• +91 8446942500
  • scmproknowledge@iscmindia.net

Heuristics – Rule of Thumb and Mediocrity

Stefan de Kok, Co-Founder & CEO Wahupa
15th March, 2021

Most of us will have heard the term “heuristics” before, but what does it mean, and more importantly, what does it mean for your plans and forecasts?

A number of decisions are based on the “Rule of Thumb” – a simplification of a complex situation. However, with digitalization and computing capabilities a number of these decisions can be based on data. In this article, Stefan De Kok elucidates when we need to trust heuristics and when not to.

Most of us will have heard the term “heuristics” before, but what does it mean, and more importantly, what does it mean for your plans and forecasts?

The Merriam-Webster dictionary defines “heuristic” as:

involving or serving as an aid to learning, discovery, or problem-solving by experimental and especially trial-and-error methods

It stems from the ancient Greek word εὑρίσκω (heurískō), which means as much as “I discover“. But you may know it better under its other names rule of thumb, trial-and-error, or estimated guess.

In business in general and supply chain in particular, heuristics are found everywhere. Many of the uses are fine, some are dangerous, whilst a few are downright destructive. In this article, I aim to teach you how to fish for the bad uses in your own company rather than give you a ready-to-consume list of fish.

Why do we use heuristics?

There are two main reasons for the pervasive use of heuristics. Biologically, humans simply cannot deal with high levels of complexity, uncertainty, and chaos. The simple rules, often established over long periods of time or even generations, help us make sense of it all. They used to help us make snap decisions that would save our lives or livelihood most of the time. Historically, with the advent of the earliest computers, highly simplified models could be run where more complete or realistic models would require too much computing power. Over time, those simplifications became ingrained and the foundation upon which even more rules and expanded models were built.

Nowadays, our situations both require and allow for more accurate models of our business and supply chains, but old habits are hard to break. Especially, if careers, expertise, and recognition have been built upon the way it has always been done, there is great incentive to maintain the status quo. For every agent of change, there will be a hundred agents of obstruction to change.

Should we hunt down and kill all heuristics?

No. Some heuristics are fine. The key is knowing which are and which are not. Here are some general conditions under which using a rule of thumb is okay:

  1. If there is no or little value in improving beyond the quality of the heuristic output.
  2. If heuristic results are consistently close enough to the optimal output.
  3. If the topic to be covered is not sensitive to model errors.
  4. If the problem has no tractable alternative solution (e.g. NP-hard problems)

Conditions one and two begs the question: “why bother?” A good example is the Economic Order Quantity (EOQ) calculation. Deviating from the optimal value has very little impact on the bottom line, and simply rounding to whole pallets for example will lead to savings or conveniences elsewhere that far outweigh any additional cost in inventory. Under these two conditions, there is often a tradeoff to be found between complexity and effort of finding a better approach versus the value it brings. But if the result is regularly only 90% optimal and the business has tight margins the heuristic will be directly hurting its competitiveness.

Violation of Condition 3 is the most common cause of truly severe issues. Some problems are inherently sensitive to small errors or deviations. Weather forecasting is a well-known case. The butterfly effect where a butterfly flapping its wing causes a hurricane on another continent is its quintessential example. In business, anything aiming for some balance tends to also be sensitive to errors. Examples are inventory forecasting or cash flow forecasting. Even a small deviation in inputs or a minor approximation in the model will lead to large swings in the output. Rules of thumb applied under such conditions will lead to havoc. The result is often clearly visible in most businesses, but the root causes are typically hard to track down. Any time there is a lot of expediting, break-ins into frozen fences, lost sales, or margin erosion of any significance, you can rest assured that heuristics are at fault, caveat really bad management decisions.

Condition 4 is the escape hatch. Sometimes we know we are using a heuristic and that it is not really good enough, but there is no viable alternative. The remedy, if any, is twofold. First, research if there truly is a no better alternative. And second, reduce any reliance on the output of such processes. If it is a major risk, either remove it or contain it.

What to do?

So heuristics could be bad. Don’t go around like a kid with a hammer looking for nails. Don’t go looking for things to fix. Instead, take a good look at the things that are causing real problems. Those things that keep you awake at night.

For those issues, can you track down where the problems start? Maybe forecast accuracy is low, or inventory is lopsided, or sales accept orders within the frozen fence. Whatever it may be, and there may be multiple issues and multiple aggravating factors.

Then look for assumptions made in the processes that are causing the issue. And you may also look for assumptions made in processes between the potential cause and the bad impact. Look for things that are generally done to make the math or logic easier. Easy is the leading indicator of assumption. And these easy paths are equally likely to exist in your commercial planning or forecasting software as they are in spreadsheets. The more generic a software system is the more likely shortcuts were needed to make it fit all sizes. Key examples are the use of a normal distribution (or no distribution at all when things are uncertain), or segmentation like ABC/XYZ, or Pareto rules. There are places where those are fine, but many more where they only lead to trouble. Since we start by looking at trouble spots chances are, if you encounter a simplification, it has a role in that trouble.

Finally, whenever you hear “we’ve always done it this way“, or “this is a best practice“, you can be certain there is a heuristic to kill.

Author:
Stefan de Kok
Co-Founder & CEO
Wahupa

Copyright © All Rights Reserved www.scmproKnowledge.in