How do you automate repeat visits without spamming customers?

The fear behind this question is legitimate. Most automated F&B messaging is a broadcast calendar wearing automation's clothes: the same promotion to the whole list every fortnight, which is why customers mute venue accounts. The distinction that matters is not automated versus manual — it is triggered versus scheduled.

A triggered message is caused by something the individual customer did or did not do. It arrives when it is relevant to them specifically, which is why triggered sends routinely outperform scheduled broadcasts by three to five times per message. It also means volume scales with behaviour rather than with your content calendar, so nobody receives a message just because it is Thursday.

Four triggers carry most of the value. Welcome, within 48 hours of a first identified visit, carrying a reason to return inside a week. Near-reward nudge, when a customer is one or two visits from a reward — the highest-converting message in most programmes, because it asks for something the customer already wants. Birthday, sent a few days ahead with a claim window, consistently returning five to eight times its cost. And cadence lapse, when someone's gap exceeds their personal norm.

Now the safety layer, which is what separates automation from spam. A hard frequency cap of one message per customer per week, across all flows. A suppression rule on every send — anyone who transacted in the last 72 hours is excluded from win-back, anyone who just redeemed is excluded from the near-reward nudge. A quiet period around sending hours. And a global kill switch for the days you are running a manual campaign, so the two do not collide.

The failure this prevents is specific and common: a customer visits on Saturday, and on Monday receives 'we miss you'. That single message does more damage than a month of good sends, because it proves you are not paying attention. Suppression rules exist to make that mechanically impossible rather than a thing you remember to check.

Judge the system on two numbers: revenue per message sent, and unsubscribe rate per flow. If unsubscribes concentrate in one flow, that flow is mistimed or mistargeted — fix it rather than reducing overall volume. Cutting frequency across the board to fix one bad flow is the most common over-correction, and it costs you the flows that were working.

This is the part where an AI layer earns its cost at independent scale: computing per-customer cadence nightly, choosing which trigger a given customer qualifies for, respecting the caps, and drafting the copy in a consistent voice. Done by hand, it is a part-time job; done by an agent, it is an approval queue you clear twice a week.