Weeks of scattered context. Nobody has the full picture.
Operator notes. Interviews. Slack threads. Diagrams. Process docs. Five formats. Six teams. Zero reconciliation.
operator-notes.md
Riya Singh analytics-eng
interview.md
Tom Whitfield head-of-data
process-doc.md
v2.1 (stale) updated 2024-09
arch.svg
+ companion vision-extractable
slack.json
#data-customer-health 14 days · DMs
Inputs
Drop what you already have.
→Operator notes
→Interview transcript
→Process documentation
→Architecture diagrams
→Slack / Email threads
→
L01 AURORA
Vision-extracted diagrams. Parsed structured threads. Cited claims with line-number precision. No interviews to schedule.
How It Works
Eight specialists. One signed dossier.
Ingest
Heterogeneous inputs
Extract
Claims + citations
Triangulate
Cross-source proof
Gap Analyst
What's missing
Rank
Value model
Boundary
What's out of scope
Assemble
Blueprint render
Pre-Signoff Audit
Citation integrity
Current-State Map · 13 systems · 4 layers
Workflow and architecture — mapped together.
13 systems · 4 layers · sources / models / delivery / supporting. Documented paths in cyan. The shadow channel CS actually trusts in violet. Broken connection in red. This system surface is what L02 acceptance criteria reference and L03 integration proofs cite.
Triangulation
Two sources. One uncomfortable truth.
Slack · 2026-05-26 · Riya Singh
"ACCT-447 has been firing false-positive for months. The customer_health_v2 model is reading nulls for 3 of 5 inputs."
⤳
Interview · Tom Whitfield · L32
"If that's true, the dashboard has been wrong for six months."
⚠ GAP G2 · Broken pipeline since Nov 2025
AI Opportunity Register
Every opportunity rated G / Y / R for AI suitability.
WS-005
Repair scoring model
Rank 6.075
WS-001
Single canonical definition
Rank 5.86
WS-002
AI driver attribution
Rank 5.21
WS-003
AI playbook dispatch
Rank 5.39
WS-004
Board memo synthesis
Rank 5.78
WS-007
CS Ops query surface
Rank 5.89
WS-006
Living process docs
Rank 5.29
(rejected)
Fully-autonomous churn calls
Refused at L01
Green = ready · Yellow = needs human-in-the-loop · Red = refused before specification
Value Model · WS-005
Value, bounded by confidence intervals.
Cost Avoided
Rework + escalation cycles
$25K – $100K
per year
Productivity
CS Ops manual suppression
100 – 500 hrs
per year
Revenue Protected
ARR at risk addressable
$500K – $2.0M
of $3.8M cohort exposure
Every figure cites the source quote that supports the interval — operator estimates, financial baseline, observed cohort exposure. This is what Layer 04 will measure against.
Sign-Off
A future-state blueprint — signed before anything gets built.
L01 Output · Future-State Blueprint
Keystone Customer Health · v1
Sarah Okafor
VP Customer Success
Accuracy
✓
Maya Chen
CEO & Co-founder
Prioritization
✓
Both signatures required to advance to L02
LOCKED
Automated · Repeatable
Weeks of consolidation now automated.
Weeks
→
Hours
7
AI opportunities ranked
26
Evidence-anchored claims
$3.8M
ARR exposure surfaced
100%
Citations resolved
Handoff
One signed artifact. Forward.
L01 · Signed
Future-State Blueprint
→
L02 · SPECIFY
Aurora doesn't hand off a slide deck. It hands off a binding artifact the next layer cannot deviate from.
Most discovery turns into a six-week interview marathon. Aurora turns it into thirteen days.