Every platform in this market has a demo. The demo is excellent. The data is clean. The interface is polished. The narrative your sales rep constructs is coherent and compelling and entirely accurate about what the platform can do under optimal conditions.
The demo does not show you what happens at 7:47pm on a Friday when the room starts to turn and no one has time to open the app.
These questions will.
Question 1: Show me an instance where your platform acted without a human in the loop. Not alerted. Acted.
This is the most important question you will ask. It separates every platform in this market into two categories: platforms that act, and platforms that alert a human who acts.
The correct answer is a specific example — not a roadmap feature, not a capability statement, not a use case description. A specific thing the platform did autonomously. An order placed. A dispatch executed. A pricing adjustment made. Something that happened in a restaurant without a human deciding to make it happen.
If the answer involves a human reading something and then doing something, you are looking at a Layer 2 platform. If the answer involves a roadmap or a future release, you are looking at a platform that has not yet built what it is selling you.
Question 2: What is your behavioral training corpus? How many consumer decisions is your model trained on? From what environments?
Every platform in this market will describe their AI capabilities. Very few of them will describe what their AI was trained on.
Restaurant data — transaction records, labor logs, review scores — is the raw material most platforms used. It is what they had access to. The limitation is not that restaurant data is insufficient. The limitation is that crowd contagion, price sensitivity modulation, and disengagement how energy moves through a room are not visible in restaurant data at the scale required to model them.
The behaviors that matter most for real-time hospitality protection were learned in environments that process 40,000 to up to 80,000 people at once. Those environments are stadiums, theme parks, and mass retail. No restaurant platform trained their models in those environments. Ask them to confirm.
Question 3: What is the latency between a signal occurring in my restaurant and that signal producing an action?
If the honest answer to this question is anything other than “seconds,” the platform is not built for real-time intervention.
Platforms built on Snowflake, BigQuery, or any batch-processing architecture will describe their data pipeline in terms of near-real-time refresh rates — 15 minutes, 5 minutes, sometimes 60 seconds under ideal conditions. None of these are adequate for a hospitality loss window that opens for 90 seconds to 6 minutes and closes permanently.
Ask them to walk you through the data path from floor event to action. Count the steps. Each step that involves a scheduled job, a batch process, or a human is a step that adds latency your guests will feel before the platform does.
Question 4: What happens when my General Manager turns over?
The restaurant industry has 60% annual GM turnover. This is not a personnel problem. It is the operational environment in which every restaurant technology platform must function.
Platforms that require trained operators — copilots that need a skilled pilot, management systems that need someone who knows the interface, analytics tools that need someone who knows what questions to ask — reset every time a GM leaves. The institutional knowledge that made the platform useful leaves with her.
Ask your vendor: if my GM leaves next month, what is the operational cost to the platform? If the answer involves a ramp-up period, a retraining process, or a knowledge transfer exercise, the platform is not designed for the industry it is serving.
Question 5: Can your platform detect the loss of hospitality before a guest decides how they feel?
This question will tell you whether the platform was built around the problem or around a set of capabilities that are adjacent to the problem.
The problem is not slow service, or low check averages, or declining review scores. Those are outcomes. The problem is the moment — specific, measurable, detectable — when a guest stops feeling like a guest. Before that moment hardens into a decision. Before the decision hardens into a behavior. Before the behavior becomes a record.
Ask them to describe how their platform detects that moment. If the answer describes post-experience analytics, review score monitoring, or satisfaction surveys — the platform was built to measure the outcome, not prevent it. If the answer involves real-time behavioral signal fusion from WiFi, camera, and voice detection — ask them to show you a deployment example, not a capability statement.
Question 6: What did your last three enterprise customers say changed about their Friday nights?
Not their metrics. Not their NPS scores. Not their case studies. What did their operators actually say changed about the experience of running a service after deployment?
The answer to this question will tell you what the platform actually does. Platforms that improve understanding of the operation will produce answers about better data, clearer reporting, and more informed Monday meetings. Platforms that improve the experience of running the operation will produce answers about what it feels like to be on the floor.
If the answer is about dashboards and reporting and data clarity — you know what layer the platform operates at. If the answer is about Friday nights feeling different — ask for a reference call.
Before You Renew
If you are approaching the renewal window for a platform you have been using for 12 months or more, there is one question worth asking yourself before you ask any vendor:
What has changed about the way I run my Friday nights since I deployed this?
Not what has changed about your reporting. Not what has changed about your Monday meetings. What has changed about the experience of running a live service with a full room?
If the honest answer is “not much,” the platform you are about to renew was built for a version of your problem that is adjacent to — but not the same as — the problem you actually have.
The switching cost is real. The retraining is real. The disruption is real.
The cost of staying is also real. It happens every Friday. It does not appear on any invoice.