How to forecast sales for a new store

Why is it hard to forecast sales for a store that does not exist yet?

An open store has a history: you can project its sales from what it sold before. A new store does not. All you have is its location, its format, and what you know about your other stores.

That is why every new store forecast is, at heart, a comparison: which of your stores does this site resemble, and how much do they sell? The forecast is only as good as those comparisons.

And getting it wrong is expensive. A mid-size store costs US$1–5 million to open, and leases run 5 to 20 years. An optimistic figure can approve a store that never pays back; a pessimistic one can rule out a good site that a competitor then takes.

Which methods are used to forecast new store sales?

There are four families of methods. They are not rivals: many chains start with analogs and move to fuller models as they add stores and data.

The best-known gravity model is the Huff model, formulated by David Huff in 1963. It starts from a simple idea: people tend to go to the most attractive, closest store, and that probability is split across every option in the area.

Which variables usually matter?

They vary by category, but these show up in almost every retail model, inside and outside the store:

• Population and socioeconomic profile within the trade area, measured by travel time.
• Floating population: how many people visit or pass through the area without living there.
• Direct competitors and complementary businesses nearby.
• Points of interest that drive visits, such as schools, offices, hospitals, and transit stops.
• Access and visibility: corner or mid-block, car access, and parking.
• Store characteristics: format, size, opening hours, additional services, and in-store partnerships.
• Your other nearby stores, because part of the sales may come from them (cannibalization).
• Upcoming changes, such as housing, office, or retail projects with approved building permits.

What should you forecast: sales, transactions, or units?

Forecast what the decision needs. If your committee approves based on revenue, forecast sales. If you worry about kitchen or checkout capacity, transactions are more useful. If you are planning inventory or logistics, forecast units.

Transactions and units tend to be more stable when prices change, because they do not depend on the average ticket. Sales in currency, by contrast, also move with inflation and price changes.

Each variable is modeled and tested separately. A model that gets sales right does not necessarily get units right.

Why should a forecast come with a margin of error?

A single figure creates a false sense of certainty: it sounds better than a range, but it hides how confident the model is.

A useful forecast says how much the store could sell and with what margin. That lets you ask the right question: does the store pay off even in the low scenario? If it only works in the high scenario, the decision is riskier than it looks.

When the margin is very wide, it sometimes helps to change the question. Instead of a figure, a sales range model estimates whether the store would land in high, mid, or low sales. For comparing sites, that can be enough.

How do you validate a forecasting model?

A model is validated on data it did not learn from. The clearest way is a blind test: set aside some existing stores, let the model estimate their sales without having seen them, and compare the estimate with what they actually sold.

When someone shows you an accuracy figure, look for three things: how it was measured, on which stores, and what its limit is. For example, this is how Pulster reports its reference test: 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.

Before you trust a model, also ask for:

• The error by store, not just the average: a good average can hide stores with large errors.
• The error on stores similar to the one you plan to open: same format, same type of area, same country.
• Confirmation that the test stores were left out of training. If the model already knew them, the test does not count.

Common mistakes in new store sales forecasting

These are the mistakes that most often inflate or deflate a forecast:

• Using your best store as the benchmark for every opening.
• Comparing with stores of a different format, size, or type of area.
• Forgetting the ramp-up: many stores take months to reach stable sales, and year one does not look like year three.
• Ignoring cannibalization and counting as new sales what actually comes from your other stores.
• Not updating the forecast when the area changes, such as a new competitor or construction, or never comparing actual sales with the forecast afterward.
• Treating the estimate as a guarantee instead of an input for the decision.