System Intelligence vs. Everyone About How It Works Apply for Access →
The Full Category Map

EVERY UPSTART BUILT
A SMARTER ALARM SYSTEM.
WE BUILT THE ENGINE.

Every category in this market built for one of two tenses. Legacy built for the past — reporting what happened with precision and care. Upstarts built for the predicted future — forecasting, prescribing, modeling aggregate patterns. Neither of them built for the present. Not a forecast of your room. Not a report of your room. Your room. Right now. That tense was empty when the upstarts left it.

The upstarts were describing a real thing. They predicted it well. They modeled it accurately in aggregate. They could not see it. You cannot predict what THIS guest at THIS table is feeling right now. You can see it. We see it. That is the gap the whole category missed.

The upstarts educated the market. They proved operators needed something more than a Monday report. They promised the present tense. They built very good predictions. Prediction is not seeing. We see.

// The specific failure mode

Most of them built consulting-ware — platforms that require their CS team, their QBR, their implementation specialist, and their playbook to generate value. You pay a SaaS license. You receive a retainer. Count the humans between the signal and the action. →

Beneath the consulting layer: many of them are built on third-party model APIs rather than proprietary training corpora. The question to ask any platform: if your model provider changed their API tomorrow, what would change about your product? The full architecture tell. →

Legacy — Tense: past
What happened?

Honest tools. They reported what occurred. Monday morning accounts of Friday. Nobody lied. The ceiling is the past tense.

Upstart AI — Tense: predicted future
What is probably
going to happen?

Forecasts. Predictive models. Prescriptive recommendations. Aggregate patterns that describe what should happen — not what is happening at Table 9 right now. Statistical. Approximate. One table too vague.

superGM.ai — Tense: present
What is
happening?

Not a model of your room. Your room. Camera, WiFi, voice — live, individual, right now. This guest. This moment. With enough resolution to act before it closes. The throne was empty. It is not anymore.

Camera Intelligence WiFi Network Analysis Voice Signal Detection Review Intelligence Crowd Contagion Science Yield Intelligence Empathic Intelligence™ Camera Intelligence WiFi Network Analysis Voice Signal Detection Review Intelligence Crowd Contagion Science Yield Intelligence Empathic Intelligence™
Six Categories. The Same Pattern.

EVERY CATEGORY.
ONE ALARM SYSTEM.
DRESSED AS AN ENGINE.

Each category below shows the same architecture. Legacy built honest tools. Upstarts added AI and called it autonomous. Every one of them ended with a human in the critical path, receiving a signal, and deciding whether to act. The broadcast was live. The result never executed. Six categories. Same gap. One heir.

Category
Operational Intelligence
The category the upstarts promised they would build. They did not.
Full analysis →
Legacy Players
Crunchtime
Operations management, food cost, real-time compliance
Enterprise ops management. Honest about what it is.
Zenput (Crunchtime)
Ops execution, checklists, compliance monitoring
Field ops compliance. Checklists at scale.
Jolt
Digital checklists, food safety, accountability
Operations accountability platform.
They built operations management tools. They did not claim to be AI. They did not claim to act in real time. Their operators know what they have and what it does. That honesty is worth something.
AI Upstarts
SophySays.ai →
Seed
Proactive AI that prevents operational failures before they occur.
Sling by Toast
Acquired
Smart operational awareness built for the modern restaurant.
The pitch was right. Real-time operational intelligence — detecting problems before they cost you — is exactly what operators needed. What deployed was observability: faster alerts to humans who still have to respond. The AI in the marketing was real. The execution layer it implied was not.
They told operators: we will see what you cannot see and tell you in time to act. What they built: we will see what you cannot see and tell you. The act was still yours.
The Heir
superGM.ai

We see what they cannot see. We act. The GM is dispatched when her judgment is specifically required. Everything else executes without her.

