Executives, Leaders or teams are really working in an unvarnished environment – with complete psychological safety? I think you’d agree, not many.

If you, your company and your supply chain is going to compete and continually evolve and improve, won’t ongoing innovation need to become a way of life? And that means people need to collaborate better, disrupt faster and feel completely comfortable challenging and destroying the status quo.

Now, I’m not saying that when you don’t agree with someone to tell them to go F-themselves.

What I am saying – and other folks who are a lot smarter than me – is that hiring, promoting, encouraging and fostering people and a working environment that is unvarnished will be a crucial!

So here’s to being unvarnished. To being and working in safety. To real collaboration and candor.

And to looking your status quo in the eye and saying…”F**k you!”

Questions and Answers

Questions

Did you know that most, if not all, organizations and innovations started with a question, or series of questions?

Reed Hastings concocted Netflix by asking a simple question to himself…”what if DVD’s could be rented through a subscription-type service, so no one ever had to pay late fees?” (Rumor was that this was just after he’d been hit with a $40 late fee).

Apple Computer was forged by Woz and Jobs asking, “Why aren’t computers small enough for people to have them in their homes and offices?”

In the 1940’s, the Polaroid instant camera was conceived based on the question of a three year old. Edwin H. Land’s daughter grew impatient after her father had taken a photo and asked, “Why do we have to wait for the picture, Daddy?”

Harvard child psychologist Dr. Paul Harris estimated that between the ages of two to five, a child asks about 40,000 questions. Yup, forty thousand!

Questions are pretty important. They lead to thinking, reflection, discovery and sometimes breakthrough ideas and businesses.

The problem is that we’re not five years old anymore and, as a result, we just don’t seem to ask enough questions – especially the “why” and “what if” kinds of questions. We should.

Turns out our quest for answers and solutions would be much better served by questions. To demonstrate the power of questions, let’s consider the evolution of solutions to develop a forward looking, time-phased forecast of consumer demand by item/store.

Early solutions realized that at item/store level a significant number of products sold at a very slow rate. Using just that items sales history, at that store, made it difficult to determine a selling pattern – how the forecasted demand would happen over the calendar year.

To solve this dilemma, many of the leading solutions used the concept of the “law of large numbers” – whereby they could aggregate a number of similar products into a grouping of those products to determine a sales pattern.

I won’t bore you with the details, but that is essentially the essence behind the thinking that, for the retail/store, the forecast pattern would need to be derived from a higher level forecast and then each individual store forecast would be that stores contribution to the forecast, spread across time using the higher level forecast’s selling pattern.

It’s the standard approach used by many solutions, one who’s even labelled it as multi-level forecasting. Most retail clients who are developing a time-phased forecast at item/store are using this approach.

Although the approach does produce a time-phased item/store forecast, it has glaring and significant problems – most notably in terms of complexity, manageability and reasonableness of using the same selling pattern for a product across a number of stores.

To help you understand, consider a can of pork and beans at a grocery retailer. At what level of aggregation would you pick so that the aggregate selling pattern could be used in every store for that product? If you think about it for a while, you’ll understand that two stores even with a few miles of each other could easily have very different selling patterns. Using the same pattern to spread each stores forecast would yield erroneous and poor results. And, in practice, they do.
Not only that, but you need to manage many different levels of a system calculated forecast and ensuring that these multi-level forecasts are synchronized amongst each level – which requires more system processing. Trying to determine the appropriate levels to forecast in order to account for the myriad of retail planning challenges has also been a big problem – which has tended to make the resulting implementations more complex.

As an example, for most of these implementations, it’s not uncommon to have 3 or more forecasting levels to “help” determine a selling pattern for the item/store. Adding to the issue is that as the multi-level implementation becomes more complex, it’s harder for planners to understand and manage.

Suffice it to say, this approach has not worked well. It’s taken a questioner, at heart, to figure out a better, simpler and more effective way.

Instead of the conventional wisdom, much like our 3 year old above, he asked some simple questions…

“What if I calculated a rolling, annual forecast first?” “Couldn’t I then spread that forecast into the weekly/daily selling pattern?”

As it turns out, he was right.

