Key takeaway
Restaurant email marketing is mailing your own guest list with what's on this week and the fact that you noticed somebody stopped coming. The hard part isn't choosing what to send. It's that whoever has to send it is on a shift. So I'd build the four that fire off data the till and the booking system already hold, count lapse from a guest's own last order at 30, 60 and 90 days, and write one weekly send.
What email actually does for a restaurant.
Email marketing for a restaurant means mailing a list of addresses your own guests handed over, with this week's menu, a reminder of what they last ordered, and a line to whoever hasn't been in for a while. It's worth the effort for one reason, and it isn't reach. A restaurant's best customer is a repeat customer, and email is the only channel where the guest record and the message sit in the same place you already control.
That matters more than it did, because most of a restaurant's demand arrives without anybody walking through the door. Nearly 75% of restaurant traffic is off-premises, and off-premises makes up a larger share of sales than it did in 2019 for 58% of limited-service and 41% of full-service operators, according to the National Restaurant Association. The guest ordering from a sofa is a row in a system rather than a face the host recognizes. Email is how you reach a row.
I've been building email platforms for over a decade, and what stops a restaurant is almost never the writing. Every guide hands over the same eight campaigns, a benchmark, and a dashboard to run them in, and none asks who logs into it. The marketing team is the owner or the GM, and both are on a shift at the hour the email should go out. So this page is arranged around the data you already hold, the permission that came with it, and the sends that need nobody at a screen, which is the half you can hand to an agent.
Where a restaurant's email addresses actually come from.
The addresses arrive from four systems rather than one, and they arrive on four different permissions. Which system an address came out of decides what you're allowed to send it, and the spreadsheet it lands in won't tell you.
- POS receipt capture: the guest handed over an address so the till could send a receipt, which is a permission to send that receipt.
- Your own online ordering: the guest transacted with you directly, and a marketing checkbox at checkout is the one place a restaurant reliably gets consent worth having.
- The reservation system: the booking record is yours to use for the booking, and whether it's yours to promote from depends on what the guest agreed to when they booked.
- Delivery marketplaces: the order came through somebody else's storefront, and the guest is usually their customer rather than yours.
The fourth line decides how much of this is worth doing. A restaurant running most of its volume through marketplaces has far fewer mailable addresses than its order count suggests, and no amount of segmentation fixes a list that was never collected. The only fix is to move guests onto a surface you own: your own ordering page, a table card that offers a reason to sign up, a receipt that asks. An order through your own checkout beats the same order through a marketplace twice over, once on margin and once because it arrives with an address.
Two sources look like list building and aren't. The fishbowl of business cards by the till and the guest Wi-Fi splash page both produce addresses nobody remembers giving you, and those are the addresses that generate complaints when you finally mail them. If you use either, say what the address is for at the moment you take it, and keep the marketing opt-in separate from what the guest came for.
A receipt is not a permission.
Mail a restaurant sends splits into two kinds. A confirmation, a receipt, or a booking reminder is transactional, because the guest asked for it by doing something. A promotion is commercial, and commercial mail carries obligations the receipt doesn't.
The line moves the moment you staple an offer onto a confirmation, which is exactly what most guides recommend when they tell you to put a discount code in an abandoned-order email. A message carrying both reads as a promotion if the subject line promises one, or if the receipt isn't what the body opens with. The guest opened it for the order and got sold something, and that's the version they remember.
Being a promotion isn't a problem. It's a checklist. A subject line that reads as one, a real sender identity, a one-click unsubscribe, and suppression the send path checks rather than a spreadsheet somebody processes on Sunday. The useful discipline is ordering: put the receipt or the confirmation at the top where the guest is looking for it, and let the offer sit underneath.
The limit is geography. That's the US federal floor, and it starts from the position that you may mail until somebody says stop. Canada and the EU start from the opposite default, where the consent has to exist before the first send. Since a restaurant's list is built one guest at a time anyway, the safe habit is the same in both places. Keep the marketing opt-in explicit, recorded, and separate from the transaction that produced it.
The four segments that decide everything.
A restaurant's segments are dates, not personas. What separates one guest from another is how recently they came and how often they come, and both numbers are already sitting in the ordering system. I'd build the win-back off those two before anything else on this page, because it's the one segment where the guest already knows the food.
