← Home
FohBoh | Cortex™  ·  AI for Operations

Operations Intelligence
You Can Actually Trust.

Most restaurant AI runs on raw, unverified data. Cortex runs on the KPI Vault — a sealed, immutable record of certified operational truth. Every recommendation traces back to a fact, not a guess.

Certified data layer KPI Vault source of truth No hallucinations on certified KPIs Every insight is auditable $249/loc/mo · 10 loc min

The Core Principle

The data your AI runs on
determines everything.

Garbage in, garbage out — but at scale, with confidence scores attached. Cortex is purpose-built to run on MGE-certified data. When it tells you food cost is up 3.2% at Dallas, that number traces to a sealed vault entry — not a raw POS export that has never been reconciled.

Certified data layer KPI Vault source of truth No hallucinations on certified KPIs Every insight is auditable

Where It Works

In-Store and Above-Store Intelligence

Cortex operates at two levels simultaneously — helping GMs run better shifts and giving COOs a portfolio view no spreadsheet can match.

🏪  In-Store Operations
📉
Food Cost Variance Detection
Why did food cost spike Tuesday night? Cortex traces variance to certified recipe cost data, purchase records, and waste logs — and tells you which category drove it.
👥
Labor Efficiency Analysis
Compare labor as a percentage of certified revenue — not just hours scheduled. Identify overstaffed dayparts and correlate against cover counts and ticket times.
🍽️
Menu Item Performance
Which items are dragging your margin? Cortex cross-references certified sales mix data with recipe costing to surface what to push, price, or 86.
Real-Time Operational Alerts
Surface exceptions the moment data is certified — waste over threshold, ticket time outliers, delivery ratings dropping. Flags go to the right person automatically.
🏢  Above-Store Operations
📊
Portfolio Performance Comparison
Rank every location by certified margin, food cost %, and labor efficiency on the same normalized baseline — no location gaming the data.
🔍
Location Benchmarking
Why does Location 3 run 4 points higher food cost than Location 7? Cortex isolates variables across certified KPIs — menu mix, daypart, staffing pattern.
📈
Trend Identification at Scale
Identify system-wide patterns — rising protein cost across all 12 locations, declining lunch covers in suburban units — before they show up in P&Ls.
🎯
Executive Reporting
Board-ready certified performance summaries generated in seconds from Vault data. No analyst hours. No data preparation. Just certified numbers, surfaced on demand.

Real Questions. Certified Answers.

Who Uses It and What They Ask

Cortex is designed for operators, not data scientists. Every question is answered from certified Vault data — traceable, repeatable, and honest about what it doesn't know.

GM
"Why is my food cost running 4 points over budget this week?"
Cortex retrieves certified food cost data from the KPI Vault, isolates the variance to protein and dairy categories on Tuesday and Wednesday, and cross-references against certified purchase records. It surfaces the specific vendors and items driving the gap — with certified dollar amounts, not estimates.
VAULT: D1-FC
COO
"Which of my 14 locations are underperforming versus their peer cohort?"
Cortex pulls certified margin and efficiency data from all 14 Vault entries, normalizes by cohort (volume tier, market, format), and returns a ranked view. Locations 4, 7, and 11 are running 2.8–4.1 points below cohort median on certified EBITDA margin. It then surfaces the top driver for each.
VAULT: D2-MG
Finance
"What's driving the margin compression at the Dallas cluster this quarter?"
Cortex traces through the Vault's certified quarterly KPIs — isolating labor cost increase (+1.9pts), delivery commission overcharges (+$14K uncertified), and menu mix shift toward lower-margin items. Each finding references its Vault entry ID. No estimates. No AI confabulation.
VAULT: D3-RC
Ops Dir.
"Which menu items should I consider repricing or removing heading into Q3?"
Cortex retrieves certified sales mix data and certified recipe cost records, computes contribution margin per item against the Vault baseline, and returns a ranked list. Three items are below 55% contribution margin and declining in mix share. Repricing thresholds are based on your certified cost data, not industry averages.
VAULT: D1-MN

Prompt Architecture

Prompts Are Categorized, Not Free-Form.

Every prompt is routed to a structured category that determines which Vault data it retrieves, which rules it applies, and how it validates the answer before responding.