Then, another question…

“Why do I have to create a higher level forecast to determine a pattern?” Couldn’t I just aggregate sales history for like items, in the same store, to determine the selling pattern?”

Turns out, he could.

Finally, a last question…

“Couldn’t I then multiple the annual forecast by the selling pattern to get my time-phased, item/store forecast?”

Yes, indeed he could.

Now, the solution he developed also included some very simple and special thinking around slow selling items and using a varying time period to forecast them – fast sellers in weekly periods, slower sellers monthly and even slower sellers in quarterly or semi-annual periods.

The questions he asked himself were around the ideas of “Why does every item, at the retail store, need to be forecast in weekly time periods?”

Given the very slow rate of sales for most item/stores the answer is they don’t and shouldn’t.

The solution described above was arrived at by asking questions. It works beautifully and if you’re interested in learning more and perhaps asking a few questions of your own, you know how to find me.

So, if you’re a retailer and are using the complicated, hard-to-manage, multi-level forecasting approach outlined above, perhaps you should ask a question or two as well…

1. “Why are we doing it like this?”
2. “Who is using the new approach and how’s it working?”

They’re great questions and, as you now know, questions will lead you to the answers!

Accuracy or Precision?

 

It is the mark of an educated mind to rest satisfied with the degree of precision which the nature of the subject admits and not to seek exactness where only an approximation is possible. – Aristotle (384 BC – 322 BC)

barn

My favourite part about writing these articles is finding just the right quote to introduce them. Before we get started, go back and read the quote from Aristotle above if you happened to skip past it – I think it both accurately and precisely summarizes my argument.

Now in the context of forecasting for the supply chain, let’s talk about what each of these terms mean:

Accuracy: Ability to hit the target (i.e. how close is the actual to the forecast?)

Precision: Size of the target you’re aiming at (i.e. specificity of product, place and timing of the forecast)

I’m sorry to be a total downer, but the reason this article is titled Accuracy or Precision is because you can’t have both. The upper right quadrant in the illustration above ain’t happening (a bit more on that later).

In the world of forecasting, people seem obsessed with accuracy and often ask questions like:

  • What level of forecast accuracy are you achieving?
  • How should we be benchmarking our forecast accuracy?
  • Are we accurate enough? How can we be more accurate?

The problem here is that any discussion about forecast accuracy that does not at the same time account for precision is a complete waste of time.

For example, one tried and true method for increasing forecast accuracy is by harnessing the mystical properties of The Law of Large Numbers.

To put it another way – by sacrificing precision.

Or to put it in the most cheeky way possible (many thanks to Richard Sherman for this gem, which I quote often):


Sherman’s Law:
Forecast accuracy improves in direct correlation to its distance from usefulness.


So how do we manage the tradeoff between precision and accuracy in forecasting?You must choose the level of precision that is required (and no more precise than that) and accept that in doing so, you may be sacrificing accuracy.

For a retailer, the only demand that is truly independent is customer demand at the point of sale. Customers choose specific items in specific locations on specific days. That’s how the retail business works.

This means that the precision of the forecasting process must be by item by location by day – full stop.

Would you be able to make a more accurate prediction by forecasting in aggregate for an item (or a group of items) across all locations by month? Without a doubt.

Will that help you figure out when you need to replenish stock for a 4 pack of 9.5 watt A19 LED light bulbs at store #1378 in Wichita, Kansas?

Nope. Useless.

I can almost see the wincing and hear the heart palpitations that this declaration will cause.

“Oh God! You’ll NEVER be able to get accurate forecasts at that level of precision!” To that I say two things:

  1. It depends on what level of accuracy is actually required at that level of precision.
  2. Too damn bad. That’s the requirement as per your customers’ expectation.
With regard to the first point, keep in mind that it’s not uncommon for an item in a retail store to sell fewer than 20 units per YEAR. On top of that, there are minimum display quantities and pack rounding that will ultimately dictate how much inventory will be available to customers to a much greater degree than the forecast.Forecasts by item/location/day are still necessary to plan and schedule the upstream supply chain properly, but it’s only necessary for forecasts at that level of precision to be reasonable, not accurate in the traditional sense of the word. This is especially true if you also replan daily with refreshed sales and inventory numbers for every item at every location.

