The direct channel has to earn the order
A restaurant can ask guests to order directly. It can print the request on packaging, promote a website and explain the benefit to the business. None of that removes the work waiting on the other side of the link.
Stephen Klein, CEO and co-founder of Saivory, brings that problem to ChatGTM. A customer organizing a group meal may need to calculate quantities, compare packages and decide whether the order will actually feed everyone. A direct channel that makes those decisions difficult gives the guest a reason to look elsewhere.
Klein's conversation with Michael Beck and Chad Horn connects the ordering experience to the longer customer relationship. The opportunity is larger than capturing a transaction without an intermediary. It includes making the purchase useful enough that the customer wants to return, then having a thoughtful way to stay in touch.
Start with the job the customer brought
Klein traces the company's origins to an ordering problem at Fajita Pete's. A group-order customer often knows the occasion, the head count and the budget before knowing the exact combination of products. A catering manager would begin with that context. Many online menus begin with a catalog.
He describes a natural-language approach in which the guest states the need and receives a suggested basket they can review and change. The important part is that the customer remains able to adjust it. An office with good coffee may not need coffee added to breakfast. A useful recommendation makes that choice easy.
This also requires restaurant knowledge. Portions, packages and menu availability cannot be treated as decorative details around the conversation. They determine whether the suggestion works. A fluent response that produces the wrong quantities has simply made the mistake easier to read.
A recommendation needs to make sense at this table
Klein discusses introducing guests to products they may not know, including kolaches for customers unfamiliar with the menu. He also describes recommendations that miss the scale of an order, such as suggesting a single small addition to a much larger group purchase.
That distinction matters when evaluating a system. A recommendation can increase the basket while genuinely helping the guest, or it can look like an attempt to slip more items into checkout. Relevance, quantity and a clear path to remove an item help separate those experiences.
Klein also describes grouping items to take advantage of available package pricing. In that case, the recommendation can help the guest understand value rather than merely spend more. Operators should test those situations with actual menus and pricing rules, including the awkward combinations that rarely appear in a polished demonstration.
The same discipline applies at the location level. Klein notes that stores can differ in their menus and that unavailable items need to be accounted for. The order must remain something the selected restaurant can fulfill.
Keep the simple order simple
Not everyone arrives with a complicated group meal to solve. Klein says a customer who already knows they want a particular doughnut and coffee may prefer a conventional ordering path. Saivory's approach, as he describes it during the recording, leaves room for both.
That is an important design constraint. The business does not benefit from making a familiar task more elaborate just to ensure every guest encounters AI. Assistance should appear where it reduces effort. Returning customers should be able to move quickly when their intent is already clear.
An operator reviewing the experience can therefore test two different journeys: a first-time customer with a group need and a regular who knows the exact purchase. Success on one does not establish success on the other.
The customer relationship begins after checkout
Klein does not advise restaurants to abandon third-party platforms. He acknowledges their role in generating business while describing the value of building a stronger direct mix. The direct experience and the follow-up both need attention.
He points to the customer information collected through first-party ordering as a resource for relevant future contact. That resource needs a plan. Decide what a customer should hear next, why it would be useful and how their communication preferences will be respected. More records in a system are not, by themselves, a stronger relationship.
His account of working with Shipley Do-Nuts also emphasizes listening to operators as the product develops. Priorities emerge from the actual business, and a small team has to choose what it can build well. That keeps the technology connected to its purpose: helping the restaurant offer a direct experience that deserves the next order.
