A forecast arrives
Yours, or your customer's
Silurian Assay
Not every product can be forecast.
Assay reads a demand extract and tells you what is wrong with the file, which products you can forecast, which you never will, and which you have not got the history to judge. Minutes rather than hours, and the same method every cycle.
Before a forecast becomes a plan, somebody has to decide whether to believe it.
That work happens every cycle. Tidying up the extract, comparing it to last month, working out from memory which lines never behave. It is not in anyone's process document and it has no name, so it gets done differently every time and the reasoning leaves the building with whoever did it.
It is also the cheapest place in the whole chain to catch a problem, because nothing has been bought yet.
The largest forecasting contest ever run
92.5%
of the teams that entered produced forecasts less accurate than a basic weighted average.
Two thirds were less accurate than repeating last year's pattern.
Half were less accurate than assuming next month looks like last month.
People who forecast for a living, trying hard, on clean data.
The work you do now
hours, and an answer nobody can check
Pivot tables, judgement, or both
Tidy up the extract. Compare it to last month.
Work out from memory which lines look wrong.
Hours of work. A different answer depending on who did it.
Into the plan
Agreed, then committed
Material bought. Capacity booked.
Material bought. Capacity booked.
Stock carried against products nothing could ever predict.
Time spent on the wrong products.
No way to challenge the forecast you were sent.
The same work with Assay
minutes, and an answer you can show anyone
Upload the extract
What is wrong with the file.
Which products you can forecast.
Which ones you never will.
Which ones you have not got the history to judge.
Minutes. The same method every cycle.
Into the plan, knowing
which numbers to trust
which numbers to trust
Agreed, then committed
Material bought. Capacity booked.
Material bought. Capacity booked.
Stock carried for reasons you can explain.
Time spent where it makes a difference.
Evidence to challenge the forecast you were sent.
Does this look familiar?
The 92.5 percent were not bad at forecasting. They were forecasting products nothing could predict, because nobody had told them which ones those were.
The cost never shows up in the forecast. It shows up later, as stock you cannot sell, service you cannot hold, and an argument with your customer you have no evidence for.
Better forecasting is not the answer for every product
The competition above tells you something the leaderboard does not. The gains available on sparse, irregular products were close to nothing: around three percent at the level individual products are actually planned at, against forty percent higher up the aggregation.
The products that make your forecast look bad are the products where trying harder helps least.
The lesson is not that forecasting is pointless. It is that effort spent on the wrong products produces nothing, and nobody is telling you which products those are.
When the forecast comes from your customer
Read the supply agreement. Beyond the first few months it almost certainly says the forecast creates no binding obligation and is provided for planning purposes only.
Your customer has told you in writing not to rely on it. You buy material against it anyway.
There is a reason those numbers run high, and it is not dishonesty. A customer who wants capacity held has every reason to ask for more than they need, and it costs them nothing to do it.
So you are committing money against a number that is not binding, produced by someone with a reason to round it up, and checked by nobody.
Ask how often those forecasts turned into orders last year. Most companies cannot answer, because nobody has ever counted.
What you get
- What is wrong with your file, in plain terms, before anything else happens
- Which products you can forecast, which you never will, and which you have not got the history to judge
- How much of your volume sits in each group
- A short list of products to raise with whoever sent you the forecast, with their own history as the evidence
- A record of the run you can repeat next month and get the same answer
What it will not tell you
It will not tell you what stock to hold. It tells you which products need an arrangement with a customer rather than a number, and stops there.
It will not improve a forecast. It tells you where improving one is worth the effort, and where it never will be.
And it will not tell you what you want to hear. Every company selling forecasting software has a reason to say your whole portfolio is forecastable. This is the opposite service.
What comes back
Your portfolio, split three ways, with the volume in each group.
One page you can put in front of a customer, and a record of the run you can repeat next month and get the same answer.
See a sample analysisPercent of volume
- 55% can be forecast
- 20% never will be, and need an arrangement instead
- 25% not enough history to judge
Illustrative split. Every portfolio comes back different.
Next