Author, speaker and associate researcher at Telecom ParisTech Michael Ballé shares his collaboration with Jacques Chaize, Patrick Decoster, and Ariane Bouzette that discusses lean as a science.

The lean management approach is based on Toyota’s revolutionary approach to developing products by first developing people which transformed the company from a bankrupt local manufacturer in the early 1950s to the dominant carmaker it is now.

Taiichi Ohno, one of the key innovators behind the development of the fabled Toyota Production System explicitly states his project as: “the only way to generate a profit is to improve business performance and profit through efforts to reduce costs.

“This is not done by making workers slave away, to use a bad expression from the olden days, or to generate a profit by pursuing low labor costs, but by using truly rational and scientific methods to eliminate waste and reduce costs”.

It is apparent that the originators of the TPS believed they pursued a “scientific mindset”. In his own description of the Toyota Production System, Ohno explains “when a problem arises, if our search for the cause is not thorough, the action taken can be out of focus. This is why we repeatedly ask why. This is the scientific basis of the Toyota system”.


But what kind of science are we talking about? Frederick Taylor was convinced he was inventing the principles of scientific management. How is Toyota’s approach different from the search for the “one best way” applied to all workers? Toyota’s outlook is unique because it explicitly seeks to combine its corporate destiny with personal fulfillment. Employee satisfaction is seen as the key to customer satisfaction, and the operational key to enhancing every employee’s sense of job fulfillment in on going on-the-job development.

Toyota teaches its managers to assign work, leaving space to think, and let the team member think thoroughly. The theory of engagement it teaches to its middle-managers starts with the person:

  1. I need development.
  2. I’m excited and ready to think!
  3. Think… think… think…
  4. I know what to do.
  5. I can’t do it.
  6. This is a challenge, I’ll never give up.
  7. I made it, what’s next!

Clearly, this makes many assumptions about human nature and people’s theory of mind, but the critical difference with taylorism is that it refuses the notion that experts set the system and operators operate. Toyota specifically requires managers to develop their team members through a method of learning by problem solving. In taking on work challenges, trying things for themselves, encountering obstacles and overcoming them, people hone their skills, learn to work with others find their job more fulfilling and take more pride in their work.

To sustain this personal problem solving journey, Toyota has adopted the plan-do-check-act cycle taught by quality guru J. Edwards Deming in the 1960s. Indeed, the company represents itself as an organization with the goals of “long-term prosperity and growth” sustained by the P-D-C-A cycles of each employee. This unique and very strong assumption has been replicated and distinguishes companies that truly adopt lean and have the hope for results, and those using the lean label cover their traditional taylorist approach of thinking for their employees.

A deep understanding of PDCA, is one of the early obstacles any company encounters in trying to adopt lean. Part of the difficulty lies in understanding the profound scientific dimension of the PDCA cycle in order to make effective use of it to develop people. Problem solving is not about solving efficiency glitches to make the process better, but in solving problems to develop employee’s deep knowledge about their work, how to work with their colleagues and ultimately to align their initiative and creativity with the company’s development.

The key to understanding the scientific nature of PDCA lies in focussing on the Bayesian dimension of any science. No scientific domain evolves in a vacuum. Every science rests on a priori principles which are then revised through a Bayesian empirical approach of, in Richard Feynman’s terms “try it and see.” In the XVIIIth century, Thomas Bayes hit upon the idea that beliefs could be upgraded. In the Bayesian perspective, probability represents a degree of belief, the human mind is central to human understanding.  Bayes proposed a mathematical rule explaining change in existing beliefs in light of new evidence.


E.g., you’re given a bag with white and black stones, without being told what they are. If you first draw a white stone, it’s easy to believe you hold a back of white stones. If your second draw is another white stone, your belief this is a bag of white stones will be strengthened. But now if you draw a black stone, you’ll have to change your mind and conclude you’re holding a bag of both white and black stones.

As you continue to draw stones, the frequency at which the stones are either black or white will lead you to increasingly precise notions of proportions of white and black stones in the bag. In layman’s terms, this describes the fact that our confidence in the belief of an idea should be sustained by the number of post hoc facts this idea can explain. As we encounter facts that cannot be explained by our hypothesized theory we need to reformulate our initial idea in order to accommodate both confirming and disconfirming facts.

Bayes offered a radical departure from the path of least resistance thinking that consists of ascertaining the validity of an idea by the most salient illustrations that come to mind (in effect, only picking a few number of examples that support the idea).

A lean example would be the fact that by executing any repetitive work one-piece-flow (say folding a paper, putting it into an envelope and sticking a stamp on in one continuous movement as opposed to folding all papers, putting all papers in envelopes, attaching all stamps to all envelopes) one gains around 20 or 30 % productivity. There is no known way to prove this theory unless one carries out the experiment many times in many different settings and finds the results always hold more or less true. In this case, there is no observable reason why following one-piece-flow should be more productive by the similar orders of magnitude, although repeated experiment increases our confidence in the theory that it does. Conversely, debating one-piece-flow as opposed to experiencing it will never convince any one of the magnitude of the effect.


