Restaurant CRM: turning guest data into repeat visits

What a restaurant CRM actually is

CRM stands for customer relationship management. In hospitality that is a big phrase for something small: one place that holds who your guests are, what they order and when they were last in.

The difference between that and an address list is the connections. A list of email addresses tells you somebody once bought something. A guest profile tells you the same person has ordered eight times, nearly always on a Thursday, nearly always the same dish, and that it has now been quiet for six weeks.

There is nothing you can do with the first one. There is plenty you can do with the second.

You spot some of those patterns without software. You recognise faces, you know which table always takes the ribs. What you do not see is the guest who stops coming, because by definition they are not standing in front of you. That is exactly where a system starts earning its keep.

What it is not

• It is not your till. A till records transactions, not people.
• It is not your email tool. That sends things neatly, but has no idea who was last in yesterday.
• It is not your reservation book, although plenty of reservation systems have a guest database hidden inside them.
• It is not a project with an end date. It is a habit with a database underneath it.

That last point decides whether any of this works. Treat it as a project and you buy software and stop after three months. Treat it as a habit and you look at the same overview every Monday and act on one thing in it.

The data that actually matters

You can record an endless amount about a guest. In practice you steer on a handful of fields and the rest is decoration.

Of that list, visit frequency is the least used and the most valuable. Revenue tells you what has happened. Frequency tells you what is about to happen.

Two numbers do nearly all the work: how long ago somebody last bought from you, and how often they have bought in total. Recent and frequent buyers are your base. People who used to buy often and have gone quiet are the most expensive group to lose and the cheapest to win back.

Preferences and allergies belong in the same profile even if you never use them for marketing. A guest who has told you once that they cannot have nuts does not expect to be asked again next time. That is not a data project, it is hospitality written down.

Where that data already sits

Almost no restaurant starts from zero. The problem is rarely that the data is missing, it is that it is spread across five places that know nothing about each other.

• Your till: tickets, timestamps, what went through each table.
• Your own ordering site: name, email address, delivery address and the full order history.
• Your reservation system: names, party size, no-shows, notes on the table.
• Your loyalty programme: who earns points and who actually redeems them.
• Your Google Business Profile and review sites: what guests say about you in public.
• And the rest: the notebook by the phone, the group chat, the head of whoever runs your floor.

Merging it is the real work, and the sticking point is almost always the key. On your ordering site a guest is an email address; in your reservations they are a name and a phone number. Pick one key, usually the email address or the mobile number, and make sure every system you connect passes it along. Without that choice you do not get a guest profile, you get three half ones.

What the delivery platforms do not hand over

Orders through Thuisbezorgd, Uber Eats or Deliveroo bring in revenue. What they rarely bring in is a guest you can approach yourself later. The platform sits between you and the person ordering: the point of contact, the next offer and the app on the phone all belong to them.

That is not a complaint about the platforms, it is their model. They sell reach and they hold on to it. But it does mean growth through a platform does not grow your own database. You rent every order again, and the commission is the rent.

An order that comes in directly costs you no commission and does leave you a profile. That is why your own ordering site and your CRM are in practice the same subject. Without a channel of your own, your guest database stays empty no matter how busy you are.

What to judge a system on

The question is hardly ever which brand is best. The question is which type of system fits where your revenue comes from, and whether your team will actually use it. These five points rule out most of the options on their own.

• Ease of use. If explaining it to a new member of staff needs a manual open on the counter, it will not happen. Ask for a trial account and let somebody on your team run a shift with it.
• Pricing that grows with you. Check what happens when your database doubles, and whether you are locked into an annual contract.
• Complete guest profiles. Order history, visit frequency, contact details and loyalty balance belong on one screen, not in four reports.
• Integrations that exist today. Your till, ordering site and reservations should feed in without you exporting files by hand.
• Automation. A system that only stores things gives you homework. A system that sends things takes work off you.

When you price it up, do not count the subscription alone. Add the hours setting it up will cost and any equipment you have to buy alongside it. Some systems only run on a tablet you do not own yet, and that figure rarely appears on the quote.

Prices and packages change too often to print in an article. Ask the supplier directly, and in the same breath ask what is not included.

The types of system, and what each one misses

League tables of CRM packages date faster than your menu. What does not date is the categories. Six kinds of system call themselves a CRM, and they solve different problems.

The names you meet on the floor are reservation platforms such as OpenTable, Resy and SevenRooms. On the till side it is often Lightspeed or whatever your own till supplier offers. A general CRM such as HubSpot will do anything you configure, which is an advantage if you have somebody willing to configure it and a drawback if you do not.