The broadcast was live.
The result now executes.
// You did not buy software. You bought a retainer.
SophySays.ai
Real-time alerts require real-time interpretation. The interpretation layer lives with their CS team. The platform watches. The CS team explains what the watching means. You act. You are the third employee in this relationship.
Sling by Toast
Coordination tools require someone to coordinate the coordination. The visibility platform surfaces what the team needs to see. The manager decides what to do about it. Toast s ecosystem makes switching expensive. That is a business model, not a product feature.
Category
Decision Intelligence
The upstarts said their AI agents would act on your behalf. They advise.
Full analysis →
Legacy Players
revenuemanage.com
Revenue management and performance analytics
Analytics-driven revenue management. Reports and recommendations.
Ctuit (Sysco)
Restaurant analytics, reporting, manager intelligence
BI and reporting for ops managers. Honest category.
Avero
Restaurant analytics and performance intelligence
Two decades of restaurant data. Reporting layer.
Analytics platforms. They measure what happened and surface it clearly. Nobody told them they were autonomous. Nobody expected them to intervene. The relationship between operator and tool was always: you read it, you act on it.
AI Upstarts
SignalFlare AI →
Series A
Autonomous AI agents. Proactive. Always working. No human required in the loop.
RAD AI
Seed/Series A
AI-driven decisions that execute before you have to think about them.
OpSage by CONVX
Seed
AI that works like a senior ops executive for every location.
SignalFlare in particular built something genuinely sophisticated. The intelligence layer is real. The signal detection is real. The word "agents" in the pitch implied the agents could act. They produce recommendations. Every recommendation ends with a human. That human is the gap. That is the gap they raised capital to close. They did not close it.
They went to market with autonomous. They deployed advisory. The operators who signed up for agents got alerts. The agents were always on. The actions were still the operator's.
The Heir
superGM.ai

The intelligence SignalFlare built is genuine. The execution layer it implied was missing. We are the execution layer. We are what they were describing in the pitch deck.

The broadcast was live.
The result now executes.
// You did not buy software. You bought a retainer.
SignalFlare AI
The implementation project runs 3-6 months. They call it onboarding. It is a discovery engagement — their team learns your operation so they can configure what the software should have configured itself. The outputs require a Success Manager to contextualize. The QBR is in the contract.
RAD AI
Recommended actions require someone to implement the recommendations. That implementation pathway runs through their customer success team, their playbook, and their quarterly review of whether you followed the playbook.
OpSage by CONVX
CONVX is a consulting firm. OpSage is their platform. The line between product and engagement is not clear because it was not designed to be. The software is the reason for the consulting relationship.
Category
Scheduling and Labor Management
The upstarts put AI on the canvas. The canvas was never the problem.
Full analysis →
Legacy Players
7shifts
Restaurant employee scheduling and team management
Market leader. Clean product. Makes scheduling faster.
HotSchedules / Fourth
Enterprise workforce management and scheduling
Enterprise standard. Deep POS integrations. Honest labor tool.
When I Work
Employee scheduling, time tracking, team messaging
SMB scheduling. Does what it says.
Homebase
Free scheduling, payroll, and team communication
SMB staple. Strong free tier. Labor management basics.
7shifts made scheduling faster for a million restaurant operators. That is real value delivered. Nobody lied. These are scheduling tools. They gave operators a faster canvas. The canvas was not the problem.
AI Upstarts
Nory.ai →
Series A ($14M)
The first AI-native restaurant management platform. Learns your operation and optimizes it continuously.
Harri →
Series C ($50M+)
Predictive AI that builds optimal schedules without manager intervention.
Nory raised $14M on AI-native. Harri raised $50M+ on predictive AI. Both built systems where the AI makes scheduling suggestions that a manager reviews and approves. The AI improved the suggestions. The manager is still the optimizer. The compliance violations still happen when the manager overrides or misses a constraint. The capital raised on autonomous labor optimization deployed an assisted scheduling canvas.
Nory told operators: the AI learns your operation and optimizes it. It does. And then it hands the optimized schedule to a human for approval. That human is the last mile. That last mile is where compliance fails.
The Heir
superGM.ai

A compliant schedule requires encoding compliance as a structural constraint that cannot be overridden. It requires solving against demand forecasts and labor law simultaneously. No human in the optimization loop. Not a suggestion. A constraint. That is what we built.

