Forecasting in the supply chain: in search of the philosopher's stone

Author: Svyatoslav Oleinik, ABM Cloud inventory management consultant

Between the hammer and the anvil

Perhaps the most common problem both in production and throughout the supply chain is shortages and surplus raw materials / finished products. At the very beginning of the company's life cycle, the surpluses do not significantly worry the owners. The growth curve of margin and sales initially seeks up, which suggests that all the unsold goods this month can be sold next. If the surplus can hide behind optimistic forecasts for some time, then it is easier to see deficiencies: there is nothing for the customer to ship, and all that could be earned during shipment is the loss of profit. The situation looks more difficult for retail, where there are penalties that exacerbate the problem with lost sales.

The solution to the problem seems so obvious - after all, you can simply increase the safety stock. But as soon as we begin to do this, the value of the stock begins to grow, and at the same time does not solve the problem of deficiencies qualitatively. Over time, we wonder: why with the growth of the company and production volumes, profits stop increasing? And sometimes it even happens that there is not enough working capital to pay off receivables or to purchase the necessary components for real orders. Just at this moment, the focus of attention switches to money frozen in stocks. What to do in this case? Set a task for employees to reduce inventory? In response, we hear that this is possible, but at the expense of availability ... It feels like we are between a hammer and a hard place. Intuitively, we understand that there is a solution and it is somewhere in the middle, the only problem: it is necessary to find it.
')
Fine castles on the sand

Back in 1913, American engineer Wilson published an article in which he described how to calculate the optimal lot size. To do this, you need to know the cost of placing an order, storing products and annual demand. Having determined the optimum point, it is possible to minimize the total variable costs associated with ordering and storing stocks. In other words, when ordering more often, but less, the costs associated with the process of ordering goods will increase, and ordering less often, but more - the cost of storage. Is it really?

There are several assumptions in this formula, and one of the key is that demand is known. Therefore, in order to somehow bring this formula closer to real life, it is necessary to use an “insurance” stock, which in the real structure of the inventory often exceeds the volume of the “usual” (demand between deliveries). Using the safety stock in the equation, we get a completely different value of the “optimal” stock.

Modern practitioners supplement and improve the initial idea, adding to the model the conditions for working with a multi-order order, taking into account losses from a deficit, etc. But in reality, no matter how complex and multifactorial model for calculating the "optimal" volume. If it is built on too simplified or unrealistic basic assumptions, there will be no good result. You can not build a house on the sand.

Ah ... if only our prediction was more accurate ...

I believe, for it is absurd

There is a certain religion in which managers, from production to retail, sincerely believe that a quality forecast will solve if not all, then the absolute majority of their problems. And most importantly, the expediency of this belief is very easy to prove, considering only an “insurance” reserve in money, not to mention surpluses and deficits. Things are easy - to calculate the quality sales forecast.

Every company, department, even an individual employee has a “own” approach to forecasting. In practice, there were three such levels:

1. The manager uses the average value and his experience / vision / feelings.
2. Basic statistics filtering and some moving average.
3. The system of multifactor filtering statistics and dynamic selection of the optimal prediction model, which, like the filtering system, are constantly being improved.

Indeed, the average accuracy of the forecast with the transition from one level to the next really increases. Go from the first to the second level of development is quite simple, and you can even get a good result. The rise from second to third usually takes years, and the benefit will not be as substantial as the first time. At the third level, a separate staff is required, powerful servers for calculations, expensive software, etc. And all this only in order to squeeze each subsequent percentage increase in the accuracy of the forecast.

What is forecasting really?

Forecasting is the belief that the future will look the same as the past. Those patterns and trends that worked yesterday will work tomorrow. And, as you may have guessed, it is not. Therefore, the difficulty of achieving each subsequent percentage of forecast accuracy increases exponentially. In other words, it is simply impossible to achieve an “accurate” forecast, not only mathematically, but also from the side of expenses or even common sense.

But do not talk about it to the head of the planning department :)

About battles with windmills

Nevertheless, managers do not give up attempts to improve the accuracy of the forecast by at least a few percent. As an example, several "classic" approaches to solving a forecast problem:

• buy corporate training to raise the level of staff knowledge,
• hire a renowned expert / mathematician or purchase a new equipment / software product,
• to hold other events with the help of which already this year there will be an opportunity to defeat the “evil giants”.