Choose on the basis of how your guests arrive. If your place runs on reservations and tables, your database is in your reservation system and that is your starting point. If it runs on collection and delivery, it is in your ordering site, and any system that only understands tables is half a system for you. If you do both, the question is which of the two you make the leading one.

Combining is fine. Storing twice is not. Agree which system holds the truth and have the others write into it, or in a year you will be maintaining two databases that contradict each other.

What you do with the data

Collecting is the easy part. The moment a CRM starts paying for itself is the moment you stop sending one message to everybody.

A newsletter to your whole database treats yesterday’s guest and last year’s guest exactly the same. Those two have nothing in common beyond having bought something from you once.

There are no good Dutch figures on how many guests choose a restaurant on the strength of a loyalty programme. In American industry research from the National Restaurant Association in 2025, 41% of guests said a loyalty or rewards programme weighs on where they eat. That is American data, so treat it as a direction of travel rather than a promise about your own street.

Four groups that genuinely exist

More groups are possible, but start here. Four segments you refresh every month do more than twelve you create once and never touch again.

With the drifting group, watch the difference between quiet and gone. Six weeks of silence from a guest who came weekly is a signal. Six weeks of silence from somebody who visits twice a year is normal. The guest’s own rhythm decides when you miss them, not the calendar.

• Invite the guests who took the tasting menu most often to the evening you run the new one.
• Put a suggestion in your order confirmation that matches what is in the basket, not what you want to shift.
• Only ask for a review after somebody has ordered a second time. After the first, they do not know whether they mean it yet.
• Send your birthday message a week early, so there is still a table left to book.
• Ring the ten guests who spent the most with you and whom you have not seen for six months. That is an hour of work and it is not a campaign.

None of those five starts with a discount. That is deliberate. Discounting works in the short term and teaches your regulars to wait until something is on offer again.

The automations you set up once

The argument for automation in hospitality is not that it looks clever. It is that you are not going to sit typing messages at seven in the evening, and that without automation it therefore never happens at all.

This sequence covers most of it and you can set it up in an afternoon.

1. Welcome. After the first order: thank you, here is what else we do, and a reason to come back within two weeks.
2. Reorder. After however many days suit your kind of place, a message featuring the dish the guest had before.
3. Birthday. Only if you asked for the date properly, and well before the day itself.
4. Win back. Six to eight weeks of silence from a guest who used to come more often, with a reason to return rather than an immediate discount code.
5. Review. After the second order, with a direct link to your Google Business Profile.
6. Balance. Anybody close to enough points for something free should know about it.

Put a cap on frequency. One marketing message per guest per week is already a lot for most places, and somebody who has just bought from you does not need an offer on exactly that.

Review every quarter which sequences are still running and what they produce. An automation nobody reads any more will otherwise sit there for years, because nobody is looking at it.

What the GDPR asks of you

Guest data is personal data, so the GDPR applies, which Dutch law and Dutch suppliers call the AVG. That is not a reason to do nothing, it is a reason to set it up properly once rather than find out later.

• Ask for consent where you need it, and record when and for what somebody said yes.
• Do not collect more than you use. A date of birth with no birthday message is a risk with no return.
• Put an unsubscribe link in every commercial message, and make sure unsubscribing genuinely works.
• Agree a retention period. A guest who has ordered nothing for four years does not belong in your database.
• Sign a data processing agreement with every supplier processing the data for you, and ask where it is stored.
• Make sure you can show a guest what you hold about them, and delete it when they ask.

There is more room with existing customers than people assume. Subject to conditions you may email them about things similar to what they already bought from you, as long as they can unsubscribe from every message. The detail of the rules does shift, so have your setup checked once by the Autoriteit Persoonsgegevens or a lawyer before you import thousands of addresses.

And do not buy address lists. A database you did not build yourself performs badly and earns you precisely the complaints you do not want. The guests you collect yourself have a reason to read you: they have eaten your food.

Measuring it, and what to do on Monday

A CRM does not pay for itself in messages sent. It pays for itself on four numbers.

• Share of repeat guests: how many of your orders come from somebody who has bought before.
• Time between orders: if it is getting shorter, it is working.
• Share of direct orders: how much of your online revenue bypasses the platforms.
• Revenue per guest per year: the number that sums up the other three.

Measure them monthly and put them side by side. A campaign that feels good but moves none of these four is a campaign you can stop without losing anything.

If you want to start on Monday without buying anything: export your orders from the last ninety days, sort by email address and count how many people ordered more than once. That percentage is your baseline. Then take the twenty names who spent the most and whom you have not seen for six weeks, and send them a message you write yourself.

If that works, you know exactly what you are buying a system for and what it has to do. If it does not, software would not have saved you either.