ForecastForward viewLive
What next month looks like
if nothing changes.
Revenue, profit and orders projected forward, each with a range — and the range is measured, not chosen. It comes from how badly the same method would have missed the history it was fitted to, so a steady store gets a narrow band and a volatile one gets the wide band it has earned.
Measured, not estimated EU-hosted Read-only by default
01 / The case for it
A forecast with no range is a guess with a decimal point.
Without it
Draw a line through the history, extend it, print a number. It is the easiest thing in this product to build and the fastest way to lose the reader, because the one question they have — how sure are you — has no answer on the screen.
What Profflow does
The range comes from how badly the method already missed.
The same straight line is run against the store’s own history to see how far off it lands, and the size of those misses becomes the band. A steady store gets a narrow one; a volatile store gets the wide one it has earned.
02 / What it looks like
Next month, with the range around it
FebNext month
Solid is measured, dashed is projected, and the shaded part is the range — € on both sides of it.
The range is measured from how wrong the same method has been on this store’s own past, not chosen to look confident.
A trend fitted to one Christmas forecasts a Christmas in February, so nothing here is seasonal and the screen says so. Under four periods it returns nothing and explains why — a line through three points is a line through three points.
03 / Worth understanding
What makes a forecast worth printing
A number without a range is a guess in better clothes
Any method can produce a single figure for next month. What makes it useful is knowing how far out it has been before — so the range here comes from testing the same method against this store’s own history, not from a confidence setting.
Too little history means no forecast
Below four comparable periods the product declines to forecast rather than producing something shaped like an answer. Refusing is a feature: a projection from two months of data is a shape, not a prediction.
Seasonality is learned, not assumed
A store that sells coats does not grow smoothly into July. The fit uses the store’s own twelve months, so the shape of its year is in the projection instead of a straight line drawn through it.
04 / What it is built from
Forecast is assembled
from accounts you already run.
01Monthly revenue history02Profit history03Order history04Ad spend history05Back-test error06Period length
Forecast
Revenue, profit and ordersA measured rangeWidening with distanceA floor on the data05 / How it is worked out
From the account
to the figure on the screen.
- 01FitsA straight trend to the history the store actually has.
- 02TestsThe same method against that history, to see how far off it lands.
- 03RangesThe band from that error, widening with every period forward.
- 04RefusesTo draw anything at all on fewer than four periods.
Asked of the figures
How confident is that number?
As confident as this store’s own history allows. The band is the error the same method made against the last twelve months, widened for distance — six periods out is not as knowable as one, and a flat band would claim it was.
How Profflow reasons06 / What it lets you do
What follows
from getting Forecast right.
Revenue, profit and orders
Three forecasts from the same fitted history, each with its own band.
A measured range
Derived from back-test error rather than a confidence percentage typed into a constant.
Widening with distance
Six periods out carries a wider band than one, because it is less knowable.
A floor on the data
Under four periods there is no forecast and the screen says why. A line through three points is a line through three points.
Stated as straight-line
No seasonality is claimed. A trend fitted to one Christmas forecasts a Christmas in February.
Break-even forward
What the coming period needs to return before the advertising stops paying for itself.
07 / How the figure is kept honest
The same four rules,
on every screen.
Profflow observes the accounts it is connected to. It does not write an order, a price, a campaign or a payout back to any of them.
A charge read from an account is labelled measured. Anything modelled — a forecast, a projection, a filled gap — is labelled as such on the screen it appears on.
Follow any number to the dated rows underneath it. A figure nobody can check is a figure nobody acts on.
The product prepares a decision and shows the evidence. Nothing consequential happens until you say so.
Questions
About Forecast.
- Is it seasonal?
- No, and it says so on the screen. Seasonality needs years of history this product does not have yet. What it does instead is state the method plainly and let the range carry the uncertainty.
- Why no forecast on a new store?
- Because a forecast off three data points would be the most misleading thing this product could show. Under four periods it returns nothing and explains the reason.
- Where does the range come from?
- From back-testing. The same method is run against the store’s own history and the size of its misses becomes the band. A store whose weeks are all alike gets a narrow one.
- Can I plan against it?
- It is built for that — one central figure, one honest range, and the assumptions written down. It is not built to be right; it is built to be checkable.
One operating system
The rest of the picture,
one view away.
Each part is useful on its own. Profflow holds them in one model, so a figure here carries the context from everywhere else.
See the whole productSee Forecast on your own figures.
Revenue, profit and orders projected forward, each with a range — and the range is measured, not chosen. It comes from how badly the same method would have missed the history it was fitted to, so a steady store gets a narrow band and a volatile one gets the wide band it has earned.