Start with guests who used to visit repeatedly, have missed their usual return window and can be contacted through their chosen marketing channel. Set aside recent service-recovery cases and guests already receiving another promotion. Then hold out a comparable group so you can see whether the message brought back more guests than would have returned anyway.
The right waiting period depends on how often a guest normally visits. A weekly regular who has been away for two months is a different prospect from a guest who visits once each season. Restaurant win-back campaign segmentation works best when those patterns shape both the audience and the send date.
Define repeat guests and an inactivity threshold
Before choosing a tool or writing an offer, decide what counts as a visit. For a direct-order restaurant, that might be a completed pickup order. For a dining room, it might be a completed reservation tied to a guest record. Include only activity you can identify consistently; a reservation that was canceled or a message opened is not a completed visit.
Bring the relevant records together where possible. A guest may dine in and order pickup, but appear under different contact details. Ask whether records can be matched and corrected without combining two different people. If walk-in purchases cannot be connected to a guest record, treat the resulting visit history as incomplete rather than assuming every recorded absence is real.
Use a threshold that gives a guest time to miss their normal pattern. Someone who usually orders every Friday may merit a check sooner than someone who visits for birthdays. Avoid calling every guest inactive after the same number of days merely because the software offers that filter.
Hypothetical worksheet: On November 11, an operator reviews completed visits from the previous 12 months. A repeat guest has at least two completed visits in that period. The operator separates frequent guests, with four or more visits, from occasional guests, with two or three. These are starting rules, not universal restaurant benchmarks.
| Guest pattern | Last qualifying visit | Decision on November 11 | Why |
|---|---|---|---|
| Four or more visits | September 13 through October 12 | Consider for the current send | Absent 30–59 days after a frequent pattern |
| Four or more visits | August 14 through September 12 | Consider separately | Absent 60–89 days; may need a different message |
| Two or three visits | August 14 through September 12 | Consider for the current send | Allow a longer gap for a less frequent pattern |
| Any repeat pattern | October 13 or later | Wait | Fewer than 30 days have passed |
The table shows eligible timing windows before exclusions, not a list to message immediately. A four-visit guest last seen September 20 enters the first group. A two-visit guest last seen on the same date waits until the longer window begins. Someone absent for 90 days or more can be reviewed separately; the worksheet does not assume that a longer absence makes a guest a better prospect.
Review the date ranges each time you prepare a send. A saved audience that updates automatically can be useful, but check whether it uses completed visits, the correct location and the intended lookback period. If your records cover only online orders, name the segment accordingly rather than treating it as the whole guest base.
Remove guests who need a different response
Apply contact and service exclusions before estimating your audience size. A usable email address or phone number is not, by itself, permission to send marketing. Keep guests out of a channel unless their recorded preference permits that type of message. Also remove people who have opted out or whose preference cannot be determined. Ask a qualified adviser about applicable consent and contract questions when needed.
Next, identify recent service-recovery cases. A guest who reported a missing item or a poor experience may need a personal response, not a generic promotion. In the worksheet, the operator sets aside guests with an unresolved complaint and anyone whose case was closed in the past 30 days. The team can review those records later; the rule does not assume a resolved ticket means the guest is ready for marketing.
Check for other reasons a return message would be poorly timed. Exclude a guest who already has an upcoming reservation if your records show it. Consider whether a location closure, changed hours or a temporarily unavailable menu makes the message misleading. If a guest has an active issue with an order, route it to the staff member handling that issue.
Finally, check competing campaigns. A guest receiving a birthday offer, loyalty reminder or recent promotion may see a win-back message as one more interruption. In this worksheet, anyone sent another promotion in the past 14 days waits. Set an owner for that rule and ask whether the software checks sends across channels or only within one campaign. If it cannot check across channels, reconcile the lists before sending.
Write the exclusions down beside the audience rules. That makes it easier for a manager to explain why two guests with the same last visit date received different treatment. It also makes the final count more useful: the reachable audience is the group remaining after the visit, preference, service and campaign checks.
Choose the first segment and message date
For the hypothetical November 11 send, start with the frequent guests absent 30–59 days who pass every exclusion. Their prior pattern makes a recent break easier to recognize, while the shorter gap leaves room to learn before contacting a broader group. This is a starting choice based on the worksheet's assumptions, not a promise that those guests will respond best.
Keep the 60–89-day frequent group separate. They have a longer absence and may need a different reminder or a closer look at changes in their ordering habits. Keep occasional guests separate as well: their longer normal gaps make an early reminder less useful. If the first group is very small, consider whether to wait for a larger group rather than combining audiences with different patterns simply to raise the send count.
The choice changes when the restaurant's own records show a different rhythm. If frequent guests typically return every two months, the 30–59-day group may still be on schedule; shift the first window later. If most repeat guests visit weekly, waiting two months could be unnecessarily slow. Compare the proposed threshold with actual gaps between completed visits, and account for seasonal patterns before changing it.
Choose a message that fits the reason for contacting the segment. A straightforward reminder about ordering or a current menu may be enough to test interest. If you consider an offer, decide who can redeem it, when it expires and whether it can overlap with another promotion. The guide to loyalty reward cost can help frame that decision. Do not count points or a stated reward equivalent as a reward the guest can actually redeem without checking the program rules.
Before buying software, ask whether staff can filter by completed visit count and last visit date, suppress overlapping audiences, review the final list and schedule or stop a send. Also ask what can be exported for measurement. Fleksa's restaurant marketing automation page describes email and SMS campaigns and segmentation; check the specific controls and setup against your workflow.
Measure incremental returns with a holdout
A return after a message does not prove the message caused it. Some guests would have come back anyway. Before sending, randomly divide the eligible segment into a message group and a holdout group that does not receive this win-back message. Keep the same eligibility rules for both, and avoid sending the holdout a substitute promotion during the measurement window if you can control it.
Here is a separate hypothetical measurement example. After all exclusions, 100 eligible guests are randomly split: 80 receive the message and 20 are held out. During the next 30 days, 12 messaged guests and two holdout guests each make at least one completed qualifying visit. The return rates are 15% and 10%, respectively. The difference is five percentage points. Applied to the 80-person message group, that difference suggests four incremental returning guests, not 12.
Those small counts can move sharply with one or two visits, so treat the result as a directional estimate rather than a dependable forecast. Keep the window, qualifying visit definition and split consistent when comparing later sends. Track guests, not just transactions: a guest who returns twice is still one returning guest for this measure. You can separately review completed orders or sales, but sales are not the same measure as incremental guests.
Record the send date, audience rules, exclusions, message, offer and holdout method. Note other events that might affect both groups, such as a holiday or a change in hours. If a guest cannot be matched reliably to a later visit, count that limitation when interpreting the result. An online ordering system can be one source of completed direct-order records; ask how those records can be matched to the campaign audience and exported.
Next step: Write your repeat-guest rule, first inactivity window and exclusions on one page. Then compare the marketing workflow against your operating requirements with Fleksa, including audience review, suppression and holdout reporting.
Frequently asked questions
How long should a restaurant wait before sending a win-back message?
Use the guest's usual visit pattern as the starting point. A frequent guest may warrant an earlier check than an occasional guest. Compare your proposed window with completed-visit gaps and adjust for seasonality.
Should every inactive guest receive a discount?
No. Start by deciding whether the guest is suitable to contact and what reminder fits. If you use an offer, check its redemption rules, overlap with other promotions and cost before sending.
What if my restaurant cannot create a holdout in its software?
Ask whether you can export the eligible audience, randomly select a holdout and suppress it from the send. If you cannot reliably separate and track both groups, a post-send return count alone will not show incremental impact.
