I have worked with a company related to TPM and engineering activities. While I was doing this, unknown to me, they were also moving from a localized forecast and scheduling model to a centralized model. I started asking questions when site after site I went to had inventory in the production areas, outside warehouses, and Lunch rooms. Well, the last one’s a stretch but anyway I started asking about the issue. I figured the move to central forecasting was obvious since everyone was filling up but I was surprised about the central scheduling part. Corporate governance was driving a schedule attainment metric (moving from a day to hours) may make sense in some situations but it was creating havoc in the warehouse and operation side.

I get trying to save money on SKU level forecasts out 12 to 14 weeks. Using the resources that are doing that should be used to find better signals and help the plant react or produce at the rate of consumption , not predict the future. But at least give the local sales and marketing group as well as scheduling the ability to change the schedule appropriate for what they are seeing. Taleb in Anti-Fragile, discusses this phenomenon in depth and what I take from it is two-fold. Some items are not scalable. Going from sales reps who are talking directly with the direct customers, they come up with fancy algorithyms. My first question is how much better is your system compared to naïve or other simple models run on a smart spreadsheet?

The other issue brought up by Taleb is when one makes a system bigger, the bigger the errors. This may seem simple but it actually gets worse mathematically as connections cross disparate fields. This makes risk management not just difficult but bordering on the impossible.

What is the knee jerk reaction to this inventory growth? I have seen the following:

Sales gets pushed to make the numbers either now or in future periods / quarters.

Operations get shuddered at some point due to low market demand to plan. Read Reality did not line up with the guesses.

CapEx gets squeezed, either from the infrastructure or productivity budget. The emergency CapEx balloons as the creative writing begins.

Discount the stuff that the customer didn’t want anyway.

Tweak the math in the black box so that it “better” predicts the future.

I have not seen a company dismantle the system they are currently using for forecasting for a simpler model. I have had sales / sites / and mid-level leadership to go do what they want instead of worrying about what is coming out of the black box.

I am a grown-up and know that companies, even small ones, have to spend some portion of their time forecasting demand. I would use the simplest model that got 60-70% of the demand and then use tribal knowledge or local expertise to tweak as see fit and let operations work with scheduling to make the most of their operations.