The broadcast was live.
The result now executes.
// You did not buy software. You bought a retainer.
Nory.ai
The AI-native operating system requires a 3-month onboarding engagement. During those 3 months, their implementation specialists configure the AI parameters. Those specialists are doing what the software was supposed to do autonomously. The AI is native. The setup is billable.
Harri
Predictive scheduling requires someone to approve the predictions. The approval workflow is the product. The human who approves is the ceiling. When that human leaves, they bring in their team to retrain the replacement. That is a consulting model with a SaaS price tag.
Category
Full Management Platforms
The most expensive platform in the category. Still requires her to run it.
Full analysis →
Legacy Players
Restaurant365
All-in-one restaurant management, accounting, and operations
The standard enterprise management platform. Deep accounting and ops.
Apicbase
Restaurant management, recipe costing, and operations intelligence
Menu and operations management for multi-unit chains.
Winnow
AI-powered food waste reduction and kitchen management
Specific, honest use case. Food waste detection and reduction.
Restaurant365 is what enterprise restaurant management looks like. Comprehensive, integrated, expensive, and honest about what it is: a management platform that requires skilled operators to generate its value.
AI Upstarts
Nory.ai →
Series A ($14M)
The operating system for the modern restaurant. AI that runs the operation so operators can focus on guests.
Apicbase (AI layer)
Series B
AI that manages your kitchen operations proactively.
Nory specifically pitched the AI-native operating system — the idea that the platform learns the operation and runs it with decreasing dependence on skilled operators. In deployment, Nory is a very sophisticated management tool that rewards skilled operators. It is better than Restaurant365 in specific ways. It still requires a skilled operator to unlock that value. When she leaves, the platform resets.
They said: the AI will carry the institutional knowledge so the operation does not reset with every GM departure. The AI does learn. The learning lives in the platform settings. The application of that learning still requires an operator who understands the platform. That operator still turns over at 60% annually.
The Heir
superGM.ai

The institutional knowledge problem is not solved by storing it. It is solved by making it irrelevant. An operation that executes independently of who is holding it does not have an institutional knowledge problem. That is what we built.

The broadcast was live.
The result now executes.
// You did not buy software. You bought a retainer.
Nory.ai
The AI-native operating system requires a 3-month onboarding engagement. During those 3 months, their implementation specialists configure the AI parameters. Those specialists are doing what the software was supposed to do autonomously. The AI is native. The setup is billable.
Category
Guest Intelligence and Experience
The upstarts saw the guest data. They built dashboards about it.
Full analysis →
Legacy Players
Medallia (Hospitality)
Guest feedback management and experience analytics
Enterprise feedback. Measures what guests felt after they felt it.
Revinate
Guest data, reputation management, and CRM for hospitality
Hotel-origin. Reputation monitoring and guest comms.
TrustYou
Guest feedback and reputation analytics platform
Review aggregation and sentiment analysis.
Feedback management platforms. They measure guest experience after it happens, aggregate it, and surface it to operators. This is valuable and honest. The measurement arrives after the guest decides. That is the design.
AI Upstarts
Ovation →
Series A
In-the-moment guest recovery. Know when a guest is unhappy while they are still at the table.
Revel Systems (CX layer)
Acquired
Real-time guest intelligence built into your POS workflow.
Tattle
Series A
Predictive guest satisfaction intelligence that drives operational change.
Ovation in particular pointed at the right problem: catching unhappy guests while they are still at the table. Their mechanism was a QR code-based feedback prompt — asking guests for feedback in real time. The guests who are unhappy enough to leave a review are often not the guests who respond to a QR code prompt. The mechanism requires the guest to participate. Hospitality loss detection does not ask the guest anything.
They tried to solve the post-experience problem in real time. They gave operators a faster path to guest feedback. They did not detect the loss of hospitality before it happened. They caught the moment after the guest decided. We operate in the moment before.
The Heir
superGM.ai

We do not ask the guest how they feel. We detect how they feel before they decide. WiFi behavioral signals, camera intelligence, voice detection — the guest has not formed the review yet. That window is ours.

