Rent the stack. Own the guest record.

Walk the back office of a busy Chicago group and the picture repeats: a reservation system, a point of sale, an email tool, a review dashboard, and two or three AI add-ons, each sold as the one that finally fixes things. They all work. None of them talk. Ask the owner which guests from last quarter have not come back, and the answer takes a week and a spreadsheet.

The spend is real and the payoff is not landing. In the National Restaurant Association's 2025 read, 83 percent of operators call technology a competitive advantage, yet only 28 percent say it improved profitability. On a pre-tax margin of about 5 percent for a typical restaurant, buying the wrong layer twice is not a rounding error. Ownership, not shopping, is the decision that matters. Some parts of the stack a restaurant should own. Most it should simply rent.

Sort your own stack into rent and own before the next contract renews

Start with what a restaurant should never build

Most of an AI stack is a commodity, and the honest move is to rent it. The reservation platform, the point of sale, card processing, the service that answers the phone during a rush, the model that drafts a reply to a one-star review: many vendors sell each one, they improve every quarter, and switching costs keep falling. Building your own means paying to reinvent what you can lease for a few hundred dollars a month, then maintaining it for good.

Rented AI earns its keep on time saved, not on defensibility. SevenRooms found AI cut the time operators spend reading and answering guest messages by 27 percent. Real gain, good reason to buy. It is also available to the group across the street on the same terms, which is why it is not the thing to build.

What a restaurant should actually own

The build is the connective tissue and the data under it. Two assets are worth owning outright. The first is a single guest record: one profile per person that merges reservation history, check history, allergies and anniversaries, no-shows, and marketing consent, whichever tool captured them. The second is attribution: the thread from a marketing dollar or a direct booking to a guest who actually showed, sat, and spent. No vendor gives you either one clean, because each vendor sees only its own slice. Reservations booked through your own channels are first-party data you own outright, and that is the raw material. Own the record and the attribution, and every rented tool becomes swappable around them. Rent those two and you are renting your own memory of your guests, the one thing a restaurant cannot afford to lose.

The build vs buy line, capability by capability

Two ways this goes wrong

The first trap is the all-in-one platform that promises to be your reservations, your marketing, your loyalty, and your guest database in one login. It is convenient right up to the day you want to leave, when the guest history you thought you owned turns out to live in their schema and walks out with them. Buy the modules if they earn it, but keep the master guest record somewhere you control.

The second trap is building past what you can staff. Owning the record and the attribution does not mean training a model or standing up a data team you cannot keep busy. It means a thin, well-kept layer the rented tools feed. Operators themselves say technology should augment the work rather than replace people, and the build follows the same rule: hold the parts that are yours, rent the rest.

How we would run it

When a group brings us this stack, we do not start by shopping for another tool. We map one operating week: every place a guest touches the business, and every system that writes something down about them. The same way we sequence the workflows a group fixes first, we find the fractures, the spots where one guest exists as four different people, and the point where a booking stops being traceable to a dollar. The first build is almost never a model. It is a guest record the rented tools write into, and one report that answers who came back and what they were worth. That is how we work: map first, then own the narrow thing that matters.

Most restaurants have already bought the AI. Few can point to the asset it built. The tools you rent are replaceable by design, and that is fine. The one you should refuse to rent is your own memory of your guests. The rest is a subscription.