Find your two or three quietest hours in your till data, work out who is nearby and free then, and use AI to draft a few offers built for them: a bundle or a reason to stay, not a blanket discount. Send them through your loyalty scheme, Google profile and social posts, then compare that window with the previous four weeks.
The hours are cheap to fill because your staff and your coffee machine are already paid for. That's also the trap: an offer that mostly discounts people who'd have come anyway can lose money while looking busy. The method below keeps an eye on that from the start.
Finding the quiet hours in your till data
Most café tills can export sales by hour. Pull four to eight weeks of transaction counts by hour and weekday, excluding holidays and any week with a closure or event. Paste the export into a chat assistant and ask it to lay the numbers out as a grid, weekdays across the top and hours down the side, then check the totals against your till report. If you're on Square, its built-in AI assistant (listed as Managerbot in Square's help centre, with availability varying by region) can answer "sales by hour" questions directly.
Here's what a 40-seat café's grid might look like for its Tuesday to Thursday average (an illustration):
| Hour | Transactions (Tue-Thu average) | |
|---|---|---|
| 8-9am | 38 | Commuter rush |
| 9-10am | 30 | |
| 10-11am | 22 | |
| 11am-12pm | 20 | |
| 12-1pm | 34 | Lunch |
| 1-2pm | 26 | |
| 2-3pm | 11 | Quiet |
| 3-4pm | 9 | Quiet |
| 4-5pm | 14 | Closes at 5 |
The 2-4pm window on Tuesday to Thursday averages 20 transactions a day, against 34 in the lunch hour alone. That's the slot worth working on. Pick one window at a time; a café that tries to fix every quiet hour at once can't tell which offer did what.
The dead slot isn't always mid-afternoon. A café attached to a leisure centre, say, might find its quiet hour is 10-11am, after the school-run crowd and before lunch, while 11am is busy because an over-60s fitness class finishes at 10:45. That café's best offer has nothing to do with discounts: it's a "class finish" pot of tea and toasted teacake, ready on a reserved table at 10:50 on class days, built around a timetable it can read in advance.
Who could come in at 2:30 on a Tuesday?
An offer only works if someone is free to use it. Before writing anything, list the groups likely to be nearby and available in your quiet window. Ask an assistant to help brainstorm, giving it what's around you in general terms (offices, a school, a park, a college, residential streets) rather than asking it to guess:
I run a 40-seat café. My quiet window is 2-4pm, Tuesday to Thursday.
Nearby: [e.g. a primary school, small offices, a park, flats,
a college]. Our strengths: [e.g. good coffee, home baking, fast wifi,
big tables].
List 6 groups of people who might be free and nearby in that window.
For each: why they'd come, what they'd want, what would stop them,
and one offer idea that isn't a straight percentage discount.Part of a typical reply (sample output, illustrative):
3. College students between lectures. Why: cheap place to study near campus. Want: wifi, sockets, refills. Stop them: price, and feeling they must buy every hour. Offer: "Study stamp card", the fifth afternoon drink free. 5. Retired locals. Why: company and a comfortable seat. Want: a quiet table, table service if possible. Stop them: noise from the after-school rush. Offer: 10% off for over-65s.
Two fixes before you use this. Group 3 only exists because the prompt's example list mentioned a college; if there isn't one near you, delete it, and replace the example list with what's actually around you before you run the prompt again. And group 5's offer is a straight percentage discount, which the prompt ruled out. Ask again for that group and you'll get something better suited to them, such as a Wednesday "pot of tea and a scone, table brought to you" at 2:30, before the school-gate crowd arrives.
Typical groups for an afternoon slot: people working from home who want a change of scene; parents collecting from school around 3pm; retired locals; students between lectures; office staff on a late break; dog walkers from the park. Each wants something different. The remote worker wants a table, a socket and permission to stay. The parent wants something for the children, quickly, before they melt down.
Offers that fill a slot without training regulars to wait
| Offer type | Example | Works for | Watch out for |
|---|---|---|---|
| Time-boxed bundle | "Afternoon desk": any hot drink and a slice of cake for $8, 2-4pm Tue-Thu | Remote workers, students | Regulars who'd have bought both anyway at full price |
| Loyalty bonus | Double stamps 2-4pm | Existing customers who could shift their visit | Only works if people care about your loyalty scheme |
| A reason to stay | Reserved laptop tables with sockets, free filter refill | Remote workers | Tables still full of laptops at 4:30 when the after-school rush arrives |
| A group or event | A weekly knitting circle, a toddler story session, a quiz for retirees | Groups who'd come together | Needs a host and a regular slot to build momentum |
| After-school deal | Babyccino or hot chocolate plus a biscuit with any adult drink, 3-4pm | Parents | A queue at 3:15 with one person on the counter |
| Pre-order for collection | Order by 2pm, collect a discounted afternoon box | Office staff | Uncollected orders |
Notice what's missing: "20% off everything after 2pm". It discounts every regular who already comes in then and teaches morning customers to shift, without giving anyone new a reason to visit.
