Sample data

The future widens.

Zerqano forecasts every product from its own past, and draws how sure it is: narrow for tomorrow, wider further out.

Sample data · Harbor Dome Tent, 4-person, all-season shell

Harbor Dome Tentunits sold a day, then the forecastLightGBM
TodayHow sure: the band

Every product sells its own way.

So no one model forecasts them all.

Sample data

  • Harbor Dome TentSells every day, about 24, in a weekly rhythm.

    LightGBM

  • Dayline PackSold on 31 of the last 121 days, never more than 5 at once.

    SBA

  • Beacon Camp LanternNo error measured yet, so its forecast comes without a band.

    Exponential smoothing

  • If the pack stopped sellingThe forecast would have to let go.

    TSB

A race on the tent’s own past.

Each model forecasts the last 42 days without seeing them. The one that misses least is used.

Sample data

How far off, over 42 test daysWAPE · lower is better

  1. LightGBM10.6%
  2. Holt-Winters (ETS)13.7%
  3. Last value20.4%

A complex model is used only when it beats the simple ones. When two are about as good, the simpler one wins.

LightGBM is used

Eight days in ten, inside the band.

That is what the band promises. Checked against the tent’s past forecasts, sales landed inside it on 84% of days.

Sample data

Harbor Dome Tentthe next 30 daysP10–P90
TodayTomorrow: narrowA month out: wider

84%of days inside

A bursty seller gets its own method.

The pack sold on 31 of 121 days. Models that expect a sale every day are left out of its race; SBA, built for sellers like this, forecasts it.

Sample data

Dayline Packone stick per day with a sale, 31 of 121SBA
Today

When it can’t measure, it says so.

A dash, not a made-up score.

Sample data

Dayline PackSBA

Test days
6
How far off
44.9%
Band held
—

Six test days are too few to check the band. Too little history to calibrate the range, so the band is deliberately wide.

Beacon Camp LanternExponential smoothing

Test days
—
How far off
—
Band
none drawn

Nothing measured how wrong this forecast could be, so it gets a line and no band.

When sales stop, the forecast lets go.

If the pack went 13 days without a sale, TSB would move to the front of its race. Each quiet day lowers its chance of a sale by a tenth, so the forecast falls toward zero instead of holding the old rate.

A rule in the forecasting code, shown on the sample pack

Days since the last saleforecast as a share of the old rate
SBA holds the old rateDay 13: TSB leads, 25% leftTSB falls toward zero

The real Forecasting screen.

Same tent, same race, same 84%.

The real Zerqano Forecasting page, Signals tab, with sample data: the Harbor Dome Tent, 4-person, all-season shell, LightGBM, a calibrated band, error 10.6% (WAPE) on 42 held-out days; the chart of units sold and the forecast with its P10 to P90 band; Why this model? open, comparing LightGBM 10.6%, Holt-Winters (ETS) 13.7%, Last value 20.4%; stock and planning with 216 on hand and a 9-day lead time; the 80% band held on 84% of days.The real Zerqano Forecasting page on a phone, Signals tab, with sample data: the Harbor Dome Tent, 4-person, all-season shell, LightGBM, a calibrated band, error 10.6% (WAPE) on 42 held-out days; the chart of units sold and the forecast with its P10 to P90 band; Why this model? open, comparing LightGBM 10.6%, Holt-Winters (ETS) 13.7%, Last value 20.4%

REAL SCREEN · SAMPLE DATA

Free for up to 150 SKUs

14 days of full access · No card required

Forecast your productsSee a sample planRequest a demo

Sample data · a fictional outdoor-gear store, 30 products forecast · not a customer’s results

What demand forecasting does

  1. Upload or refresh your sales history.
  2. Open a product in the Restock plan to see its demand range and forecast error.
  3. Decide whether to follow the recommendation, with the demand behind it in view.
  4. Check forecast error in Analysis to see how well past estimates held up.

Questions

Is this only for dedicated demand planners?

No. Anyone who reorders can read the demand range and forecast error beside each recommendation.

Does demand stay connected to replenishment?

Yes. The Restock plan uses each product's demand rate for days of cover, when to reorder and how much.

What if a product has no sales history?

It is shown as not planned, with a message that demand can't be estimated and a link to check its sales history.