Predictive models
Find out what drives sales at your stores, how much a new store could sell, and how much your current stores will sell. You use it inside Pulster.
Sales, transactions, or units: burgers, bags of cement, or whatever your business needs to measure. Each variable is trained and tested separately.
The models
Store performance drivers
What you get
The internal and external characteristics most closely associated with your strongest and weakest stores.
When it fits
Before opening more stores, or to improve the ones that sell less.
Its limit
It shows what is associated with success; it does not prove what causes it.
Where you use it: Pulster NOVA
New store sales forecasting
What you get
An estimate with its margin of error.
When it fits
When you need a figure for your budget or to compare options.
Its limit
It depends on how similar the new store is to the ones you already have.
Where you use it: Pulster NOVA
New store sales ranges
What you get
The most likely range, such as high, mid, or low sales.
When it fits
When a single figure is not reliable, but a range still helps you decide.
Its limit
It is the most likely range; the store may end up outside it.
Where you use it: Pulster NOVA
Store-level sales forecasting
What you get
A forecast by store and period, with its margin of error. It uses your history and outside changes, such as competitors and mobility.
When it fits
To plan targets, inventory, or campaigns.
Its limit
A major change that was not in the data, such as construction or a new competitor, can shift it.
Where you use it: Pulster RADAR
Guide
Step 1 · What do you want to solve?
Step 2 · What would help you most?
This guide is a starting point. Before we begin, we review your data and tell you which model can be built with confidence.
How it is built
We review your data
Whether it is enough and which model fits.
We add our layers
Population, competitors, mobility, and more.
We build the model
With your data and market data.
We test it
On data it did not use, and calculate its margin of error.
You use it
Inside Pulster NOVA or Pulster RADAR.
What we need
From your company
From Pulster
Accuracy
Forecasting is hard, and we say so. Every estimate comes with a margin of error, calculated from how well the model performs when we test it on data it did not use.
96%
accuracy forecasting first-year sales at existing stores, validated in a blind test with a leading quick-service restaurant chain.
The model was not trained on those stores. The average error per store was under 4%. This test was run on existing stores and does not guarantee the same accuracy for new stores.
This result comes from the new store sales forecasting model. It does not automatically apply to other models or variables.
In the first meeting, we understand your case and see whether your data can support the model. If it can, we explain the next steps.
We tell you before we start. Sometimes we can build a sales range model instead of a single figure; other times it is better to wait for more data.
We test it on stores or months it did not use for training and compare its estimates with what actually happened. That is how we calculate its margin of error.
We review it case by case: it depends on how comparable your stores are and the data we have for that market.
Building the model is quoted based on your case. After that, you use it inside Pulster NOVA or Pulster RADAR.