There are those out there who would argue that my entire premise is flawed. That I’m not considering the fact that with advances in artificial intelligence, big data and machine learning, it will actually be possible to process trillions of data elements simultaneously to achieve both precision and accuracy. That I shouldn’t even be constraining my thinking to daily forecasting – soon, we’ll be able to forecast hourly.

Let’s go back to the example I mentioned earlier – an item that sells 20 units (give or take) in a location throughout the course of a year. Assuming that store is open for business 12 hours out of every day and closed 5 days per year for holidays, there are 4,320 hours in which those 20 units will sell. Are we to believe that collecting tons of noise (whoops, I meant “data”) from social media, weather forecasting services and the hourly movement of soybean prices (I mean, why not, right?) will actually be able to predict with accuracy the precise hour for each of those 20 units in that location over the next year? Out of 4,320 hours to choose from? Really?

(Let’s put aside the fact that no retailer that I’ve ever seen even measures how accurate their on hand records are right now, let alone thinking they can predict sales by hour).

I sometimes have a tendency to walk the middle line on these types of predictions. “I don’t see it happening anytime soon, but who knows? Maybe someday…”

Well, not this time.

This is utter BS. Unless all of the laws of statistics have been debunked recently without my noticing, degrees of freedom are still degrees of freedom.

Yes, I’m a loud and proud naysayer on this one and if anyone ever actually implements something like that and demonstrates the benefits they’re pitching, I will gleefully eat a wheelbarrow of live crickets when that time comes (assuming I’m not long dead).

In the meantime, I’m willing to bet my flying car, my personal jetpack and my timeshare on the moon colony (all of which were supposed to be ubiquitous by now) that this will eventually be exposed as total nonsense.

Fundamentals

Imagine you’re a hotshot US high school basketball player during the late 1960’s. You’re all American and can pretty much choose any collegiate program you want, with an almost for certain scholarship attached and a better than decent chance that someday you’ll play professionally in the NBA.

You finally decide to join the famed UCLA program, under the tutelage of the acclaimed John Wooden, aptly named The Wizard of Westwood – an eventual winner of 10 NCAA titles in 12 seasons, including a still record 7 in a row.

I think it’s safe to say you’d be both nervous and excited.

First practice day arrives and you’re intrigued to see what you’ll learn from the Wizard. What techniques does he employ? What’s his secret sauce? How has he been so successful?

Wooden arrives, promptly introduces himself and his staff and new recruits begin to practice a drill that Wooden starts every season with…

learning to put on your socks and tie your laces.

Astonished, your first practice with the top program in the country would include practicing, over and over, how to put on your socks and tie your shoe laces.

John Wooden understood how important fundamentals are. If the players got blisters from their socks or improperly laced shoes, they wouldn’t be able to move as well. If they couldn’t move as well, they couldn’t rebound or shoot as well and they’d miss shots. And if they missed rebounds and shots, they’d lose.

Master the fundamentals, Wooden realized, and the team would excel.

It’s impressive thinking and very insightful for all us working in supply chain planning.

Flowcasting is based on fundamentals. I remember talking recently to an Executive from a technology-focused and hyped firm who proclaimed to me, “Flowcasting is not new or innovative”. After all he proclaimed, “the concept has been around for more than 15 years and the book is 12 years old”.

Ok, so what’s that got to do with anything? Does something have to be new, exciting and, to date, unproven to add value? I think not and I know you do too.

Flowcasting is based on sound fundamentals that will stand the test of time. A forecast of consumer demand, at customer touchpoint only, is used to calculate and translate that only unknown into all resource requirements in the extended retail supply chain (or any industrial supply chain).

Joe Orlicky, one of the Godfathers of time-phased planning, MRP and DRP, said it best, almost 50 years ago when he proclaimed ”Never forecast what you can calculate”.

I’ve never met Dr. Orlicky or Dr. Wooden but I believe that they are kindred spirits. They understood the power and importance of fundamentals.

Now you do too.

A Symphony of Placid Beauty

Game6-FinalPosition

July 23, 1972. Reykjavík, Iceland. An event would occur that day that would rock the chess world and would continue to be talked about, even to this day. It was game 6 of the acclaimed world chess championship match between the World Champion, Boris Spassky of Russia, and the challenger, Bobby Fischer of the United States.

