Mencora
AI-powered Business Automation System
An automation system combining workflows, structured data and AI-assisted analysis to support business processes.
Mencora is an experimental system exploring how structured workflows and AI-assisted analysis can support repeatable business processes.
Problem
Business processes frequently combine repetitive coordination, fragmented information and decisions that still need human context.
Constraints
- Human approval gates every Gmail Draft; the system never auto-sends.
- Google Sheets quota remains a structural constraint.
- Scoring is a heuristic, not a statistical or ML model.
Solution
An automation-oriented system that brings workflows, structured data, integrations and assisted analysis into a single process.
The technical layers that turn the product idea into a working system.
Workflow orchestration
Eight n8n workflows coordinate launch, prospecting, analysis, review, drafts and reads.
Structured analysis
Responses API returns strict JSON with low reasoning and no web tools.
Operational data
Mencora CRM uses Google Sheets for campaigns, prospects, analyses and drafts.
Human review
Explicit states gate Gmail draft creation and persist its ID.
Select a node to inspect its role and relationships.
The verified draft-only path, orchestrated by n8n and gated by a human.
Gate communication with human approval
- Context
- AI-assisted copy is not authorized communication.
- Decision
- Require APROBADO before WF07 creates a Gmail Draft.
- Trade-off
- The workflow includes a manual queue and never sends.
- Revisit when
- Only if the safety policy changes deliberately.
MEN-INC-01Only one batch result persistedResolved
- Symptom
- Multiple model outputs produced one stored row.
- Root cause
- n8n Run Once / Execute Once semantics.
- Fix / mitigation
- Reworked item execution for batches.
- Verification
- Validated with 3 drafts, then 5 drafts.
- Lesson
- Workflow execution semantics are data correctness.
- Status
- Resolved
MEN-INC-02Sheets quota failuresMitigated
- Symptom
- Workflows and frontend refetches created read pressure.
- Root cause
- Google Sheets request quota was limiting.
- Fix / mitigation
- Retries, waits, cooldowns and reduced refetch.
- Lesson
- Mitigation does not remove a structural limit.
- Status
- Mitigated
MEN-INC-03Duplicate campana_idMitigated
- Symptom
- A physical duplicate remained in Sheets.
- Root cause
- Persistence lacks uniqueness enforcement.
- Fix / mitigation
- WF08 deduplicates on read and keeps greatest row_number.
- Verification
- 29 rows became 28 unique campaigns.
- Lesson
- Read repair does not fix stored integrity.
- Status
- Mitigated
MEN-INC-04Compass errors after tab switchingResolved
- Symptom
- Returning to the tab triggered more reads under quota pressure.
- Root cause
- Automatic focus/reconnect refetch amplified load.
- Fix / mitigation
- Disabled focus and reconnect refetch.
- Lesson
- Frontend cache policy affects datastore load.
- Status
- Resolved
MEN-INC-05Duplicate AI execution observedPartially validated
- Symptom
- Analysis historically ran more than once.
- Root cause
- Exact root cause was not fully confirmed.
- Fix / mitigation
- Applied Execute Once and cooldown mitigations.
- Lesson
- Uncertain causes must remain labelled uncertain.
- Status
- Partially validated
- n8n batch semantics require deliberate item-level design.
- Structured output makes automation inspectable.
- Human review and draft-only delivery are intentional boundaries.
- Compare Gemini / Perplexity only as a future experiment.
- Expand commercial validation without implying customers or revenue.
00–08 set including launcher, analysis, review and list reads.
A documented campaign completed the draft-only path.
Five eligible rows persisted in the documented batch.
WF08 retained the greatest row_number and sorted newest-first.
Mitigated with retries, waits and refetch policy; the structural limit remains.