Workflow Scanner
Reconstruct real workflows from system logs. Find bottlenecks, judgment points, and automation opportunities before you automate.
Value and Reality,
Scanned Together
Values alone don't move operations. Operations alone can't optimize.
Dual-axis scanning maps judgment and execution simultaneously.
Two scans, one model
This Is What You Get
Two maps, side by side. One shows what you say. The other shows what you do.
Avg 42h legal bottleneck
Cost cited 2.3x more in approvals
"What the organization says it values"
"What the organization actually does"
Where these two maps diverge, MARIA OS detects structural risk.
One Scan, One Report
Values, reality, gaps, bottlenecks, and gate-validated recommendations — all in a single deliverable.
Stated vs Practiced alignment
Bottleneck wait time (hours)
Where values and reality diverge
Gate-validated actions
Interview Last: Facts Before Opinions
Read traces first. Ask humans last. Build a Reality Map from evidence.
Connect
- Read-only access
- Least privilege
- Audit log
- PII masking
“Connected to Slack, Jira, Google Drive with read-only OAuth”
Artifact Scan
- Parse documents & files
- Extract ownership chains
- Tag decision artifacts
“Found 847 documents, 12 ownership chains, 34 decision points”
Event Log Scan
- Timeline reconstruction
- Latency measurements
- Handoff detection
“Detected 4 bottlenecks: Finance review avg 14h, VP approval avg 22h”
Implicit Decision Detection
- Unwritten approval gates
- Shadow escalation paths
- Informal authority maps
“3 hidden gates: CFO informal veto, Team lead chat approval, Friday freeze”
Replay Interview
- Walk-through with owners
- Confirm or correct findings
- Fill evidence gaps
“Owner confirmed: 'Yes, nothing ships without Sarah's Slack thumbs-up'”
Chaos In, Structure Out
Real-world data is unstructured by default. MARIA OS converts it into auditable, evidence-linked structure.
Judgment Controlled, Execution Parallelized
Split the same workflow into two graphs: one for decisions, one for tasks.
Graph Builder Output
Purchase Order Approval Flow
This is what your PO approval actually looks like. MARIA found 2 bottlenecks and 1 hidden gate between Risk Assessment and Budget Check.
Estimated cycle-time reduction: 42% after gate restructuring.
Optimize Existing, Create New
Existing flows get faster. New flows start safe. All changes are versioned and auditable.
Invoice Approval Workflow — as observed
Auto-extract invoice data
Route by amount
Parallel: Finance + Manager
Evidence auto-bundle
Invoice Approval Workflow — recomposed
From Memory to Graph
MARIA OS doesn't replace workflows. It understands, structures, and evolves them.
Current State
7 steps, 3 hidden gates, no audit trail
With MARIA OS
6 steps, full governance, evidence-backed
Same process. Structured governance. 93% faster.
AI Proposes, Gates Decide
Every optimization candidate must pass 5 gates. No exceptions.
Proposed Optimization
Auto-approve invoices under $5,000
Risk Tier
Responsibility
Evidence
Policy
Execution Safety
Gate Stack Result
Awaiting evaluation...
Every optimization MARIA proposes must pass through 5 safety gates.
As Risk Rises, Humans Take Over
The ratio is not a setting. It is determined by Risk, Evidence, and Responsibility Gates.
Agent collects, organizes, detects gaps
“Gather all PO history and vendor data”
Agent evaluates within limits
“Auto-approve standard office supplies”
Agent proposes, human confirms
“New vendor onboarding — $10K contract”
Agent organizes evidence, human decides
“Terminate supplier relationship”
Agent stops and explains, human bears responsibility
“Major strategic acquisition”
As judgment weight increases, human share rises naturally.
The ratio is not a setting. It is determined by Risk, Evidence, and Responsibility Gates.
As Judgment Progresses,
Human Ratio Emerges
Decisions don't complete in an instant.
As evidence builds, human ratio naturally rises.
Responsibility Is Not Implicit.
It Is Confirmed with a Click.
HR = 0.30R + 0.25(1-E) + 0.25H + 0.10(1-V) + 0.10P
Human Ratio = 0.42
Contract Amount Change Approval
Risk Tier: Medium · Scope: 1 customer · Reversible
Responsibility is not yet assigned.
AI assists the judgment.
Humans accept the responsibility.
That moment is visible in the UI.
Even When AI Runs,
Responsibility Stays with the Human
Immutable · Non-deletable · Valid until execution ends
Task C paused — Unexpected input detected
Decision owner attention required. Responsibility anchor referenced.
AI runs the execution.
But responsibility never leaves the anchor.
Responsibility Has a Lifecycle.
It Shrinks, but Never Disappears Silently.
Approval
Execution
Completion
Validation
Archive
Responsibility closes on evidence, not on time.
Even after archive, responsibility can re-open for:
A new Responsibility Anchor is created. Past anchors are never reused.
Accept responsibility at decision.
Hold it through execution.
Close it on evidence.
MARIA OS designs the full lifecycle.
Contact
Workflow Scanner Inquiry
Share your goal, deadline, constraints, and target systems. We will return scope and execution steps.