Category 01 — Diagnostic
Why did X happen?
"Why is food cost high this week?"
Root cause analysis against certified Vault data. Retrieves the relevant KPI history, isolates the delta, and attributes the variance to specific certified drivers.
Category 02 — Comparative
How do I compare?
"How does Location 3 compare to Location 7?"
Cross-location or cross-period benchmarking on certified KPI data. Normalizes by cohort before comparison — volume, format, and market are controlled.
Category 03 — Trend
What's changing?
"Which costs are trending up across my portfolio?"
Pattern detection across time-series certified Vault entries. Surfaces directional changes before they appear in monthly financials.
Category 04 — Prescriptive
What should I do?
"Which items should I reprice in Q3?"
Recommendations derived from certified margin data, not industry benchmarks. Every suggestion anchored in your Vault — your costs, your mix, your contracts.
Category 05 — Predictive
What's likely ahead?
"Will I hit my labor target this period?"
Forward projections based on certified historical Vault data and current-period inputs. Cortex is explicit about confidence levels — it won't assert certainty on uncertain data.
Category 06 — Audit
Can you show your work?
"Where does that food cost figure come from?"
Every answer traces back to its Vault source entry. Returns the certified data point, its seal date, and its chain of custody — full transparency on demand.

Patent-Pending Infrastructure

The KPI Registry and KPI Vault

The infrastructure that makes Cortex trustworthy isn't the AI model — it's what the model draws from. The Registry and Vault are the immutable source of truth for system-wide continuity.

KPI Registry™ — Patent Pending

The Canonical Definition of Every KPI

The Registry is the single authoritative source for what every KPI means across the entire system. Food cost % means the same thing in every prompt, every report, every AI answer, every certification — defined once, never redefined locally.

Eliminates definitional drift — "food cost" at Location 3 cannot mean something different than at Location 11

Governs every AI prompt — when Cortex retrieves a KPI, it uses the Registry definition, not the AI's internal weights

System-wide continuity — add a new location or new data source and the Registry ensures consistency from day one

Version controlled — if a definition changes, every historical Vault entry that used the prior version is preserved under the original definition

KPI Vault™ — Immutable Record

The Sealed Historical Record of Every Certified Value

The Vault is where certified KPI values live permanently. Every value written to the Vault is SHA-256 sealed and timestamped. No one can alter a sealed entry. It is the ground truth the AI reasons from.

Immutable by design — sealed entries cannot be edited or deleted. A correction creates a new version, preserving the original

AI retrieval anchor — Cortex's CRAG loop retrieves from the Vault first. If the answer isn't there, Cortex says so rather than inferring

Cross-module continuity — the same Vault entry Sentry uses to certify a fee overcharge is the same entry Cortex uses to answer an operations question

Audit-ready at all times — every recommendation traces to a Vault entry ID, its seal date, and the Registry definition it was interpreted under


How Cortex Answers

CRAG — Corrective Retrieval-Augmented Generation

Cortex doesn't generate answers from model weights alone. It retrieves from the Vault first, validates against the Registry, then generates — and corrects itself if the answer doesn't ground out in certified data.

STEP 01
💬
Prompt Received
The question is parsed and routed to the appropriate prompt category. Intent is classified before any data is retrieved.
STEP 02
🗄️
Vault Retrieval
Cortex queries the KPI Vault for the certified data relevant to the question. The Registry definition governs which entries qualify.
STEP 03
Validation Loop
Retrieved data is validated against the Semantic Truth layer. If Vault coverage is insufficient, Cortex flags uncertainty rather than inferring past the data.
STEP 04
💡
Grounded Response
The answer is generated with Vault entry IDs cited. Every number in the response is traceable. If Cortex can't ground it, it says so.

System-Wide Continuity

One version of truth.
Across every location,
every module, every prompt.

The KPI Registry and Vault aren't just database infrastructure — they're what makes AI trustworthy at scale. When the same definition governs every answer across 50 locations, you stop arguing about the numbers and start acting on them.

🔗

Sentry and Cortex share the same Vault. Fee recovery certifications and operations insights draw from the same immutable source — no reconciliation needed.

📐

Registry definitions travel with every new location. Onboard location 51 and it immediately operates on the same KPI definitions as locations 1 through 50.

🔒

Sealed Vault entries are permanent. Auditors, counsel, and board members can independently verify any AI-sourced figure at any time — without FohBoh's participation.

💬

AI that says "I don't know" is more useful than AI that guesses. Cortex surfaces its own uncertainty — flagging when Vault coverage is insufficient rather than inferring past the data.