New store sales forecasting for site selection

A new store has never rung a sale, so there is no operating history to extrapolate from. Forecasting fills that gap by reasoning from comparable locations and known performance patterns, producing a structured estimate of expected revenue before any actuals exist to confirm it.

A forecast that hides its assumptions behind a precise-looking decimal is closer to a guess than an estimate. The output worth trusting shows its work. It states a range and the confidence behind it, and it lists the inputs that move the number, so a reviewer can push on the parts they disagree with instead of taking the total on faith.

A point estimate, say 2.4 million in year-one revenue, sounds authoritative and is almost always wrong by some margin. The trouble is not the arithmetic. A single figure discards what you do not know: how close the analogs really are, how the trade area will behave, how fast the store ramps, how much demand transfers from a nearby unit.

A range, say 2.0 to 2.8 million, keeps that uncertainty in view. It shows the committee where the downside sits and lets them size the lease and the buildout against the low end rather than the hopeful middle. The band narrows as the analogs get closer and the assumptions firm up, so its width doubles as a readout of how much you actually know.

Most credible new-store forecasts begin with analogs, existing stores that resemble the candidate on the factors that move sales. You pull units with similar trade-area demographics, drive-time access, competitive density, format, and daypart mix, then treat their actual performance as the basis for the new site.

• Match on what actually moves sales: trade-area population and income, road access, nearby competition, and store format. Region is a weak proxy. A unit ten miles away can be a poor analog while one across the country sits close.
• Adjust for the gaps. No analog is identical, so when the candidate has weaker access or a denser competitive set, the analog AUV gets marked down and the reasoning is recorded next to the number.
• Work from a pool, not a favorite. Several analogs give you an AUV range, a spread of real annual unit volumes, and that spread is where the forecast range comes from.

New stores rarely open at full volume. Volume climbs as awareness builds and a base of regular customers forms, a process that can run months or even a couple of years before the unit matures. A forecast that quotes a mature run-rate as year-one revenue overstates the early cash flows that pay back the buildout.

A ramp curve models that climb. It might assume a store opens at sixty to seventy percent of mature volume and reaches steady state late in year two, with the exact shape drawn from how your own past openings behaved. Time-to-maturity matters as much as the maturity number, because a slow ramp changes the payback math even when the eventual AUV is healthy. Keep the two separate in the forecast: what the store settles at, and how long it takes to get there.

Gross sales at the new site is the wrong number for a network operator. Part of that volume is not new demand at all; it transfers from existing units nearby. Counting it as growth double-counts revenue you already had and inflates the case for the deal.

A cannibalization-adjusted forecast nets out that transfer. Geod estimates it with a network-gravity model, working out how much of the candidate trade area your own stores already serve and how demand redistributes once a new unit opens. The figure that comes out is net-new revenue across the network instead of the gross ring around the new store. A site that looks strong on its own but mostly drains the unit two miles away is a transfer dressed as an expansion, and the adjusted number is what exposes it.

Every forecast rests on assumptions: which analogs you trusted, how steep the ramp is, how much demand transfers, what the trade area will really do. What separates a defensible forecast from a number people argue over is whether those assumptions sit in writing next to the result.

Geod surfaces its inputs instead of burying them. The drive-time trade area, the demographics and competition inside it, and the cannibalization estimate each carry their inputs and data vintage, and they roll up into a brief a committee can actually read. A forecast wants the same discipline: state the band, state the confidence, and spell out what would have to be true for the high end or the low end to land.

One caveat, stated plainly. Geod is not a managed forecasting service tuned to your point-of-sale history. It gives you explainable demand inputs, trade-area sizing, competition, and cannibalization, the scaffolding a credible range is built on. If you need deep analog forecasting calibrated to a large, mature fleet with years of POS detail, a managed forecasting vendor may suit that job better, and the two approaches can sit side by side.

A forecast earns trust by what it does against reality, and a new store eventually supplies that reality. Post-opening validation is the step most teams skip and the one that pays off over time. Once the store has traded long enough to be representative, compare the forecast to actual sales and feed the error back into the model.

A common measure is MAPE, mean absolute percentage error, the average size of the gap between forecast and actual across your openings, regardless of direction. When MAPE drifts upward, the model is losing touch with how stores really perform. Look at where the misses cluster, whether by format, market type, or ramp speed, then recalibrate the analogs, the ramp curve, and the cannibalization assumptions to match. Each opening you validate tightens the next forecast, which is the reason to run forecasting as a loop instead of a one-time number.

A fit score and a sales forecast answer different questions, and treating them as the same thing trips up a lot of site-selection conversations. A score ranks fit, measuring how well a site matches your model of a good location on a scale you can compare across candidates. A forecast sizes performance, estimating how much revenue the site is likely to do in dollars, with a range and a confidence level attached.

Most of the time you want both, in sequence. Use the explainable score to screen and rank the field down to a short list, then forecast the survivors to size the deal and stress the economics. A high score is a reason to look harder. A forecast is what you bring to the committee when the question turns to how much to spend.