At the workplace everyone has an opinion about everything. The question is: how robust are these opinions? In general, opinions are the result of a priori belief we pick up from experience or people around us, supported by one or two salient, vivid facts. For instance if the next-door plant has been shut down because it’s production has been moved to a low-cost country, it’s quite sensible to believe that globalisation is bad for employment. As a factory worker, you’d need a lot of evidence to accept that globalisation is overall good for your business. More than 3,000 psychological studies show that our default thinking mode is to select facts confirming what we already believe, and therefore, learn rather slowly.

The Bayesian nature of PDCA’s strength is that it teaches every employee to develop a different relationship with what he or she knows. There are no absolutes, just cases of greater or lesser certainty about what we know. This mental agility is the main resource that makes the entire company more adaptive as well as engage employees in their own work. In this sense, PDCA really is about think… think… think… think and then try it and see.

  • Plan:the first step of learning through problem solving is problem finding. It’s important to realise that scientific thinking does not work in a vacuum but starts with an intuition, a best guess derived from our a priori knowledge. In lean, there are several important principles that point towards specific problems, such as a customer satisfaction issue, an ergonomic strain, a long lead-time, or a process producing defectives. For the Bayesian process to kick in, it’s essential to start by stating:
  1. What is the question we’re answering? What is the lean principle involved?
  2. What is our best guess answer?
  3. How are we going to measure whether this is the correct answer?

This is really important: our intuition is our starting point. We need to honestly have a plan about how we’re going to solve the stated problem. We need to look at the situation, repeatedly ask “why?”, consider many factors and so on. We also need to know how we’re going to measure whether our idea works or not. Without a clear idea of measurement we know for sure we’re going to fall back on selecting the facts that support our intuition and neglect the rest.

  •  Do: try something and see how it goes. Do is about trying many small experiments to confirm or infirm our original intuition – not about committing resources to solve the problem. This means quick and dirty attempts, at first, then leading to more refined experiments. The aim is not to solve the problem, but to gauge the level of robustness of our original intuition.
  • Check: do and check are an on-going process. Check is essentially about listing cases in which our experiments confirm our intuition and which experiments show that something else is going on. As both piles of stones grow, we can derive a feeling of confidence in our original thought. Listing disconfirming cases is essential to the act of learning – without that, we get fake learning, and ideological self-reinforcement of what we already believe. Yes, this is uncomfortable, but that’s the point. The very aim of the exercise is to develop true knowledge, which is an understanding of which notions are trustworthy, which are likely but fragile, and which are plain doubtful.
  • Act: adopt, abandon or adjust, act is a judgment call about whether to pursue with our original intuition, to modify it, or to abandon it altogether. The answer at this stage is rarely unambiguous. The environment of the experiments is generally hard to separate from the test itself, and the question always remains open of how many more tests one should do to conclude. Essentially, the key point to keep in mind is we’re not seeking a philosopher’s truth but working knowledge that can carry us until the next check point without doing something foolish. No knowledge is ever set in stone, all statements are qualified and subject to change as new facts are unearthed, although some ideas have got legs and have been proved robust over time (not all of them correct).

The lean vocabulary shuns the word “solutions” to which it prefers “countermeasures” because, precisely, there are never any final solutions – all conclusions are subject to further experiments and enquiries. In the meantime, although we make do with humble countermeasures, many of those are far more robust than the ideological solutions being pushed around by true believers. As Taiichi Ohno pointed out: “we are all human and we are wrong half of the time”. The trouble is knowing which half. Here again, Ohno is explicit: “there are so many things in this world that we cannot know until we try something”.


Learning occurs in the process of shifting from thinking to try-and-see and back. This is a deep learning alternative to the taylorist assumption of “just accept what you’re told.” In real life, there’s always a bit of both going on. One should definitely take into account what one is told in formulating the original idea, intuition, after all, is a subconscious process of putting facts and ideas together from what one experiences. Yet rote learning, tradition and existing are only part of it – the starting point of the learning process, not its end.

For example, when one of the authors was the CEO of an industrial group, he worked with the main plant’s scheduling manager to increase service and reduce inventory. In 2008, product scheduling was managed by the MRP system, and because of an important yearly rebate promotional campaign, three months of inventory had to be stockpiled to be ready on day one of the campaign. As the company considered lean to improve its service performance and reduce its inventory, it started with the a priori notion that a leveled pull system would improve the situation.


Year one

Think: set up a leveled pull system

Try and see: focus on a few high runners and pull them through the production process. In a few months, the finished goods inventory is divided by an order of three and on-time-delivery improves by 20%. Yet, having a regular production of high runners work, some products fall short because customers buy these at infrequent intervals but large quantities.


