Measuring 7-Eleven marketing with test and control areas

Short answer: Choose one area that runs the campaign and a similar area that doesn't, then compare how much sales per store per day changed in each. The difference is the effect you can reasonably credit to the campaign. It is only trustworthy if price, promotions and shelf placement didn't change during the test.

Looking only at "before vs after" isn't enough, because convenience store sales rise and fall with seasons, holidays, weather and what competitors do. If the campaign month was already a strong month, every campaign looks like it worked.

Step 1: Choose a matched pair of areas

• Pick areas with a similar number of stores in stock and similar sales per store per day before the test
• Keep store types alike, for example office districts against office districts, not against residential areas
• Avoid areas where many new stores are about to open during the test

TikTok ads can be targeted by location, so they can separate test and control areas. Creator content spreads everywhere, so count it as a shared effect, not one limited to the test area.

Step 2: Collect data before you start

Use at least 4–8 weeks of data before the start to see whether the two areas move together. If they already rose and fell differently before the test, change the pair.

Step 3: Lock other variables

During the test, change only one thing: campaign or no campaign. If price, in-store promotions, placement or packaging change at the same time, you can't tell what moved sales. If a change can't be avoided, make it in both areas and write it down.

Step 4: Calculate

Hypothetical example:

Both areas grew, but the control area grew because of general factors, so the campaign effect is about 0.08, not 0.12. Looking only at before and after would overstate it by half.

Step 5: Record disturbances

Note everything that happens during the test, such as stock-outs in some stores, competitor promotions, long holidays or news about the category. These don't make the result useless, but they help explain unexpected numbers.

When you can't get data by area

If the distributor can only provide monthly order totals, compare the order trend before and after the campaign roughly, and label it as an estimate. In our 7-Eleven work we plan this kind of evaluation from the start, because store-level data may be incomplete, and we don't conclude more than the data supports.

Limits

• Figures in the table are hypothetical
• No two areas are identical; the result is a better estimate than a guess, not a scientific experiment
• Content that goes viral across areas also reaches the control area, so the measured effect is usually lower than the real one

FAQ

How long should a test run? At least one cycle in which shoppers have a chance to see the content and visit a store, usually 4 weeks or more. Products bought less often need longer.

How many stores do we need? The more the better, because per-store sales in convenience stores are usually small numbers. With too few stores, random variation hides the effect.

Does this work for TikTok Shop? The same principle applies, but TikTok Shop already has order data in the system, so tests can run in shorter cycles.

Sources

• This design is the approach the Flowup team uses to plan evaluations; it does not rely on external data.

Related guide: Listed in 7-Eleven Thailand but Not Selling?

What Flowup can do for you: Services · Contact Flowup · Insights