n8n Automation SaaS
One controlled operating surface for channel conversations, AI assistance, human handoff, and business actions, with tenant-owned records and reviewable automation boundaries instead of scattered workflow logic.
A configurable automation platform combining tenant-facing and operator panels, a business API, dedicated workers, and version-controlled n8n workflows for customer conversations and operational tasks.
Project brief
The product context, ownership, and current evidence.
Target users
Businesses and platform operators coordinating Facebook, Instagram, WhatsApp, customer conversations, leads, bookings, orders, knowledge, and AI-assisted responses across separate teams or tenants.
Intended buyer outcome
One controlled operating surface for channel conversations, AI assistance, human handoff, and business actions, with tenant-owned records and reviewable automation boundaries instead of scattered workflow logic.
Owned scope
Product architecture · Full-stack engineering · AI and automation systems
Evidence status
API, worker, two web panels, migrations, channel/AI/business modules, tests, and workflow artifacts are present in the private repository. This review does not establish a live end-to-end production SaaS.
Problem
A workflow editor alone cannot safely own tenant authorization, durable customer records, AI configuration, retries, human handoff, and business-state changes. The product needed clear boundaries between its application database, API, background worker, existing n8n runtime, and channel providers.
Approach
Built a TypeScript monorepo with a Fastify API, BullMQ worker, PostgreSQL/pgvector schema, Redis coordination, customer and super-admin Next.js panels, channel adapters, AI and retrieval modules, server-side media integration, and a sanitized six-workflow n8n bundle with validation and deployment tooling.
Current outcome
The repository implementation and its CI-oriented verification path are in place. The n8n workflow bundle validated locally on 2026-09-21. Live workflow import/cutover, Meta permissions and credentials, storage and backup checks, staging load/recovery exercises, and a controlled pilot remain external production gates.
Maturity & lifecycle
What is complete, active, and still planned.
API, worker, two web panels, migrations, channel/AI/business modules, tests, and workflow artifacts are present in the private repository. This review does not establish a live end-to-end production SaaS.
Evidence reviewed
Discovery
CompleteCustomer messaging, tenant/business ownership, channel integration, human handoff, and operational failure cases were defined.
Scope & architecture
CompletePostgreSQL was assigned durable business state; the API owns authorization and mutations; Redis/workers handle asynchronous work; n8n remains an orchestration runtime.
Design & prototype
CompleteCustomer and operator panels, workflow contracts, and channel-to-business actions were implemented as reviewable repository slices.
Build & integrate
CompleteThe API, worker, two panels, database migrations, AI/RAG modules, channel adapters, and version-controlled n8n workflow bundle were implemented.
Verify & launch
In progressRepository verification and workflow-bundle validation exist; live n8n import, Meta approval, Media Storage acceptance, staging recovery/load, and pilot checks remain.
Handover & iterate
PlannedProduction cutover evidence, runbooks, provider operations, monitoring, and post-pilot improvement follow the external acceptance gates.
Real-world solutions
How controlled automation fits real customer operations.
Each pattern links a specific operating problem to the implemented application boundary and makes remaining provider acceptance visible.
Commerce and social selling
Operational problem
Product questions, media, stock context, and order requests arrive through several social channels while staff need one consistent record of the customer conversation.
Solution pattern
Use channel-normalized messages, business-owned catalog data, guarded AI response planning, order actions, and staff takeover in one tenant-scoped workflow.
Live Meta permissions, connected channel accounts, and provider messaging behavior still need credentialed production acceptance.
Appointment and service operations
Operational problem
Incoming enquiries, availability questions, follow-ups, and bookings are easy to lose when they stay inside chat threads or disconnected automations.
Solution pattern
Capture conversations as durable records, apply validated booking and lead actions, schedule follow-ups through workers and n8n, and expose the state to staff.
Customer support and handoff
Operational problem
An automated reply can conflict with a human agent or continue after a sensitive issue needs staff ownership.
Solution pattern
Keep AI and HUMAN modes explicit, suppress pending AI output after takeover, and retain conversation and audit context for the operator.
Multi-brand or agency operations
Operational problem
Different businesses need separate users, channel credentials, knowledge, quotas, and reporting without leaking data across tenants.
Solution pattern
Scope API authorization, business records, encrypted provider configuration, media mapping, queue namespaces, and usage views by tenant and business.
System map
How the product boundary fits together.
- 01
Customer and operator panels
Tenant-facing configuration and conversation workflows are separated from platform administration.
- 02
Application API and PostgreSQL
Authorization, business mutations, audit, conversations, actions, and durable state remain in the application boundary.
- 03
Redis and dedicated worker
Queues and short-lived coordination support retryable background work without making Redis the business source of truth.
- 04
n8n and channel providers
Versioned workflows orchestrate integrations through application contracts; provider acceptance and live cutover are still pending.
Product benefits
Practical value without overstating the evidence.
Business data stays authoritative
PostgreSQL holds tenant, conversation, order, booking, and usage records; n8n orchestrates work through application contracts rather than becoming an unreviewable second business database.
Automation can yield to people
The conversation model supports human takeover and suppresses pending automated replies when a staff member owns the turn.
One workflow across channels
Channel adapters normalize Facebook, Instagram, and WhatsApp events so AI, business actions, media, and audit rules can share a consistent application boundary.