Fischer opened 1. C4, the English Opening – the first time in his entire career that he’d deviated from his beloved 1. E4 opening move. Spassky quickly transformed the opening into the Queen’s Gambit Declined, for which he was one of the world’s foremost experts in this line of defense.

What followed was a masterpiece. Fischer’s moves were new, exciting and novel on a variation that had been played by chess grandmasters for centuries. The moves were pure, clean and deceptively simple, yet powerful and profound.

When Spassky resigned after Fischer’s 41st move, not only did the crowd stand and applaud, so too did Spassky. They all knew that they’d witnessed a masterpiece. A game of such beauty and purity – still talked about to this day, as chess perfection.

Dr Anthony Saidy, a Fischer confidant and assistant described it magically when he proclaimed, “it was like a symphony of placid beauty”.

Fischer’s play in game 6 captures his signature style: crystalline – transparent but ingenious and incredibly profound and powerful. Nigel Short, the highest ranked British Grandmaster of all time sums up Fischer’s play nicely, “The thing that strikes me about Fischer’s chess,” he says, “is that it’s very clear. There are no mysterious rook moves or obscure manoeuvrings. There’s a great deal of logic to the chess. When you look at it you can understand it – afterwards. He just makes chess look very easy.”

There are a lot of parallels to Fischer’s chess, particularly game 6 from 1972, and Flowcasting.

Flowcasting, as you know, seamlessly connects the supply chain from the consumer to the factory in a natural and logical way. That’s easy to understand and most people seem to get that.

However, like analyzing Fischer’s moves, the nuances of the process are deceptively powerful and profound. I’ll outline a couple of important ones, though there are others and they follow the Fischer-like mantra – deceptively easy to understand, simple, yet profound.

The forecasting approach used by the leading Flowcasting solutions does not use any sophisticated algorithms. Instead it uses and builds on profile based forecasting techniques that have been around for decades – the subtle improvement (like some of Fischer’s novelties in game 6) is the use of differing forecast time periods by SKU, converting them to integer forecasts for slow selling items and then consuming these forecasts as actual sales happen.

Why is that placid-like beauty? Because if you study retail, and real sales history, you’ll uncover that the vast majority of products sell less than 26 units per year per store for virtually any retailer. Trying to find a fancy algorithm that can predict when these sales will occur is a fool’s game.

The simple combination of integer forecasting, consumption and daily re-planning simplifies the solution to deliver results. And, once explained to planners, makes sense to them and is easy to understand and manage. Much like Fischer’s moves, the ideas and concepts are deceptively simple and profound.

Consider also the simple and profound concept within Flowcasting of daily, net change re-planning. At our most recent implementation the solution works like this: for any product that had a sale or change that occurred yesterday, then the entire supply chain is re-forecast and re-planned from consumer to factory for that item.

In retail, on a daily basis, only about 5-15% of the products experience a change daily. Only these are re-planned daily, adjusting the future flows of inventory to ensure you remain in-stock and your inventory is productive. All projections are easily and simply converted into the language of the business so that the entire organization is working to a single set of numbers. One other benefit of net change, daily re-planning is that it also dramatically reduces system processing requirements.

Flowcasting is an easy concept and solution to understand and most people wholeheartedly agree with the premise. The trick to making it successful is to understand the nuances, embrace its simplicity and instill, over time, that kind of thinking in your planning organization.

Once you do, then, from experience, your supply chain and indeed your company will work like a Symphony of Placid Beauty!

 

Ungrain

In 1983 Benjamin Libet, researcher in the Department of Physiology at the University of California, performed one of the most famous and controversial experiments in the history of neuroscience.

In simple terms, Libet’s experiment measured and timed the response of the neural circuitry of the brain, based on some very basic commands – like moving your left wrist, followed by your right wrist.  What he discovered is that there is a time lapse between the decisions our neural circuitry makes for us and our awareness of the situation.

What that means, in a nutshell, is that for basic operations and requests the brain has already been hardwired, or ingrained, into a conditioned response, basically without thought.  The brain has seen this movie (or ask) so many times that the response is automatic.

For us folks who are working on changing people’s behaviors and habits, we can relate.  People become ingrained in current practices, processes and ways of thinking and it usually takes considerable time and effort to change – the thinking and the response.

