A full delivery-app dashboard does not tell you how many orders the app added. Some guests may be trying your restaurant for the first time; others may have ordered from your website or by phone anyway. The useful number is the estimated increase in total orders across channels, not just the number placed through the app.
Start by separating first-observed and repeat guests, then compare total orders over a sensible period. If you can run a controlled holdout, use it to estimate what would have happened without the app promotion. If you cannot, calculate a range based on explicit assumptions. Neither method identifies every guest’s intent, but both give you a better buying decision than treating every marketplace order as new demand.
Identify first-time and repeat cohorts
A cohort is a group of guests or orders that share a starting point. For this decision, use the first order you can observe and the channel on which it appeared. Track what happens afterward across your direct and marketplace channels.
Ask your marketplace account contact which order and guest-level reports are available, what date range they cover, and whether the same guest can be recognized on repeat orders. Pull direct-order history from your ordering provider and include phone or in-person records if you can reliably connect them. Your manager or bookkeeper can supply order totals, refunds, food costs and campaign charges.
Keep these groups separate:
- First observed on the marketplace: No earlier order appears in the records you can match. This is not proof that the person is new to the restaurant.
- Known direct guest ordering on the marketplace: A match appears in your direct records before the marketplace order. The order may have shifted channels, but the match alone does not prove it would have happened without the app.
- Repeat marketplace guest: A guest with an earlier marketplace order. Repeat behavior can be valuable, but it is not automatically incremental.
- Unmatched or unknown: An order you cannot connect to a useful guest history. Keep it visible rather than forcing it into the new-customer group.
Use only identifiers you are permitted to use, and ask providers how their matching and reporting work. Different names, phone numbers, household members and missing history can all distort a match. A person may have eaten in your dining room for years without appearing in an online-order file.
Record both guest counts and order counts. One new guest who orders three times is one first-observed guest and three orders; those are different measures. Also record when the observation window begins. A guest who ordered before your records start may look new when they return.
Choose a comparison period that reflects service
Pick a period long enough to include normal weekday and weekend patterns. Compare like days and service hours, not a busy holiday weekend against a quiet midweek stretch. If your marketplace campaign runs only at dinner, review dinner orders across all channels as well as the full-day total.
Before the period starts, write down the app placement or promotion being evaluated, its dates, and the channels included. Include direct website orders, phone orders and other marketplaces where you have dependable counts. Decide how you will treat canceled orders and refunds, and apply that rule consistently.
Note meaningful changes such as a menu launch, changed delivery radius, staffing shortage or bad-weather closure. These can move total demand without being caused by the marketplace. A comparison with several changes at once can still guide a decision, but it deserves a wider estimate range.
Capacity matters too. If the kitchen is already turning away orders at peak times, an app order may displace another order rather than add to total output. Compare busy and quiet service windows separately, and check whether prep capacity or delivery coverage changed. An increase at lunch may be useful even if dinner is already full.
Your restaurant website and direct online ordering records can form part of the baseline. Make sure the reports cover the same dates and use comparable order definitions before adding them together.
Estimate shifted orders with a holdout or a range
A holdout is a comparable group that does not receive the marketplace activity being measured. For example, if a marketplace offers a controllable promotion, ask whether similar delivery areas can be assigned to promotion and holdout groups for the same dates. Compare the change in total orders across all channels in each group, not just the app orders in the promoted area.
Choose areas with similar earlier order patterns and enough activity to make the comparison useful. Guests may cross area boundaries, see other advertising or order from multiple addresses. Small counts and unequal neighborhoods can also make the result unstable. If you cannot make the groups reasonably comparable, do not present the difference as a precise causal result.