The broadcast was live.
The result now executes.
// You did not buy software. You bought a retainer.
Ovation
Guest feedback recovery requires someone to read the feedback and recover. The real-time platform fires the alert. The operator acts. The gap between alert and action is still a human who is on the floor managing seventeen other things. Their CS team helps you design the recovery workflow. That workflow design is consulting.
Revel Systems (CX layer)
POS-integrated intelligence requires someone to integrate the intelligence into the operational workflow. That integration is a professional services engagement. The PS engagement is where the value lives. The software is the reason for the engagement.
Tattle
Predictive satisfaction intelligence requires someone to act on the predictions. The action path runs through their customer success team, their recommended operating procedures, and their quarterly review of whether the procedures improved the predictions. That is a consulting loop with a data layer.
Category
Conversational AI and Copilots
The copilot still needs a pilot. The pilot is still at capacity.
Full analysis →
Legacy Players
Toast AI features
Integrated AI recommendations within the Toast POS ecosystem
POS-native AI layer. Recommendations within existing workflow.
xtraCHEF by Toast
Invoice processing and food cost automation
AP automation. Specific, honest use case.
POS-native AI features. Useful within the workflow they already occupy. Not claiming to run the operation. The AI is a feature, clearly labeled.
AI Upstarts
AskColette →
Seed
Like having a senior consultant available 24/7. Your operators always know what to do.
Kintow
Seed
Proactive AI that guides operators through service without needing to be asked.
AskColette and Kintow both built products operators find genuinely useful when they have a moment of stillness to engage. The limitation is structural: the copilot is reactive. It responds to invocation. At 8pm on Friday, the moments of stillness required to engage a copilot do not exist. Kintow specifically pitched proactive guidance — the AI guides operators without needing to be asked. In deployment, it responds to voice commands. It waits to be commanded.
Proactive means the system initiates. What deployed was responsive: it answers when asked. Those are different products. One requires bandwidth from the operator. One does not. They built the one that requires bandwidth and called it proactive.
The Heir
superGM.ai

We do not wait to be asked. We detect what is happening, classify what it requires, and dispatch the operator only to the situation that requires her judgment. She is not consulted about everything. She is reached only when her specific capability is needed.

The broadcast was live.
The result now executes.
// You did not buy software. You bought a retainer.
AskColette
The copilot requires a playbook. The playbook tells you how to ask it the right questions to get the right answers. The playbook exists because the right questions are not obvious. A product that requires a guide to use is a consulting engagement with a chat interface.
Kintow
Voice AI guidance requires the right prompts. The right prompts are in the onboarding materials. The onboarding materials are the consultant s job description encoded as documentation. The voice AI is the interface. The consulting engagement is the value.
Before You Sign or Renew

THREE QUESTIONS.
ANY PLATFORM.
ANY CATEGORY.

These questions work for every platform in this assessment. The honest answers will tell you what you are actually buying — and what you are not.

— Question 01
The action question

“Show me one thing your platform did in a live deployment — not alerted, not recommended, did — without a human in the decision loop.”

Why it matters: If the answer involves a human reading something and deciding to act, you have an alert system. If the answer is a specific autonomous execution, you have something closer to what they claimed.

— Question 02
The Friday question

“What does your platform do between 7pm and 9pm on a Friday, during peak service, when your operator is already at capacity and cannot pause to engage with software?”

Why it matters: Most platforms are designed for operators who have time to engage. Peak service is the moment that matters most and the moment most platforms are least useful. The honest answer reveals the ceiling.

— Question 03
The dependency question

“If we cancelled all our QBR sessions, declined your CS check-ins, and used only the self-serve platform — would the value be the same?”

Why it matters: If the answer is no — if your team is required to generate value — then you are buying a retainer, not software. The question tells you where the intelligence actually lives.

For questions specific to each platform — SignalFlare, Nory, Harri, AskColette, Black Box, Ovation, MarginEdge — visit their individual assessment pages. Each company gets five questions built around their specific claims.

Full question bank →
Already Using One of These?

YOUR RENEWAL IS COMING.
READ THIS FIRST.

The gap between what the upstarts claimed and what actually deployed — documented here, by category, with the specific question that exposes it. Before you sign another year with an upstart that delivered advisory and called it autonomous, ask the question on this page.

Before You Renew →
How We Work

THE OPERATORS WHO
BELONG HERE HAVE ALREADY
STOPPED WAITING.

The upstarts spent five years telling operators the room could drive results automatically. Those operators believed them. They deployed alert systems and watched the signals fire while their teams responded — or did not. They know exactly what is missing. They felt it every Friday.

We do not name the operators we work with. We ask that they do not name us to their competitors. The advantage compounds in silence. The platforms on this page will watch their numbers move and not understand why.

// The operators who belong here
Tried an AI upstart. Discovered a smarter alarm.
Still waiting for the signal to produce a result without a human in the way.
Have a competitor they watch who seems to be operating on a different level.
Understand that 60% GM turnover is a system problem, not a hiring problem.
Are ready for the results engine the upstarts described and could not build.
Application Review

MOST OPERATORS
WHO APPLY
WILL NOT BE SELECTED.

We work with operators whose operation, culture, and competitive position fit what we built this for. We review every application individually. We select from the backlog.

If you are reading this because a competitor sent it to you, they may already be in production. We don’t confirm or deny active deployments.

Applications reviewed individually · Not all are accepted