The perfect forecast doesn't exist (and that's good news)
The perfect forecast doesn't exist. Our manifesto: stop chasing accuracy, and invest in the decision at the moment the plan breaks.
By Mansour Niang, CEO & co-founder, Clevizio·Published July 6, 2026·Updated October 18, 2026·8 min readKey takeaway
A good supply chain forecast is right about eight times out of ten, and going from 80% to 82% accuracy changes almost nothing: that is the 80% wall. The two remaining misses are the disruptions the model did not see coming, and they cost between 4 and 12 points of margin a year.
Thirty years chasing the perfect forecast
For as long as supply chain has existed as a discipline, it has pursued the same goal: forecast better. More data, better models, finer algorithms, and recently artificial intelligence. Every generation of tools promises the forecast that will finally be right. Yet the supply chain directors we have interviewed, more than a hundred in total, all describe the same reality: they still run their operations by reacting to deviations, not by calmly following a plan that comes true. The race for accuracy has no finish line.
The 80% wall
There is a mathematical reason for this. Over time, a good forecast is right about eight times out of ten. Going from 80% to 82% accuracy takes considerable effort and changes almost nothing day to day, because the problem was never the eight times the forecast is right. It lies in the two times it is wrong, and those two times are by definition the ones the model did not see coming: the supplier failure, the demand spike, the production hiccup. No amount of extra accuracy anticipates them.
Why will forecasting models always be wrong?
This ceiling has a fundamental explanation: a model learns from the past, and a disruption is precisely what is not in the past. The failure of a previously reliable supplier, a transport incident, a sudden market reversal leave no regular pattern in the history. The context makes things worse, since Resilinc measured 38% more disruptions in 2024 than in 2023: the share of reality that escapes the models is growing. Improving the algorithm is still useful for ordinary variability. Against the events that really cost money, it is structurally insufficient.
How much do forecast errors cost?
Those two times out of ten are expensive. In the time it takes to understand the deviation, get the right people together and decide, the bill piles up in express freight, excess stock, lost orders and reshuffled production. Across the scopes we have studied, this cost represents between 4 and 12 points of margin a year, and it does not appear on any line of the P&L. Nobody puts a number on it, so nobody fights it. It is the invisible leak in the supply chain.
What the obsession with accuracy costs
This race also has an opportunity cost, which studies help estimate. McKinsey estimates that disruptions cost companies the equivalent of 45% of one year's profits over a decade, and as early as 2021 Interos put average losses at $184 million per organization per year. While budgets are concentrated on a few extra points of accuracy, the moment that destroys margin, the moment of reaction, remains without tools, without process and without measurement. Beyond a certain threshold, the obsession with accuracy becomes futile, and above all it diverts attention from where the money is actually being lost.
Changing the playing field
The conclusion is not to abandon forecasting, which remains useful for sizing inventory and production. It is to stop expecting from it what it will never deliver: protection against the unexpected. The real performance lever is shifting from the quality of the plan to the quality of the decision made when the plan no longer holds. It is a different playing field, where you no longer measure forecast error, but the speed and accuracy of the response.
Decide, capitalize, repeat
On this playing field, three things matter. Seeing the drift coming before it turns into a crisis. Costing the impact of every possible response before deciding, so you can choose the least expensive one. And keeping a memory of the trade-off, so the next disruption of the same kind gets resolved faster. This conviction is what gave birth to Clevizio. We are not trying to make your forecasts perfect; we make you better at the exact moment they are wrong.
- Thirty years chasing the perfect forecast
- The 80% wall
- Why will forecasting models always be wrong?
- How much do forecast errors cost?
- What the obsession with accuracy costs
- Changing the playing field
- Decide, capitalize, repeat
Clevizio
Clevizio is the decision layer that plugs into your existing tools: it detects disruptions, simulates your options and keeps a record of every trade-off. See how the decision copilot works.
Mansour Niang
CEO & co-founder, Clevizio
A question about your disruptions? Book 30 minutes with me to estimate what they really cost you, with no commitment.
Frequently asked questions
Should we stop forecasting?
No. Forecasting remains useful for sizing inventory and framing production. The point is to stop expecting it to protect you against disruptions, which it cannot do, and to equip separately the moment when it is wrong.
How do you measure the quality of a decision rather than the accuracy of a forecast?
By tracking avoided cost: the difference between what a disruption would have cost without a fast response and what it actually cost. It is a decision metric, not a forecasting metric, and it is the one that weighs on margin.
Finally put a number on what disruptions cost you every year.
Let's start by estimating what your disruptions really cost you. It's free, it takes 30 minutes, and you leave with a figure nobody in your company knows today.