Using AI to write and target the offer
Once you've chosen one or two offers, AI earns its keep on the copy and the targeting:
- Copy for each channel. Ask for a Google Business Profile offer post, an Instagram caption, a loyalty-app message and a counter sign for the same offer, each at the right length, in your café's voice. Give it three of your past posts as examples of that voice. A first draft of the counter sign often reads like this (sample output, illustrative): "Unwind with our Afternoon Desk deal! Artisan cake and a barista-crafted coffee, just $8, all week long." That's three problems in one line: "artisan" and "barista-crafted" aren't how the café talks, and "all week long" contradicts the terms. After a quick edit: "Afternoon desk, 2-4pm, Tue-Thu. Any hot drink and a slice of today's cake, $8. Stay as long as you like." Check every draft against the terms line, because the AI treats terms as a detail and rewrites them.
- A shortlist of customers to invite. If your loyalty or till system can export customers with visit times (and marketing consent), paste the export and ask for a spreadsheet formula to flag people who have visited at least twice in the afternoon, or who used to visit weekly and haven't for 30 days. Check the formula on a few rows before trusting the list. One trap is common enough to plan for (illustrative): a café's export listed visit times as text, "14:32", rather than as spreadsheet times. The AI's formula compared them with 2pm and 4pm, found no matches, and the shortlist came back with three names out of 600 customers. Nothing looked broken; the list was just suspiciously short. Converting the column with TIMEVALUE, or asking the assistant "my times are stored as text, adjust the formula", produced 84 names. If a list comes back far smaller or larger than you'd guess, check the data types before the formula.
- Terms in plain words. "Tue-Thu, 2-4pm, one per person, not with other offers." Staff need to be able to explain it in a sentence.
Square's Plus plan ($49 a location a month) bundles loyalty and marketing tools with the till, which is one way to send targeted messages without a separate email tool. The best AI tools for cafés, grouped by task covers other options.
Putting the offer where people will see it
- Google Business Profile. You can post an Offer, which needs a title, dates and a time, and can include a description, photo, coupon code, link and terms. Set the date range; without one, posts are archived after six months.
- Loyalty app or text message to customers who've opted in to marketing, sent at the moment it's useful: 1:30pm on a Tuesday, not 8am.
- In the café itself. A counter sign and a line from staff at the morning rush ("we're lovely and quiet from 2 if you want to work") reaches people who are already regulars.
- Instagram and local groups, with a photo of the actual table and cake, not a stock image.
Getting the café ready for a busier afternoon
A quiet hour that suddenly isn't quiet exposes things nobody planned for. Before the offer goes live:
- Stock for the window. If the bundle includes cake, there has to be cake at 3:30. Many cafés bake for the morning and lunch and run low by mid-afternoon. Check what's typically left at 2pm and adjust the bake, or build the offer around what you reliably have.
- Cover at the pinch point. An after-school deal concentrates people into 15 minutes. If one person is on the counter, either stagger the offer (3:30-4pm) or move a break so two people are on.
- Tables turned and cleared. Remote workers and parents with buggies need space at the same time. Decide which tables are for laptops and which aren't, and say so on a small sign.
- Staff who can explain it. Ask an assistant for a one-line script for the counter and a two-line answer to "can I use it on Friday?". Put both on the back of the till. Filled in, they might read: "If you fancy working here this afternoon, it's any drink and cake for $8 from 2 till 4, Tuesday to Thursday" and "Sorry, the afternoon desk deal is just Tuesday to Thursday, 2 till 4. Here's a card so you've got the times."
- A way to count redemptions. A dedicated button on the till for the offer is the simplest. Without it you can't tell redemptions from normal sales, and the maths in the next section falls apart.
None of this is AI work, but it decides whether the AI-written offer turns into a good afternoon or a queue of annoyed customers and an empty cake stand.
Signs an offer should stop early
Four weeks is the normal test, but pull an offer sooner if the first week shows regulars switching from full-price visits to the offer in large numbers, a queue building at the pinch point that staff can't clear, or the window busier while the day's total takings stay flat. Each means the offer is moving money around rather than bringing in new trade.
