
“Failure is an option here. If you’re not comfortable with failure, you’re not comfortable with innovation.”
– Jeff Bezos
Unilever is a consumer-packaged goods giant. Several years ago, they had a serious production problem at one of their factories near Liverpool, in the north-west of England. They were making washing powder in the standard way – the way washing powder is still made today, where boiling chemicals are forced through a nozzle at very high speeds and as the pressure drops, they disperse into vapor and powder. Then, the vapor is siphoned off and the powder collected in a vat, where additional chemicals and ingredients are added and then sold as washing powders.
The problem Unilever had was that the nozzles didn’t work very well or smoothly – they kept clogging up. They were inefficient (causing numerous production delays), kept blocking and produced varying grains of powder, of different sizes. This was a major issue for Unilever, not just because of maintenance and lost time, but also in terms of the quality of the product which was rightly disappointing customers. They needed to develop a better nozzle. Much better.
So, they turned to an internal team of crackpot mathematicians. Unilever, like many large industrial giants, could afford the best and the brightest and had on staff a team of math folks who were experts in high-pressure systems and fluid dynamics. They were also experts in the physics of what is called “phase transition” – the process that governs the transformation of matter from one state to another (e.g., from liquid to gas).
The expert math team did what experts often do – they delved deep into the problem and developed sophisticated equations and solutions. After a long period of study and design, they came up with a new and what they believed was brilliant, design.
Problem was, it didn’t work. The powder granularity remained inconsistent, and the process was still inefficient.
In desperation, Unilever also turned to their team of biologists – a team with no knowledge of fluid dynamics and phase transitions. This team had something else that would prove crucial – a design approach that was rooted in experimentation. A test, fail and learn approach. One that seemed to understand that failure was part of learning, and often a necessity for innovation.
Instead of a brilliant design, they would make some minor tweaks, try it, see what worked and didn’t work, adjust and test again. Then, rinse and repeat, until they had a better nozzle that worked in practice, not just on paper.
They would take ten copies of the nozzle and apply a small change to each one, then test each one to see which ones failed, but importantly, which one performed the best. They would then take the “best” nozzle and do the same thing again: create ten slightly different copies and then repeat the process. Then repeat it again. And again. And again.
After 45 generations they had finally developed a nozzle that was excellent.
A world class nozzle had been built because of testing, discarding and improving upon four hundred forty-nine ‘failures’.
Progress/innovation had been achieved not by a grand, brilliant design, but instead by rapid interactions with the real world.
An approach, it turns out, that is fundamental to designing something better, even in supply chain management.
When it comes to inventory planning, people are embracing the concept of Flowcasting. That’s good. In my opinion, the breakthrough that’s made Flowcasting possible, is the capability to forecast, replenish and managing slow and super-slow selling items at store level.
The best solution to this chronic issue that had plagued planning system providers, was developed and architected by our long-time colleague, Darryl Landvater of the Oliver Wight Americas Team. And, largely by using the same approach as the biologists from Unilever – a test and learn approach, refining based on real-world learning’s, then doing it again and again until an elegant, intuitive solution was proven.
I won’t bore you with the details, but repeated tests and learning’s helped him understand and correctly conclude that managing slow sellers at the store level is simultaneously a forecasting and replenishment issue. His solution is the only approach that I know of that manages both the forecast and replenishment plan at store level in unison. It ensures a valid simulation of reality and, critically, that inventory can be maintained close to the store safety stock level for these types of items. That’s critical for a retailer, since most retailers have a significant percentage of slow sellers.
I’ve never asked Darryl how many ‘failures’ he endured on his journey to developing his slow seller solution, because it doesn’t really matter.
I’m just glad he did.