Monday, February 24, 2014

Fear of the Black-box

imageAlmost a decade ago, I was involved in implementing a replenishment planning system for a large retailer. The new system used statistically projected demand forecasts, available and projected inventory, and dynamically computed lead-times to recommend the replenishments. It took care of order sizing to comply with vendor contract terms and pack-sizes, and rounded up/down using demand volume, frequency, and variability. The intent of  implementing such an advanced system was to relieve the inventory planners from the daily grind of reviewing and approving thousands of recommended replenishment suggestions. The idea was to automate the largest volume of regular replenishment decisions so that the human intervention could focus on the most difficult replenishment issues, which were kicked out for human attention by the system based on a custom criteria. This was obviously supposed to reduce the inventory in the network, reduce capital requirements and enhance profitability. Care to guess the reaction from the planners?

It was almost a revolt against the system! They had been running their business for decades and some had obviously been successful at improving the results over time. Nobody appreciated the idea that a machine could do a better job than they did. Definitely, nobody wanted to delegate the controls over to a black-box that they did not understand or controlled directly, while their success depended on the results produced by this box. It was finally decided that even though it was not needed, the planners will continue to review and approve all replenishment suggestions whether or not they were identified for human intervention. So much for the technology!

Gut versus data. Humans understand the gut, perception, intuition. But we are still working on understanding and accepting data. Computers (and data) are a relatively new phenomena that humans have had to deal with. Therefore, even though most corporate executives would claim that their decisions are all based on old-hard data, the reality is still a little mixed up.

Want proof, just check the latest report from RSR. In a report on how retailers are using data, RSR reports that almost half the supply chain decisions involve human “intuition”, and that is for the Retail Winners! For the laggards, the percentage is even larger.


Intuition/experience probably has a role in a function like merchandising, say in deciding on the new product or assortment, where the success of the decisions relates to some extent the emotional response that these new products/assortments would produce in the consumer. But in a function such as supply chain, which is largely objective and driven by hard data, the percentages above reflect our inherent fear of the “black-box”. This likely also points to the extent to which our retail supply chains are potentially less-than-optimal.

As an executive who would rather run their operations with data, what could you do? Change the success metrics. Most people change their behavior based on how they will be measured. In the retailers story above, we started measuring by planner, the number of manual interventions and the impact of such interventions: What happened to the inventory availability, turn-over, and overall profitability for a product/class if they were left untouched versus after the manual overrides. In most cases, the data showed that manual intervention deteriorated the overall results for the planner. And that was that, the battle was won, the black-box got established finally, albeit with some human intervention.

Related Articles:

Other Sources:

Want to know more about supply chain processes and supply chain strategy? Check out my books on Supply Chain Management at Amazon.
© Vivek Sehgal, 2014, All Rights Reserved.


  1. Hi Vivek, yes, but...

    Whenever we treat one part of the supply chain in isolation it does seem that there is an 'obvious' answer that can be provided by a black box approach. But so many decisions we make in one function have a knock-on effect in other functions, which we seldom take into consideration when we focus on a 'narrow' set of decisions.

    Since you use inventory as an example, I know of several cases where optimizing inventory has led to less than optimal use of manufacturing assets. Of course the next step is to 'rationalize' the manufacturing capacity, but what if you get caught with a rapid and big shift in mix that has little net effect on inventory value, but a big shift in capacity needs?

    All too often we assume in supply chain that we have 'a' lead time - a single value - or a yield, or a throughput, or a batch size, or ... when in reality all these values are ranges. And then we calculate an 'optimal' one-number plan from these approximate input data, generating a result that is ostensibly more accurate than the input data.

    We should be focused on a risk adjusted trade-off analysis across the supply network rather than a narrow functionally focused optimization. Of course people need systems to crunch the data. But I do not believe that we are served well by an approach that takes the person out of the loop. We need to couple machine speed with human judgment to make these trade-offs across competing objectives, the relative value of which changes based upon business conditions.

    - Trevor Miles (Kinaxis)

  2. Great article, Vivek. Especially in an area such as merchandising or purchasing where people are hired and promoted for their "understanding of the business", it's always a challenge to get people to let go of some of the old behaviors (and fears) and rely on the data. I think this bias is heavily ingrained in American culture, all the way back to the John Henry myth, right up to the present Robocop; the sentiment being that in a contest between human and machine (or computer) we always want the human to win.

  3. Well said. I had a similar experience in 2001-2002, while implementing a B2B interface with one of our customers. Our systems could acknowledge an order in real time with an Available-To-Promise date. Our planners, however, did not trust the system, and asked us to insert a 24-hour delay so that they could check themselves.
    At first it seemed they were right: we had a few instances where the recommended date was incorrect. Looking closer, however, we realized that the problem was caused by incorrect master data, which the planners (who else?) were supposed to maintain. Oh well... :)

  4. Living the dream (nightmare) everyday.
    How to effectively change from 'intuition' to 'information' driven attitude is the key.. and probably the mystery of change management as well.

    Great post. Landed into your blog thanks to random google direction on a related topic. Excellent work.