- New: one order or one visit with no second one yet, which makes the second-visit nudge the highest-value message you'll ever send them.
- Repeat: two or three visits, still deciding whether you're a habit or a place they tried once.
- VIP: four or more visits, or above whatever spend threshold you set, and the segment that should never get a discount it didn't need.
- Lapsed: no visit in 30, 60, or 90 days, counted from their own last order rather than from a calendar month.
Most restaurants segment by what they're selling instead, so everybody gets the brunch email and everybody gets the Tuesday special. The segment that changes revenue is the one defined by silence. A guest who came three times in March and hasn't been since is a completely different problem from somebody who has never walked in, and only one of the two is cheap to solve.
A list under a few hundred addresses doesn't have segments, though. It has people. At that size four thin sends lose to one good one, so write the weekly email properly and leave the splitting until a segment is a group rather than a table of six. Lapsed is the exception worth running from month one, because it's the only segment that grows whether you look at it or not.
The four emails that run themselves.
Four emails do most of the work, and here they are in the order a guest meets them. Every one of them fires off something the restaurant's systems already know, which is what separates them from the specials email.
- Welcome: sent within the hour of signup, saying what they'll get from you and how often.
- Post-visit: a day or two later, one question and a reply address a human reads.
- Birthday: sent a week ahead rather than on the day, because the booking gets made in advance.
- Win-back: fired by their own silence at 30, 60, or 90 days, with the reason to return doing the work instead of a discount.
Those four run unattended and the specials email doesn't, which is the sensible division of labor. The operator writes the thing only they can write, once a week, in whatever gap exists between the delivery and service. The lifecycle mail runs off its triggers and nobody opens it again until the menu changes. Most restaurants end up doing neither of those well, because they're trying to run the eight campaigns somebody put in a listicle.
A trigger is only as good as the event behind it, and that's the point automation stalls without one. A restaurant whose visit data never leaves the point-of-sale system has automation in name only, because nothing outside the till knows a guest came in on Friday. The first piece of work isn't writing the win-back email. It's getting the event out of the till and into the list, which is a scheduled export, a webhook, or an API call depending on what your systems offer. Do that once and all four of these become the easy half.
Your point of sale system holds years of order history. We built our platform to put this information to work. I framed the result on SaaS District: the software uses that data to "automatically populate content in emails" so you never have to write a weekly newsletter from scratch.
Sending as yourself, and surviving Mother's Day.
What an inbox provider judges is the domain in your From address, its SPF record, its DKIM signature, and its DMARC policy. A restaurant that lets its booking or ordering platform send under the platform's own domain is building the platform's reputation, and takes none of it with them when they switch.
The requirements are published and they're short. At 5,000 or more messages a day to personal Gmail accounts you need SPF, DKIM, and DMARC, where an enforcement policy of p=none is enough, one-click unsubscribe on marketing mail, and a spam rate in Postmaster Tools held under 0.30%, in Google's sender guidelines. Yahoo asks for the same authentication, the same 0.3% ceiling, and unsubscribes actioned within two days in its own best practices. Most of that is a DNS record and an afternoon.
Here's the restaurant-specific part. You're a burst sender. The list gets built slowly across two years of receipts and then mailed hard in the week before Mother's Day, Valentine's Day, and New Year's Eve, which is exactly the pattern that produces complaints from people who forgot they ever gave you an address. The ceiling is a rate, so a handful of complaints concentrated on the biggest send of the year is worse than the same complaints spread across twelve months. Yahoo's two-day expectation is the tightest number published, so build the unsubscribe handling to the tighter one and stop thinking about it. That's the exact shape I've seen spike a complaint rate fastest: a list gone quiet for months suddenly getting three or four emails a week, which describes a restaurant's holiday run precisely.
Authentication buys delivery, not attention. A list that doesn't remember you fails at much the same rate from a perfect setup, and the repair isn't in DNS. It's mailing something small every month so the name in the From line is familiar by the time the big one arrives.
Who is actually going to run this.
Everything above is a small amount of recurring work attached to data that already exists. Four triggers, four segments, and one email a week.