The work of forecasting staff is usually highly valued because they perform one of the most difficult tasks. They are constantly working at the turn of the famous, improving their calculations and mathematical models. From the results of their work depends on the direction of the entire company, the number of lost sales and money frozen in stocks. They also indirectly affect sales. They are in constant search of this mythic balance, which helps to extract maximum profit with minimum investment.

The difficulty is only smoke, in which it is convenient to hide from others, but it is easy to suffocate yourself.
In the classical system, it depends on the forecast how much and when we will produce, and on the accuracy of the forecast and the availability of resources - how much surplus reserves we will have to keep and what percentage of lost sales we will have to endure.

Therefore, when you ask your employees why so much money is frozen in stocks, they will always be able to exquisitely and very hard to explain that in these conditions they make the best of what is possible. They do not gossip, on their side of the mathematical calculations: "Do not believe us - here's the data for you - calculate yourself better." Most likely, no one in the company will be able to make a better forecast. But further improvement or search for errors within the system while simultaneously increasing the complexity of the calculations will be made more difficult even by the developers themselves.


In practice, there was a case when the system worked for several years with an error in the system calculations, but since the algorithms were complex, almost none of the employees could understand them. Those employees who were at the origins of the development of algorithms either did not work in the company, or did not remember exactly how it all functions, since they were engaged in other developments. What is interesting, everyone was sincerely sure that in fact, the system works somehow differently. How exactly, each had their own idea. Only after a detailed audit of the system, the error was discovered ... All this time, the company was losing money due to the excessive complexity and opacity of mathematical calculations.

Albert Einstein said: “Every educated fool can make things bigger and more complicated. You need a touch of genius and a lot of courage to move in the opposite direction. "

The clouds are gathering

In fact, it’s not only you that do forecasting. This is a favorite occupation of all the supply chain links from the final point of sale in retail and to the supplier of raw materials. And since between the links of the chain there is usually no free exchange of information, the distributor believes that what retail orders and there is a real demand. But in reality, this is not the case: if we analyze the structure of the ordered quantity, we will see that there is a lot of “safety stock” there, display and some other additional parameters.

A distributor, planning his stocks, predicts sales (shipments) and adds to them a certain safety stock. The calculated quantity is ordered from the manufacturer. The manufacturer plans to sell (shipments), which actually consist of a certain distorted form of the forecasted demand by retail + insurance stock of retail + insurance stock of the distributor and, like a cherry on a cake, the manufacturer’s own safety stock ... Like a snowball, with each order further along the chain supply, the concept of "demand" is overgrown with distortions in the information. And since we produce what we have predicted, then:

"Minor fluctuations at the beginning of the supply chain lead to significant fluctuations at its end . " This phenomenon, better known as the whip effect, is well described in Peter Senge’s book The Fifth Discipline on the example of beer supplies.

Looking problem in the face

Since it is difficult to deal with the consequences, and the root causes are an order of magnitude easier, let's try to understand what actually prevents us from increasing efficiency. Is the root of evil really a problem with the prognosis and effect of the stick?

Problems with the forecast?

Very often, we focus on “battles with windmills”, releasing the basic properties of the forecast out of focus:

1. Always inaccurate. You can argue about the percentage of accuracy, but the fact remains.
2. The farther into the future, the less accuracy. Forecasting for the week ahead, using the average sales of the previous week, is likely to be more accurate than the forecast for the week, which will come in a month, even if in this forecast to use more sophisticated forecasting models.
3. The more details, the less accuracy. Predicting the sale of a chain of stores is easier than selling at a separate outlet. Predicting sales for a separate SKU per week is easier than for the same SKU, but on some particular day of the week.

But if the forecast is so bad, why do we even use it?

If a client placed an order and is ready to wait a month, and our production cycle, taking into account the purchase of raw materials, is three weeks, then we do not need to predict. Having received the order from the client, we can simply purchase raw materials and do what the client wants. And all this without unnecessary inventory and without lost sales. But since the client’s waiting time is shorter than the production cycle, we have to guess the future. Therefore, the forecast is an attempt to gain time, which we lack.

