Estimate your mobile app ROI

From the customers you already have buying more often and spending more each time. That is the whole mechanism. A slice of your base installs the app, and because it sits on their home screen and remembers their card, the gap between their orders shortens and their baskets get bigger. You can add zero new customers and still see a real return. The calculator models exactly that and nothing else, which is why new-customer acquisition never appears as an input.

No. They apply only to the customers who adopt the app, which at the default settings is a fraction of your revenue, not all of it. Non-adopters do not change their behaviour because an app exists. One simplification is worth knowing: within the adopting cohort the lifts are applied to all of that cohort’s spend, as though they never buy on desktop again. In reality some of their orders still happen on the website, so the true lift is a little smaller. The causal-share haircut absorbs part of that, and it is one reason we keep that setting conservative.

Because a share of revenue smuggles in the answer. If you tell a model that 10% of revenue runs through the app, you have already assumed the outcome you were trying to calculate. A share of people is a fact you can check against your install numbers, and everything downstream is then derived rather than asserted.

They are the two assumptions that move the answer most, so we expose them rather than bury them. The first says your app cohort was already spending more than your average customer, because installs are never a random sample: the people who bother are the ones who already love you. Note which way it moves the result, because it is easy to mistake for a conservatism: raising it RAISES the estimate, in proportion. The second throws away the share of the extra spend that was coming anyway, since a loyal customer placing their usual order in the app instead of the website has not made you a penny. Both defaults are illustrative starting points rather than measured figures, and both are meant to be overwritten if you have run a holdout or geo test.

What remains of every $100 you sell after what the product cost you, after shipping it, after payment fees and after returns. It is not gross margin, which counts only the cost of the goods and ignores the cost of getting the order to someone. The difference is not cosmetic: run the default scenario on gross margin and it reads as a small profit in year one; run it on what is genuinely left and it is several thousand dollars down, with payback most of a year later. That one number multiplies the only figure that counts, so this calculator asks for the honest version, and asks for it once rather than offering you a choice of definitions. Picking a category loads a starting figure, but this is the number most worth replacing with your own.

No, and that is a limitation we would rather state than hide. Nothing in the model links your promotion budget to the adoption rate you set, so the two are independent inputs. You can assume a lot of adoption on very little spend and the model will not stop you. It will warn you, because adoption near the top of the range takes sustained push, CRM and merchandising behind it.

Lifetime-value effects beyond the three-year horizon, which usually favour the app and which we would rather omit than estimate. Support-cost savings, for the same reason. And anything specific to your stack, catalogue size or launch timeline. This is a scenario estimate: not a forecast, and not a floor. Actual results may be higher or lower.

They are illustrative starting points, not measured benchmarks, and the tool labels them that way. We have no published dataset behind the basket lift, the frequency lift, the adopter-quality multiplier or the causal share, so we do not claim one. They exist so the form is never blank, and so the downside and upside cases have somewhere defensible to move between. Every one of them is meant to be overwritten with a number you have measured. The customer results further down this page are real, named and dated, but they are evidence that a lift exists, not evidence of how large it would be in your store.