Year two

Think: the production capacity dedicated to high running products is enough to produce the average day’s demand, but not enough to fulfill an unexpected large order of products sold infrequently to large customers.

Try and see: create a new category of middle-runner products which will be held in inventory and produced as soon as a customer asks. Since the demand is infrequent, production can level the inventory reconstitution. As a result, stock value goes back up some, but on-time-delivery improves by another 15%. It is apparent this approach does not work for the Russian market:  there is a marked seasonality and the inventory needed to satisfy the upsurge of summer demand is considerable – as well as always short.


Year three

Think: since the levelled pull principle is now confirmed by many cycles of experimentation, the idea is to treat the Russian market as a perfect customer that would order all year round the same amount, creating a stock calculated to correspond to the lowest historical sales in Russia. In this way, as the summer season starts, the factory needs only make the extra parts for Russia as opposed to produce it all at the expense of capacity for the other markets.

Try and see: weekly scheduling of quantities during the low demand season, and then according to real demand during the high demand season. On-time delivery increases again by 20% as inventory values hit an all-time low. However, levelling production schedules reveal components arrive in large irregular batches, causing production interruptions and huge component inventories.


Year four

Think: Applying pull principles to the supply chain in order to level frequent smaller deliveries.

Try and see: Renegotiating contracts with suppliers to better level shipments and the use of regular milk-runs. This is a continuing effort through years five and six, with experimentation with cross-docking, and relocalising parts closer to production, which progressively and radically change the entire supply chain.

Throughout this learning process, the scheduling manager learns to see his job differently, to work with his professional tools such as MRP, and to develop specific tools to level the flow of finished products and schedule production. The scheduling manager learns to better work with purchasing and procurement to solve problems across the supply chain. This produces a spillover effect of new tools and from upstream to the downstream, and eventually to suppliers. The Bayesian nature of this learning process can be seen because the lean tools were never applied as best practices at the outset. Through careful parsing of cases where each experiment worked and cases where it didn’t, the scheduling office developed unique case-by-case knowledge sustaining a continuous improvement curve, with significant financial impacts as shows the following cash curve, with a cash conversion ratio of 200%.



Lean is a radical departure from other forms of organisation because it starts from the individual’s learning journey, not from making the person adhere to organisational processes defined by others. The basic assumption is that people will do their job, but to do so with care and attention they must find their own fulfillment in the work, which creates a space for individual initiative and innovation. Here the lean approach redefines what a job is in work and continuous improvement:



Lean is scientific in that it is profoundly Bayesian, rules, solutions, processes don’t matter as much as the thinking process to study and differentiate specific cases and gain more detailed knowledge of all possibilities. The word “solution” doesn’t exist in the lean vocabulary. We talk about countermeasures to insist on the “probabilist” and not-quite-final nature of all action.

The founding fathers of lean were seeking a space for the craftsman pride and care in his or her work within the process-driven necessities of economies of scale. What they have found in Bayesian problem solving is an infinite space where employees can see their efforts take form and shape in both their working environments and in the products they build. This, requires creating a work environment where independent thinking and respect for what is already known is valued; where both following the process as well as trying new things is encouraged; where both a sense of efficiency as well as the freedom to fail without blame. The paradoxes that appear to Toyota observers are the very same contradictions that plague science, it takes a lot of effort and resources to reach narrow conclusions hard to generalise, no answer is ever definitive and all opinions are always subject to peer review. The very mental flexibility this process drives is precisely what the Toyota founders sought.

What makes lean a science is the training of every employee to the scientific mindset. The lean principles of customer satisfaction, just-in-time, jidoka, kaizen and standardised work are nothing more than questions one can ask of any enterprise process: how satisfied are your customers? What is your level of just-in-time? How quickly are defects recognised after being produced? What improvements are occurring in the work cycle? These questions trigger improvement, which are tried and tested repeatedly until changes are either adopted, adjusted or abandoned. The entire company continuously improves its quality in staying close to its customers preferences and reduces costs by avoiding mistakes caused by faulty reasoning and obsolete beliefs.

As with science where the ultimate responsibility for producing new knowledge rests with each individual scientist, in lean the responsibility for raising quality and productivity starts and lies with all employees in the workplace. Management’s role is to earn trust, provide necessary skills and understanding and secure their active participation in improvement activities. None of this can happen unless management recognises the necessity that each person master the learning discipline of PDCA and, in doing so, develop what Peter Senge has called the personal mastery of “continually clarifying and deepening our personal vision, of focusing our energies, of developing patience, and of seeing reality objectively.” The ideal lean organisation is not the one with the most effective, streamlined processes, but one where every person in the company is clear on what learning activity they’re involved in as part of their day to day job, thus, creating collectively a true learning organisation.