-
Your Demand Chart Hid a 132-Unit Problem. The FT’s Chart Rules Catch It
Four datasets can share the same summary statistics and still tell four different stories. So can your SKU history. Here is how to name the relationship in your…
-
What Shape Is Your Plant? In Two of the Four, Your Bottleneck Lies
Every factory game teaches you the slowest machine sets the pace. None teach you that in a V-plant, the trunk can run at 100% and convert every unit…
-
Six Times They Tried to Prove Sophistication Wins
Forty years of open forecasting competitions, and the winner’s margin over a dumb average stayed thin. Then M6 asked whether accuracy makes money. The correlation was 0.04.
-
Four Kinds of Weird in Your Price File: EUR 43,017 of Negotiating Room
A supplier asked for 20.4%. The published indices supported 2.3%. Here is the R workbench that found the gap, and the three things it honestly could not find.
-
$9 Billion for Nothing: the Stealth AI Bet That Reaches Your Loading Dock
Six labs most supply chain professionals have never heard of have raised $15.56 billion. 58% of it went to two labs that have shipped nothing at all.
-
Your Supplier Isn’t Offering a Discount. They’re Offering You 29.8%
The price-break EOQ math says buy 600 units instead of 87 and save $6,566 a year. Then a carrying rate above 29.8% flips the answer. Here are the…
-
AI Model Distillation, Explained: The Compression Trick Behind the $589 Billion Monday
A single Chinese app knocked $589 billion off Nvidia in one day. The trick behind the panic has a boring name and a fascinating story.
-
I Raced Amazon’s Forecasting AI Against a 66-Year-Old Formula. The Formula Won.
I put Amazon’s zero-shot forecasting AI against five classical models on 148 real retail series. A formula from 1960 beat a 205-million-parameter model by 14 to 16%.
-
I Let AI Agents Play a 5-Year Beer Game: Why Sharing Data Beat Buying a Smarter Model
I ran the classic Beer Game for 260 weeks with AI agents in the chairs. Giving the chain real customer demand cut in-game cost by up to 40…
-
A 3-Point Gap and a 42x Price Cut: The Open-Weight AI Landscape, July 2026
The best open-weight model now trails the closed frontier by three points of measured intelligence and costs a third of the price. For anyone routing thousands of tasks…
-
Your Exception Report Is Lying to You: How to Make ‘1% False Alarms’ Actually Mean 1%
A static ±3σ band promised 0.27% false alarms and delivered 3.80%. Here is how to turn your exception threshold from a statistics setting into an honest attention budget.
-
Factory Physics: The Laws Your Factory Floor Already Obeys
Throughput, WIP, and cycle time aren’t three dials you can set independently. They’re bound by physics, and ignoring that costs you weeks of lead time.
-
S&OP: Everyone Signed Off on the Number. It Was Still Wrong by 8.2%.
A consensus forecast measures agreement, not truth. Here is what a textbook S&OP cycle looks like, why the signed-off number still ran 8.2% high, and the four moves…
-
Does Your Forecast Beat a Sticky Note? The Placebo Test for Demand Planning
Your forecast has exactly one job: beat a sticky note that says ‘same as last year.’ Most don’t. Here’s the placebo test that proves it.
-
GLM-5.2 and the Benchmark Trap: How One Score Becomes Two Headlines
A Chinese open-weight model just topped the leaderboards and undercut GPT-5.5 on price by 7x. Then the benchmark tables started disagreeing with each other.
InPhroNeSys
Actionable Insights