Libet, however, didn’t stop there.  Further work, research and experiments concluded that there were generally only two ways to change the history of the brain as it relates to a specific ask or task.  They are asking WHY and making a JOKE of the situation.

Let’s look at an important supply chain planning example and focus on the WHY.

To date, most retail planners, consultants and solution providers have firmly cemented and ingrained the thinking that to systemically create a forward looking time-phased forecast by item/store (or webstore) requires that you forecast at multiple levels and then spread the higher level forecasts down to the lower (store) level.

Initially the thinking was that the aggregate level forecast would be more accurate, and that is usually the case.  But some people realized that the higher level forecast was of no value – it’s the lowest level of forecast that drives the integrated supply chain.

Asking and wondering WHY enough times eventually surfaced that the higher level forecast was really only helpful in determining a selling pattern, especially for slower selling products where a pattern was difficult to detect.

Our colleague, Darryl, not only understood but asked WHY it was necessary to forecast at a higher level.  Couldn’t the pattern be determined, at the selling location, without the need and complexity of forecasting at a higher level.

Eventually, he arrived at a simple, sensible solution.  In a previous newsletter, I outlined the key elements of the approach but the key elements of the approach are:

  • An annual forecast is calculated, along with a decimal forecast (by day and week) for the 52 weeks that comprise the annual forecast
  • A category or department level selling pattern is calculated at the store location (or other’s if needed)
  • Simple user forecast thresholds are applied against the annual forecast to determine the forecast time period and how to determine the selling pattern – including using the category/department level pattern from above for slow sellers (to get the sales pattern)
  • The same thresholds determine whether to convert the decimal forecast to integers
  • For the forecasts that will be converted to integers, a random number between 0 and 1 is calculated, then the small decimal forecasts are added from there and once the cumulative forecast hits 1, then an integer forecast is 1 is used in that period, and the counter and randomizer starts again…this logic is applied to the 52 week forward looking forecast

Now, while the above is tougher to write to help understand, our experience in outlining this to people is that they not only understand, but it makes intuitive sense to them.

This solution was originally key functionality of the RedPrairie Collaborative Flowcasting solution and is now available within the JDA solution set, aptly named JDA Slow Mover Forecasting and Replenishment.

Yes, but does it work?

The graph below outlines the sales forecasts of our recent implementation of Flowcasting at a Canadian hardgoods retailer, using this exact approach:

Slow-sellers2

As you can see, a significant number of products are slow or very slow sellers (54% sell less than one unit per month at a store).  However, using this approach the company was able to improve in-stock by 6%, while also reducing and improving inventory performance.

Having an integer-like forecast for all these item/store combinations is important since it allows them to calculate time-phased DC and vendor replenishment plans, along with complete capacity and financial projections – allowing them to work to a single set of numbers.

In addition, the solution is so much simpler in terms of understanding, flexibility and processing requirements.

Given the above, people should embrace this solution full tilt.  This should be a no-brainer, right?

Nope, wrong.

Our old villain, ingrained, has helped cement the view of higher level forecasting in retail.

It’s ironic, and a little sad, that a number of people and companies who advise and help companies change and learn new and presumably better ways have not embraced this approach, and instead are still pushing old, tired and ineffective solutions.

They need to ungrain their thinking (ungrain is the opposite of ingrain and yes, I made this word up!).

My advice is simple: if you’re a retailer who is forecasting at a higher level, or you’re someone who’s pushing this approach, please stop.

Learn. Understand. See it yourself.  Ask WHY.  And, importantly…

Ungrain the old and begin to ingrain the new.

A beautiful mind

Do you remember the movie “A Beautiful Mind”?

The film is based on mathematician Dr. John Nash’s life, and, during one part, attempts to explain how Nash got the idea for his equilibrium theory as a part of game theory. In the scene Dr. Nash is at a bar with three pals, and they are all enraptured by a beautiful blond woman who walks in with her friends.

While his friends banter about which of them would successfully woo the woman, Dr. Nash concludes they should do the opposite – Ignore her. “If we all go for her,” he says, “we block each other and not a single one of us is going to get her. That’s the only way we win.”  That’s the moment when he formulated his idea.