The first sign, in numbers (illustrative): in week one the after-school deal was redeemed 58 times, which looked like a success. But 41 of those redemptions were loyalty members who had usually come in at lunchtime, and lunch was down by nine transactions a day on the three offer days. That's 27 lunch transactions lost at full price against 58 discounted ones, most of them the same people. The café changed the deal to "with any adult drink, 3:15-4pm only, for children in school uniform", which kept it for the school-gate parents it was meant for, and redemptions fell to about 20 a week, nearly all of them new afternoon visits.
Worked example: filling 2-4pm at a 40-seat café
Continuing the illustration: the café runs the "Afternoon desk" bundle ($8 for a hot drink and cake that normally cost $9.50 together) and an after-school deal for four weeks, Tuesday to Thursday.
- Before: 20 transactions a day in the window, 240 over the 12 comparable days in the previous four weeks.
- During: 336 transactions over 12 days, so 96 more than before. The offers were redeemed 140 times.
- Reading it honestly: if 96 extra transactions came in but the offers were used 140 times, about 44 redemptions came from people who'd have visited anyway. At $1.50 off each, that's $66 of discount given away.
- The gain: 96 extra transactions at an assumed gross margin of about $5 each is roughly $480 over four weeks. Less the $66, that's around $414, with no extra staff cost because the team was on shift anyway.
Also check what happened to the lunch hour and the morning. If they dropped, some customers simply moved their visit to get the deal, and the real gain is smaller. In this illustration they didn't, so the offer stays, and the next month tests whether the after-school deal alone would do most of the work.
Keeping the gain once the offer ends
A quiet-hour offer is most valuable when it turns into a habit. Build the second visit in from the start: a stamp card that only fills in the afternoon, a "see you Thursday" for the knitting group, a sign-up for the loyalty scheme at the first redemption. Using AI to bring in more regulars covers the follow-up side, and building a loyalty programme with AI personalisation goes further if your scheme is the channel doing most of the work.
A simple way to measure the habit (illustrative): of the 96 extra visits in the worked example, note how many came from people who signed up for the loyalty scheme at their first redemption, say 31. Four weeks after the offer ends, count how many of those 31 are still visiting in the afternoon at full price. If it's 12, the offer has created a dozen new afternoon regulars, which is worth more than the $414 in the four weeks it ran. If it's two, the offer only works while it's running, and the question becomes whether to keep it on permanently as a named menu item.
Review every four weeks with three questions: did transactions in the window rise against the same days before, did other hours fall, and how much discount went to people who'd have come anyway? Drop any offer that fails two of the three. If you're still deciding where AI fits in the café more broadly, what AI can and can't do for an independent café is a good place to start.
Café owners also ask
Won't regular discounts make the café look cheap?
Blanket discounts can, and they teach regulars to wait. Offers tied to a time and a purpose rarely do: an afternoon work bundle, a parents' after-school deal or a quiet-hour loyalty bonus reads as hospitality, not desperation. Keep offers specific, time-limited and few, and give them a name, so they feel like part of the café rather than a sale sign in the window.
Can AI send the right offer to the right customer automatically?
Some loyalty and marketing tools can segment customers and send messages on a schedule, and AI can help you write the segments and the copy. Start manually for the first month: pick the segment, check the list, send, and measure. Automate only once you know which offer works for which group, or you'll automate a message nobody wanted.
Can I text or email every customer in my till system?
Only those who have agreed to receive marketing. Data-protection law such as the GDPR, and rules on electronic marketing in many places, generally require consent or an existing relationship with a clear opt-out. Most loyalty sign-ups include a marketing opt-in; use that list, include an unsubscribe option in every message, and ask your adviser if you're unsure what applies to you.
Further reads
- How Spas Use AI to Personalise Offers From Treatment History — Personalised offers from visit history, in another local business.
- How Pilates Studios Use AI to Turn Intro Offers Into Memberships — Turning a one-off offer into repeat visits.
- Email Marketing Tools With AI: What a Small List Costs per Month — What sending offers by email costs for a small list.
- How Bakeries Can Use AI to Predict Demand and Cut Unsold Stock — Matching what you bake to the quieter afternoons.
- Can a Café Use AI to Take Bookings and Answer Messages? — Handling the messages and bookings that offers bring in.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Google Business Profile Help (post types and offer fields); Square's unified plans announcement (Square Plus includes loyalty and marketing tools); Square Managerbot help article. Worked example is illustrative. Checked September 2026.