Every product built for this job assumes somebody sitting at a desk with a browser open, which is a description of an agency rather than of a restaurant. Retrofitting an agent onto a dashboard-first platform is bolting a motor onto a bicycle: the parts fit, and it's still the wrong frame.
Look at what the work actually is. The four triggered emails are a handful of API calls and a rule each. The segments are queries over visit data, once you've got it out of the till and into the list. Nitrosend doesn't read your POS; it starts where the data lands. The weekly specials send is one instruction and a list of dishes. Nitrosend is an AI-native email platform, MCP-first: each of those is a tool an agent can call before it's a screen anybody has to open, contacts are unlimited on every plan including Free, and BYO sending keys on Pro and above keep the mail leaving through your own provider account.
Judgment doesn't automate, and it shouldn't. What's worth promoting this week, which regular deserves a phone call instead of an email, and whether to mail at all in a week when the kitchen is short two people stay the operator's decisions.
If the win-back is the one that pays for the rest, wire that up first: your own guests, their own silence at 30 days, and an email that goes out while you're on the floor. The free tier is 8,000 emails to start, then 500 a month, unlimited contacts, and full MCP, API, and CLI access with no card. Run your whole email stack from one agent command, and spend the evening in the kitchen instead.
Sources
- National Restaurant Association, off-premises dining research: that nearly 75% of restaurant traffic is off-premises, and that off-premises is a larger share of sales than in 2019 for 58% of limited-service and 41% of full-service operators.
- Google, Email sender guidelines: the 5,000-messages-a-day threshold for personal Gmail accounts, the SPF, DKIM, and DMARC requirement, one-click unsubscribe on marketing mail, and the 0.30% spam-rate ceiling.
- Yahoo Sender Hub, best practices: Yahoo's equivalent authentication requirements, its 0.3% spam-rate ceiling, and the two-day window for actioning unsubscribes.
Common questions
From the systems you already run, before you build anything new. The receipt prompt on the till, your own online ordering checkout, and the reservation system between them hold most of the addresses you're ever going to get, and the checkout is the one place you can put a marketing opt-in that means something. Add a table card or a receipt line giving a reason to sign up, and say at that moment what you'll send and how often.
Once or twice a month for the email you write yourself, plus whatever the triggered emails fire on their own. The rhythm matters more than the number. A list that hears from you monthly stays warm, and a list that only hears from you in the week before Mother's Day has forgotten who you are by the time it arrives. Watch unsubscribes per send: a steady rate is health, and a spike is an answer about frequency or content.
Usually yes, and the reason is the From address rather than the feature list. Ordering and booking platforms tend to send under their own domain, so the sending reputation you build there is theirs and none of it moves with you. A separate email platform sends from your own domain, holds guests from all four of your sources in one list rather than one list per system, and lets you segment on visit recency across the lot.
Often not, and never automatically. The order came through somebody else's storefront, so the guest is usually the marketplace's customer rather than yours, and what you can see of their contact details is set by that marketplace's terms. Treat marketplace volume as the reason to move guests onto your own ordering page, where the address and the permission both belong to you.
What they'll get from you, how often, and one thing to do next. Send it within the hour of signup, because it's the only email whose timing the guest chose, and keep it to a few lines: who you are, what turns up in the inbox, and a link to book or order. A small welcome offer is fine, and it's worth less than a clear expectation.
You can, and it changes what the message is. Once the subject line would lead a guest to expect a promotion, or the confirmation isn't what the body opens with, what arrived is a promotion with a receipt on top: it needs a one-click unsubscribe, a real sender identity, and suppression that holds. Put the order details at the top, keep the subject line about the order, and let any offer sit underneath.
Those are two different problems. Landing in spam is usually authentication or complaints: the sending domain has no SPF, DKIM, or DMARC records, or too many recipients on a big seasonal send are marking mail they don't remember signing up for. The promotions tab isn't a penalty, it's a category, and marketing mail belongs in it. Fix the first with DNS records and a cleaner list, and stop worrying about the second.
Watch three things rather than a benchmark: repeat visits from the guests you mailed, revenue per send against the time it took to write, and a spam complaint rate held under 0.3%. Open rate is the number everybody quotes and the least trustworthy one, because privacy proxies fetch the tracking pixel before any human sees the message. Judge the program on the direction those three move across a quarter.