FR();
03 / Engineering case study

Mencora

AI-powered Business Automation System

An automation system combining workflows, structured data and AI-assisted analysis to support business processes.

Functional MVP · ExperimentalVerified Sep 20266 technologies
AutomationAPIsAIData ProcessingWorkflowsn8n
AUTOMATION SYSTEM
InputWorkflowReview
Structured datan8n
Assisted analysisAI
Context

Mencora is an experimental system exploring how structured workflows and AI-assisted analysis can support repeatable business processes.

01

Problem

Business processes frequently combine repetitive coordination, fragmented information and decisions that still need human context.

02

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.
03

Solution

An automation-oriented system that brings workflows, structured data, integrations and assisted analysis into a single process.

Under the Hood

The technical layers that turn the product idea into a working system.

01

Workflow orchestration

Eight n8n workflows coordinate launch, prospecting, analysis, review, drafts and reads.

n8n
02

Structured analysis

Responses API returns strict JSON with low reasoning and no web tools.

03

Operational data

Mencora CRM uses Google Sheets for campaigns, prospects, analyses and drafts.

04

Human review

Explicit states gate Gmail draft creation and persist its ID.

System Explorer

Select a node to inspect its role and relationships.

Workflow

The verified draft-only path, orchestrated by n8n and gated by a human.

Technical decisions
01 / 05

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.
What broke
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
Engineering learnings
  • n8n batch semantics require deliberate item-level design.
  • Structured output makes automation inspectable.
  • Human review and draft-only delivery are intentional boundaries.
Next iteration
  • Compare Gemini / Perplexity only as a future experiment.
  • Expand commercial validation without implying customers or revenue.
Evidence
WORKFLOWS8 functional workflows

00–08 set including launcher, analysis, review and list reads.

E2ECampaign → Gmail Draft

A documented campaign completed the draft-only path.

BATCH VALIDATION10 companies · 5 drafts

Five eligible rows persisted in the documented batch.

DEDUPE29 rows → 28 campaigns

WF08 retained the greatest row_number and sorted newest-first.

LIMITATIONSheets quota

Mitigated with retries, waits and refetch policy; the structural limit remains.

Technical stack

Mencora · Technical stack

AutomationAPIsAIData ProcessingWorkflowsn8n
Current stateFunctional MVPExperimental