AI remains configurable
Provider and model selection, prompt versions, retrieval, training candidates, usage, fallback, and review controls are represented in the application rather than buried in individual workflows.
Claim boundary
Constraints and unresolved risks
- The application repository is private. Credentials, customer messages, internal service endpoints, and infrastructure details are not public portfolio evidence.
- Six version-controlled workflow artifacts validated locally, but this does not prove they have been imported, activated, or accepted on the live n8n runtime.
- Facebook, Instagram, and WhatsApp provider permissions, real webhook subscriptions, and end-to-end messaging need credentialed production tests.
- Production Media Storage mapping, backup/restore, load and failure-injection exercises, and controlled pilot operation remain unverified.
- The billing model is provider-ready; no payment provider has been selected or accepted for live checkout.
Evidence base
What supports this case study
Repository implementation review
The private monorepo contains the Fastify API, BullMQ worker, customer and super-admin panels, PostgreSQL migrations, channel/AI modules, tests, and operational documentation. This is source evidence, not public deployment evidence.
n8n bundle validation
The repository validator accepted bundle 1.0.0 with six sanitized workflow JSON artifacts on 2026-09-21. Live import and activation are separate gates.
Recorded CI verification
The implementation status records isolated PostgreSQL/pgvector and Redis migration checks, workflow validation, TypeScript, automated tests, and application builds. These records do not substitute for production provider acceptance.
Production readiness boundary
The implementation status explicitly keeps live n8n cutover, Meta approval, Media Storage verification, backups, staging load/recovery, and pilot checks open.
Engineering highlights
- Customer and super-admin panels around a tenant-scoped Fastify API
- Durable PostgreSQL business records with separate n8n workflow state
- BullMQ worker and Redis coordination for retryable background work
- Facebook, Instagram, and WhatsApp adapter code with human handoff
- Configurable AI providers, training versions, retrieval, and usage controls
- Six validated, version-controlled n8n workflow artifacts
Technology and domains
01 / Product boundary
The workflow editor is not the business database.
The platform separates customer-owned state from automation execution so rules remain testable and recoverable.
- 01.1PostgreSQL owns tenants, businesses, channels, conversations, actions, and usage.
- 01.2The API enforces tenant/business authorization, validation, audit, and idempotency before a business mutation.
- 01.3n8n and the worker orchestrate or execute work through application contracts rather than directly owning business truth.
02 / Customer operations
Conversations and business actions share a controlled path.
The tenant-facing product joins channel messages, staff handoff, knowledge, media, leads, orders, bookings, and follow-ups.
- 02.1Normalized channel adapters reduce channel-specific logic in AI and business-action workflows.
- 02.2Human takeover prevents pending automated replies from racing a staff response.
- 02.3Dynamic business records keep product or service facts under customer control instead of letting AI invent them.
03 / AI and reliability
AI and async work have explicit review and retry boundaries.
Provider routing, prompt versions, retrieval, worker jobs, and operational controls are visible product concerns.
- 03.1AI provider configuration, usage tracking, prompt candidates, rollback, and scoped retrieval live outside individual n8n nodes.
- 03.2Redis/BullMQ jobs carry retry and dead-letter behavior while durable business changes remain in PostgreSQL.
- 03.3Operator views expose queue, workflow-bundle, infrastructure, and security signals without treating a green dashboard as provider acceptance.
04 / Verification and cutover
Repository readiness and live service acceptance are different milestones.
Implementation evidence supports an engineering case study, while external systems still determine production readiness.
- 04.1The six-workflow bundle passed repository validation at the recorded checkpoint.
- 04.2The status record describes migration, typecheck, test, build, and tenant/security integration gates for the application code.
- 04.3Live n8n import, Meta permissions, real storage and backups, load/recovery exercises, and a controlled pilot remain to be accepted.
Use cases / Client relevance
Where this product evidence can support a real engagement.
These applications reflect demonstrated product reasoning and the stated maturity boundary—not an unsupported production or client-result claim.
Explore AI-Enabled Product Engineering ↗- Connect commerce conversations to approved product data, staff handoff, and order intake across social channels.
- Build service enquiry, booking, lead, and follow-up workflows without leaving the only record inside a chat thread.
- Introduce AI-assisted support while preserving human ownership of sensitive or escalated conversations.
- Give multi-brand operators tenant-separated channel configuration, knowledge, quotas, and audit history.
Questions teams ask
Scope, safety, and evidence boundaries.
Is this automation SaaS live for customers?
The repository implementation is built, but this case study does not claim a live end-to-end production launch. Existing n8n runtime import, Meta approval, storage, backup, staging, and pilot acceptance are separate gates.
Does n8n store the application business data?
No. PostgreSQL is the SaaS source of truth. n8n owns workflow definitions and execution state and uses approved application contracts for business actions.
Which customer workflows can it support?
The implemented product models channel conversations, AI-assisted replies, human handoff, leads, orders, bookings, follow-ups, media, knowledge, and operator oversight. Actual channel-provider use requires connected accounts and live verification.
Are payments already integrated?
No payment provider has been selected. Billing-ready records do not establish provider checkout or webhook acceptance.
Related engineering notes
Related engineering notes
Have a related product problem?