A simpler before-and-after comparison is often easier to run, but it is more exposed to seasonality and other changes. Without a credible holdout, use assumptions rather than pretending to know each guest’s alternative. The worksheet below is a hypothetical one-month example, not a measurement of Fleksa or any marketplace.
| Worksheet item | Illustrative input | What it tells you |
|---|---|---|
| Marketplace orders | 100 completed orders | Observed volume, not necessarily new demand |
| First observed on marketplace | 25 guests | Guest count; some may have ordered before the available history |
| Assumed shifted direct orders | 40 orders | Estimated orders that would otherwise have come direct |
| Estimated incremental orders | 60 orders | 100 minus 40; depends on the shift assumption |
| Marketplace contribution | $12 per order | Sales less assumed variable order costs and marketplace charges |
| Direct contribution forgone | $16 per shifted order | Estimated contribution from the direct orders displaced |
| Fixed campaign cost | $300 for the month | Additional campaign charge included in the decision |
The 25 first-observed guests are not the same as 60 incremental orders. Some may already know the restaurant, and repeat guests may place incremental orders too. The 40 shifted orders are an assumption for the financial calculation, not a count proven by guest matching.
Under these assumptions, 100 marketplace orders contribute $1,200 before the fixed campaign cost. The 40 displaced direct orders would have contributed $640. After the $300 campaign cost, estimated incremental contribution is $260: $1,200 minus $640 minus $300.
The example includes assumed variable order costs and marketplace charges in the $12 figure, and variable direct-order costs in the $16 figure. It includes the additional fixed campaign cost and assumes no exception costs, such as unusual refunds or extra staffing. It excludes existing fixed costs that would not change with the campaign. Replace each input with your own figures and include any additional fixed, variable or exception costs the decision would create.
Now stress-test the shift assumption. If 20 of the 100 orders shifted, the same calculation gives $580. If 60 shifted, it gives a loss of $60. That range shows why counting app orders alone is not enough: the decision changes when the likely substitution rate changes.
Set an acquisition-cost limit and decide what to do
Set a maximum spend from the contribution you expect the activity to add, while allowing room for uncertainty. In the worksheet’s middle case, marketplace contribution less forgone direct contribution is $560 before the fixed campaign charge. At the stated $300 charge, the estimate is positive; at more than $560, it would be negative before any additional costs. A cautious owner might set a lower limit because the shift rate is uncertain and exception costs were assumed to be zero.
This is a limit for that specific month and its assumptions, not a universal cost per new customer. Do not divide campaign spend by first-observed guests and call the result a proven customer acquisition cost: some guests may not be new, and the campaign may not have caused every observed order. If you track later orders, report the observation period and the contribution from those orders separately. Avoid assuming that every first-time guest returns.
For your own worksheet, ask each provider for a written breakdown of promotion charges, order-based fees, payment processing, delivery or dispatch charges, and any other costs that apply. A quoted processing rate alone is not the all-in operating cost. If a guest pays a delivery charge, show it as a customer cost and account for how much, if any, offsets your expense.
Direct ordering also has costs. Fleksa’s plans page, checked November 2026, lists Fix at $199 per month per location with 0% Fleksa commission on direct online orders. Processing, delivery or dispatch, hardware or add-ons, and applicable sales tax are extra. That is not a marketplace quote or a complete per-order cost. For a closer look at the separate channel math, use the direct versus marketplace contribution guide.
Use the worksheet to review your current setup for one defined period. Then ask Fleksa about direct ordering and verify US availability for your location if it is relevant to your comparison.
Frequently asked questions
Is a first marketplace order always an incremental order?
No. The guest may have ordered directly or visited before, and may have ordered again without the app. First-observed status is useful for tracking, but it does not establish what would have happened otherwise.
What if I cannot match guests across channels?
Keep unmatched orders in their own group. Compare total channel orders over a consistent period and calculate results under several plausible shift assumptions. Treat the result as a range, not a verified customer count.
Should I stop a campaign with negative estimated contribution?
Review the inputs and any longer-term repeat-order evidence first. If the estimate remains negative under assumptions you consider reasonable, pause or change the campaign and measure the next period rather than relying on marketplace volume alone.