The idea that pops into Dr. Nash’s head at that moment is very instructive in the innovation process.  Often, real innovation happens because you are in a situation and you’re paying attention, or listening, and you just connect the dots.  It’s the subconscious mind at work, finally coming to grips with something you’ve likely been pondering for a while.

It’s a great film and a beautiful story.

Here’s another beautiful story of essentially the same approach that was used to create the breakthrough thinking and solution in demand planning at store level – which, as we know, drives the entire Flowcasting process.

In retail, forecasting at store level, systemically, has been a major challenge for a long time.  Not only do most retailers have millions of store/item combinations, they also need to deal with virtually every imaginable sales pattern.  But, by far, the largest challenge, is the large number of slow selling items – accounting for 50%+ of virtually any retailers assortment.

The main issues with slow selling items is twofold: finding a selling pattern amongst sparse data, and ensuring that the forecast reflected the somewhat random nature of the actual sales.

The hero in our true story is named Darryl.  Darryl is the architect of the RedPrairie Flowcasting solution (now part of JDA) and, specifically, the profile-based, randomized integer forecasting approach that has simplified retail store level forecasting to a beautiful, elegant, intuitive process that does something incredible – it works and is very low touch.

The baseline forecasting process works like this:

FcstApproach2

In Darryl’s approach, unlike that of other attempts, he first calculates an annual forecast by item/store.  Then simple user defined sales thresholds automatically doing the following:

  1. Determine what time period to use to forecast in (weeks, months, quarters, semi-annual)
  2. Determine which level of already pre-aggregated history to use to spread the annual forecast in the time period
  3. Determine whether to convert the forecast into integers – which he randomizes by store/item, ensuring that the same item across many stores will not have an integer forecast in the same week

How did he think of this?  Well, similar to Dr. Nash, he found himself in a situation where someone said something very interesting and it sparked his thinking and helped him connect the dots.

Rumour has it that Darryl was walking around a Canadian Tire store years ago and was talking to the owner of the store.  They approached a section of the store and the owner grabbed a particular product and said something like, “I don’t know when we’ll sell these, all I know is that we’ll sell two every quarter”!

BOOM!  The idea for a different time period for forecasting by item/store popped into Darryl’s head and this event triggered the thinking and eventual development of the baseline forecasting process.

This is a significant development – so much so that it has been patented and is now available with the JDA product solution set.  What it has done is obsolete the need for multi-level forecasting approaches that, to date, have been the norm in attempting to create store/item level forecasts.

This approach is simple, intuitive, elegant and is computationally blazingly fast – another key requirement in retail store level forecasting.

Oh, and it also works.  We implemented this exact approach during our very successful implementation of Flowcasting at Princess Auto.  The solution is forecasting items in all varying time periods and is creating store/item forecasts for products that sell from 1 unit a year at store level, to over 25,000 units a year.

Even more important is that the people that would become demand planners (with no prior knowledge or experience in demand planning) would understand and become proficient using this approach.  Just another benefit of simplicity.

John Nash looks like any other bloke.  But, without a doubt, he’s got a beautiful mind.

The hero in our story, Darryl is just like John.  If you met him, you’d immediately think he’s another Vermont farmer who’s good with hydraulics.  But behind those coveralls and hay-stained hands is…

A beautiful mind.

Flipping your thinking

When students at Segerstrom High School in California attend calculus class, they’ve already learned the day’s lesson beforehand — having watched it on a short online video prepared by their teacher, the night before.

So without a lecture delivered by a teacher, students spend class time doing practice problems in small groups, taking snap quizzes, explaining concepts to the class, and sometimes making their own videos while the teacher moves from student to student to help kids who are having problems.

It’s a new form of learning called Flip – because the idea has flipped traditional education on its head – homework is for the lecture, while the classroom, traditionally reserved for the lecture, is for practice and deeper learning and collaboration.

Flipped learning is catching on in a number of schools across North America, as a younger, more tech-savvy student population – including teachers – now make up the typical classroom.

When it comes to supply chain planning, the concept of flipping applies nicely and most people, and most companies, could benefit greatly by flipping their thinking.