And another tricky question:

Does the forecast really determine the level of money that is frozen in stocks?
Perhaps you should not seek to improve the accuracy of the forecast, constantly complicating mathematical models, ignoring the basic properties of the forecast, and use them. For example, if we reduced the supply / production leverage, we would reduce the need for a “normal” (between deliveries) stock of products. And since we have reduced the forecast arm, its accuracy has increased (the farther into the future, the less accurate). The more accurate the forecast, the less the safety margin is needed.

Or another example: we have an “optimal” production batch of 1000, with a production cycle of 1 day, average daily shipments - 10 pcs. Question: Does the forecast help significantly reduce the average stock? The answer will be: no, because the main stock level is determined by the “optimal” lot. If we want to reduce the stock - it is necessary to reduce the size of the “optimal” party, which already somehow conflicts with the very concept of “optimality”. Perhaps there are other criteria that determine the optimum party ...

By building the entire system from the forecast and focusing too much on it, we miss the improvement opportunities and levers that really determine the level of stocks, namely, reaction speed, frequency, relevant information and only last of all, the sales forecast.

In search of a solution: Is the forecast really not needed at all?

• Not needed at operational level. Using the forecast leads to the negative consequences described above. It is necessary to get away from pushing out (production agrees with the forecast with the hope that sometime this product will still be sold) to stretch. It is necessary to configure the system so that if necessary, we can produce what the client really wants. No more, no less. At the same time, it is necessary to focus, first of all, on the reliability and speed of response of the production system, and only then on improving the accuracy of the forecast, since it:

• Needed, but at a strategic level. This is a necessary tool for strategic planning, accounting for macro factors and emerging trends. It is not bad for analyzing and adjusting the direction of the company's movement, planning production facilities, space, equipment, transport, personnel, etc.

A forecast is only a tool with its strengths and weaknesses. The main thing is not to fall under the famous proverb: “If a hammer is your only tool, then for you all problems look like nails.”

Perhaps you should not spend your life searching for the philosopher's stone, trying to turn lead into gold - we know how it ended. The secret of success is in a more rational use of both lead and the small gold that is available.

Based on the problems described, there is a need for a solution that:

1. Minimizes distortions in information and does not transmit them further along the chain, while planning production work based on actual sales, and not forecast. This will create a stream of correct information, and as a result, the right materials.
2. Allow to reduce the leverage of production planning. And with a simple but effective signal system, it will protect the flow of correct information and materials.
3. Increase ROI. Having ensured uninterrupted flow, we will be able to get away from the “battles with windmills” and focus on the things that really determine the level of stocks and the availability of the necessary raw materials and components.

How to manage stocks based on demand?

If we analyze all key management methodologies from the influence on the flow, we can come to the conclusion that they do not contradict, but complement each other. Accordingly, in order to succeed in modern conditions, a company needs to expand its set of management tools.

American consultants Kerol Ptak and Chad Smith in the book “Planning the material needs for Orliki”, the third edition of which was published in 2011, presented to the world a new development - the inventory management methodology based on the demand of DDMRP. This concept of inventory management in the entire supply chain includes the best components from classical MRP, lean manufacturing (Lean), Six Sigma (6 Sigma), Theory of Constraints (ToC). One of the key ideas laid down is to look for the root cause of the problems that arise and use only those tools that are still relevant and really work. The methodology is designed to optimally adjust the transparent flow of materials and information throughout the supply chain, allowing you to keep only the right amount of stocks at strategic points.

The methodology itself consists of 5 steps:

1. Strategic positioning. Determine where and whether to store stock at all.
2. Profiles and buffer levels. We answer the question: how many stocks need to be stored?
3. Dynamic tuning. What rules should change the amount of stock?
4. Orders based on real demand. We do not produce if we do not sell in the future.
5. Transparent and joint execution. A simple but effective signaling system that will protect the production flow and work towards achieving a common goal - maximizing ROI.

The methodology has gained popularity in Western companies, such as Unilever, LG International, Oregon Freeze Dry, etc. This case study describes how one western company managed to reduce inventories by $ 2 million by implementing the DDMRP methodology.