Guides
Attracting Coffee Shop Customers by Testing One Change in One Daypart for Two Weeks
Attract more coffee shop customers by testing one change in one quiet daypart for two weeks, counting customer objections, and checking permissions first.
What to take away
- Run one experiment at a time in one daypart. Two changes at once give you a result with no cause.
- Start with the block that has spare capacity. New demand during a peak you cannot serve becomes a longer queue.
- Write down, before you start, the result that would make you keep the change and the result that would make you stop.
- Most of the available gain is in conversion of people already walking past, not in reaching strangers across a city.
- Two weeks is the minimum window. One week measures the weather.
Pick the block before you pick the idea
Export transactions per open hour for a month and look at the shape. You will usually find one dense block and a long stretch where the room is quiet and the rent runs anyway.
Transactions per open hour
- Morning peakdense block
- Middaymoderate
- Afternoonquiet stretch
- Eveningquiet stretch
Aim experiments at the quiet stretch first. That is where you have bar time, seats and attention to spare, so a new customer costs you nothing extra to serve. Aiming at the peak instead produces a longer wait, since the ceiling there is throughput rather than interest.
The experiment protocol
- Choose one block and one change.
- Write the hypothesis in one sentencewhat you expect to move, and by roughly how much.
- Write the stop conditionthe result that would make you revert.
- Brief the whole team, because an experiment nobody in the shop knows about is not running.
- Run two full weeks, unchanged.
- Read three numberstransactions in that block, average sale in that block, and how long the peak queue takes to clear.
- Keep, revert or adjust, and write down the reason.
Step six includes the peak queue deliberately. A change aimed at the afternoon can still slow the morning, usually by adding a menu item or a step the bar has to carry all day.
The experiment protocol
- Choose one block and one change
- Write the hypothesis in one sentence
- Write the stop condition
- Brief the whole team
- Run two full weeks, unchanged
- Read three numbers
- Keep, revert or adjust
Changes worth testing, from cheapest upward
Changes worth testing
| Change | What it tests | The measure |
|---|---|---|
| Board rewritten so a stranger can read it | Whether confusion is costing conversion | Transactions from new loyalty sign ups |
| A visibly fast item, named as such | Whether the wait is the objection | Transactions in the peak, and queue clearing time |
| Dairy alternatives and prices made visible | Whether people are joining a queue to be disappointed | Counter questions, counted by staff for a week |
| A reason to come in the afternoon | Whether the quiet block has latent demand | Transactions in that block |
| A pre order route promoted to nearby offices | Whether group orders can move out of the peak | Number of standing office arrangements |
| Opening earlier by half an hour | Whether demand exists before you open | Transactions in the new half hour |
Start at the top of that table. The first three cost nothing and address objections most shops never realize they are creating.
Changes worth testing
Change
- Board rewritten
- Confusion costing conversion
- Visibly fast item
- Wait is the objection
- Dairy alternatives visible
- Queueing to be disappointed
- Afternoon reason
- Quiet block has latent demand
- Pre order route
- Group orders can move out of peak
- Open earlier
- Demand exists before you open
What it tests
- Board rewritten
- New loyalty sign ups
- Visibly fast item
- Peak transactions, queue time
- Dairy alternatives visible
- Counter questions counted
- Afternoon reason
- Transactions in that block
- Pre order route
- Standing office arrangements
- Open earlier
- Transactions in new half hour
The measure
- Board rewritten
- Visibly fast item
- Dairy alternatives visible
- Afternoon reason
- Pre order route
- Open earlier
Count the objections before designing anything
For one week, have whoever is on the counter write down every question a customer asks. Not the orders, the questions.
Count customer objections
- Have counter staff write every question
- Not the orders, the questions
- Run for one week
- Expect the same four or five
- Fix the silences on the board
You will get the same four or five all week. They are your silences: things the board or room does not answer. Fixing a silence converts people already standing in front of you, the cheapest customer there is.
The board can only promise what the shop can actually produce and hold. Check any new item against what the shop stocks and rotates before it goes up.
Do not spend on reach a walking catchment cannot use
A coffee shop serves people who can reach it in a few minutes on foot. Advertising across a city buys attention from people who will never walk that far.
Spend instead on being findable and legible at the moment of decision: a listing that matches the shop, a window that says what you are, and a team that can describe the place in one line. The frequency argument behind all of that is in how a coffee shop builds a base of regulars.
Do not promise what you are not permitted to do
Before promoting a new service, check it is one you may run at that address. Extended hours, outdoor seating, trading at an event and adding food preparation can each need an approval, and running a campaign around an unapproved service is an expensive way to find out. Track those in the register of permissions the shop keeps.
The same applies to what the campaign says. Claims about origin, freshness or method have to be truthful and supportable, and review solicitation is regulated. The Federal Trade Commission's guide to soliciting and paying for online reviews sets out what it expects, and Google's guidance on improving your local ranking covers its own product.
When the experiments stop paying
If several honest experiments in a row produce nothing, marketing is probably not the constraint. The limit may be peak capacity, so the answer is bar throughput. It may be the team: recognition and consistency are service outputs that depend on how the shop hires and trains.
Or the catchment may simply be worked out, and that points to a new channel or another site and belongs in what expansion actually demands.
Common questions
How many experiments can I run at once?
One per block, and preferably one in total. Overlapping changes produce results you cannot attribute, and unattributable results teach nothing.
What if a change works but only in week two?
Extend to a third week before deciding. A single good week can be a holiday, a road closure or the weather.
Is a discount a valid experiment?
It is, and it is the one to run last. A discount usually tells you that people will take a lower price, which you already knew, and it costs margin on customers you had anyway.