Let’s take CPG manufacturers.  When it comes to demand planning, they have it difficult.  Trying to forecast what their retail and other customers are going to do and want is difficult and it’s not getting any easier.  The empowered consumer, changing and dynamic retailer-led strategies are just two examples of shifts that are making it almost impossible to predict the demand, with any level of reasonableness.  The result?  Additional inventory and buffer stock required to respond, “just in case”.

There are a number of studies that prove this point.  Forecast accuracy has not improved and, in most cases, it’s getting worse.

Supply chain practitioners and experts are responding in the typical fashion.  We need better algorithms, fancier formulas, maybe even artificial intelligence and some big data sprinkled on top in order to find a better forecasting engine.

Sorry folks, that’s not working and as consumers and customers become more demanding and expectations rise, it’s going to get worse.  What’s needed is to flip the thinking and to change the paradigm.

CPG manufacturers, for the most part, are forecasting what should be calculated.  The demand plan they are trying to predict for their customer, should be provided to them in the form of a supplier schedule.  And that schedule should reflect the latest knowledge about the consumer, and any and all associated strategies and tactics that will entice the consumer’s buying patterns and/or product flows.

Forecasting consumer demand is, as has been proven, simpler and easier that trying to predict dependent demand – that is, the resulting demand on DC’s and plants based on ordering rules, lead times, and other constraints that tend to “pollute” the dependent demand plan.

When it comes to demand planning, Joe Orlicky had it right some 40 years ago: you should never forecast what can be calculated.

Of course, what we’re talking about is a retailer using the Flowcasting process to plan all flows from supplier to consumer – factoring in any and all constraints that translate the consumer forecast into the purchase projection from retailer to supplier.

Why is this so much better than the traditional approaches?  First, the entire retail supply chain (or any industrial supply chain) is driven by only one forecast – consumer demand.  All other demands can and should be calculated.  The effect is to dramatically simplify planning.  The retailer and manufacturer are working to a single, shared forecast of what’s expected to sell.

Second, the entire supply chain can be re-planned quickly and effortlessly – making the supply chain agile and dynamic.  Changes are and can be viewed almost in real-time and the changes are automatically translated for all partners in the supply chain – in units, cube, weight, dollars, capacity or any language needed throughout the supply chain.  The result is that the entire supply chain is working to a single set of numbers.

Third, when you embrace the idea of Flowcasting as it relates to planning, you get so much more than a better forecast.  Unlike traditional approaches that are trying to mathematically predict the demand, the supplier schedules that are a resultant of the Flowcasting process, calculate the demand by aggregating product flows.

Therefore, trading partners can see, well into the future, projected product flows between any two locations and this provides tremendous insight and flexibility to improve and smooth flows, as well as proactively put in place solutions to potential flow issues before they happen.  The retailer and manufacturer can actually work, using the same system and process, as if they were one company – all oriented to delight and deliver to the consumer, in the most profitable manner possible.

Finally, in addition to providing product flows the approach also produces projections of sales, inventory, purchases, receipts and, as mentioned, flows in any language of the business – units, cube, weight and capacities for operations folks and dollars for financial folks and Management in order to get better control of the business and ensure that plans stay on track.

If you’re planning the retail supply chain, you get so much more when you forecast less.

So, what is the path forward for manufacturers?

They need to flip their thinking and understand that they are trying to forecast what should be calculated – and that this practice will soon be obsolete.

Next, they should engage and work with their key retail and other customers to help educate their customers that a process like Flowcasting not only helps them (in the form of a supplier schedule and complete visibility), it provides even more value to the retail customer.  In fact, to date, it’s the only planning approach that consistently delivers in-stock levels of 98%+, even during promotions – crushing the industry averages of around 92%.

Once they are successful, a CPG manufacturer, over time, can be working with their top retail customers and receiving valid, up-to-date, supplier schedules that in most companies account for 70-85% of their volume.  The additional demands can then be forecasted using the latest approaches – demand sensing, etc.

Imagine, for a moment, what that would mean to the retail industry and the CPG manufacturers in general.  The impact would be enormous – from increased sales and profits, to significant reductions in inventory and working capital.  Not to mention the impact to consumers and customer loyalty.

Is all this possible?

Sure, but to make it happen the first step is to flip your thinking.