Define the job before choosing the tool
Glue between SaaS tools with readable history → low-code. Domain engine inside your product → custom. Mixed → n8n plus small services. Writing “automation platform” on a pitch deck does not make custom cheaper.
List irreversible actions (CRM writes, payments, WhatsApp sends). Those need audit and kill switches regardless of stack. AI Automation at Brevia starts from that list, not from a framework preference.
If the brief is “Meta lead to WhatsApp in under five minutes,” that is classic n8n/Make/Zapier territory — see Lead follow-up automation.
Signals the job is glue
Stable public APIs, field mapping, branching by city or source, human-editable filters, and ops wants to own changes without Git. That is n8n’s sweet spot.
Signals the job is software product
Multi-tenant logic, complex pricing, sub-second internals, or automation sold to your customers as a feature. Hire engineers; use n8n only as a prototype.
What n8n gives Indian SMEs that custom often postpones
Visual debugging, credential stores, retry patterns, and a library of connectors. Time-to-first production path is days, not months, when APIs cooperate.
Self-hosting keeps data in your VPS while still avoiding engine invention — Self-host n8n in India. That satisfies many “we need custom for security” requests without a rewrite.
Mistakes still happen — bad webhooks, missing idempotency — documented in Common n8n mistakes. Low-code is not no-discipline.
- Ops can read execution history without IDE literacy
- Faster change cycles for marketing campaign filters
- Lower bus-factor if one developer leaves
- Clear package floors for budgeting
When custom code is the honest answer
Heavy computational workflows, specialised protocols, offline warehouses, or embedding automation in a customer-facing product justify code. So do extreme throughput cases where a visual runner becomes the bottleneck.
Custom also wins when you already have platform engineers, CI/CD, and on-call — and the workflow is core IP. Buying n8n for that team can still help for peripheral glue.
Do not start custom because a freelancer is cheaper “per hour” on paper; unfinished custom is the most expensive automation.
When Playwright is neither n8n nor “full custom product”
Browser scripts for API-less portals are a third category — scoped Playwright jobs triggered by n8n, not a reason to rewrite your CRM sync in Puppeteer.
Hybrid architecture we recommend most often
n8n owns triggers, branching, notifications, and SaaS writes. A small authenticated HTTP service owns pricing, eligibility, or ML scoring. The canvas stays readable; the hard logic gets tests in Git.
Code nodes inside n8n help for fifteen-line transforms. When the code node becomes a novel, extract a service.
WhatsApp sends stay in deterministic nodes with templates — whether the decision engine was low-code or custom. Compliance does not care which repo you used. See WhatsApp Automation.
Interface contract that keeps teams sane
JSON in/out, idempotency keys, versioned endpoints, and timeouts. n8n retries become safe only if the service is idempotent.
TCO over twelve months, not week-one demos
n8n: OTP build + VPS or cloud + owner hours. Zapier/Make: OTP + task/ops fees. Custom: eng months + infra + on-call + slower change requests billed as projects.
Brevia planning floors: Zapier ₹15k+ OTP; n8n/Make/WhatsApp/chatbot ₹18k+; Playwright ₹25k+. Custom quotes only after architecture discovery — How much does AI automation cost in India.
Migrate platforms when task bills hurt — Zapier to n8n migration — before leaping to custom.
Change-request economics
Marketing will ask for new branches monthly. In n8n that is a node edit; in custom it is a ticket. Price the expected change volume into the decision.
Bus factor and documentation
Visual workflows plus a runbook beat undocumented Python on a founder laptop. Custom can be excellent with discipline — budget that discipline explicitly.
Security and compliance are not automatic on either path
Custom does not equal secure; n8n does not equal insecure. Credential storage, least privilege, and PII handling are process choices. Self-host n8n if residency worries you; still patch the VPS.
Unofficial WhatsApp libraries are unsafe in both worlds — WhatsApp Business API vs unofficial.
DPDP-minded logging: avoid dumping full payloads into public Slack. Same rule for code and canvases.
- Separate staging and production credentials
- Audit who can publish workflow changes
- Encrypt secrets; never commit them
- Document data classes touching LLMs if any
Auditability for finance and clients
n8n execution logs often satisfy SME auditors faster than digging through ad-hoc script logs — if you retain them. Custom needs equivalent structured logging from day one.
Hiring and ownership models
Low-code partner + internal ops owner scales for most Chennai SMEs. Custom needs engineers who stay. Freelancer-only custom with no retainer is how systems rot.
Interview questions differ: for n8n, ask about idempotency and error workflows; for custom, ask about queues, testing, and deploy rollback. Vendor checklist: How to hire an AI automation company in India.
HubSpot/Zoho patterns are usually low-code first — HubSpot CRM automation, Zoho CRM automation.
What Brevia will not recommend
Rewriting a working Zap into custom Java “for modernity,” or forcing n8n when you are building a multi-tenant automation SaaS. Fit the tool to the job.
Decision matrix you can paste into a Notion doc
Choose n8n/Make/Zapier when APIs exist, ops must edit, and time-to-value matters. Choose hybrid when one logic island is complex. Choose custom when automation is product IP or scale demands it.
If Sheets is still your CRM, fix that before either path — When to stop using Sheets as CRM.
Booking and WhatsApp spines rarely need custom platforms — Appointment booking automation.
One-week proof that avoids ideology
Prototype the happy path in n8n. If you hit a wall that is truly domain-hard, extract that node to code. If the prototype ships, you just saved a quarter of engineering fantasy.
Red flags in vendor proposals
Custom-only proposals with no low-code option for glue, or n8n-only proposals that ignore an obvious need for a tested pricing service. Balanced partners discuss both.
How we scope this choice from Chennai
Discovery names irreversible actions, volume, owners, and whether automation is glue or product. Then we recommend n8n, Make, Zapier, hybrid, or a custom track under AI Automation.
You leave with a written recommendation, not a framework war. Floors stay published so finance can compare.
Book a Consultation or Contact Us with your current stack diagram — even a photo of a whiteboard is fine.
Artifacts for the build-vs-n8n meeting
API list, peak weekly volume, who will edit after go-live, and one success metric. Missing any of these usually means you are not ready for custom.
What success looks like in ninety days
Production path live, owner trained, error rate acceptable, and a clear backlog for hybrid extraction if needed — not a half-finished microservice farm.
Ready to automate one painful workflow?
Tell us your stack — n8n, Make, Zapier, WhatsApp, or Playwright. Chennai HQ · India delivery · reply within one business day.