n8n vs Make vs Zapier for lead routing: what we actually deploy
Two real lead-routing builds, one on n8n and one on Make, and the rules we use to pick between the three on cost, branching, AI steps and failure handling.
We have built lead-routing pipelines on all three. Here is how we choose, with two real builds as the reference points, and the costs that actually matter.
Lead routing sounds simple: a form is filled in, the right person hears about it, the lead ends up in the CRM. In practice it is the automation most businesses get wrong first, because it touches five systems, runs unattended, and nobody notices when it silently stops.
The two builds we compare against
A revops pipeline on n8n. A client needed every inbound lead classified and in the CRM within minutes. We self-hosted n8n on a small AWS EC2 instance and built: Google Form → Google Sheets → GPT-4o-mini classifies the lead as Hot, Warm or Cold with a one-line reason → Slack notification to the right channel → HubSpot contact and deal created → QuickBooks customer record. End to end in under two minutes, with no one touching it.
A lead-to-revenue pipeline on Make. A multi-location residential cleaning company runs on GoHighLevel with BookingKoala for scheduling and Google Local Services Ads for leads. Make polls Gmail every 15 minutes, parses the Local Services Ads emails, creates tagged contacts and opportunities in GoHighLevel by service area (ten of them), and hands off to a seven-stage pipeline with six automated workflows that run from first contact to win-back, with stop-on-response logic so nobody gets a follow-up after they have already replied.
Different tools, both the right call. Here is why.
What actually differs between them
How you pay. Zapier charges per task (each step that runs). Make charges per operation, which is similar but finer-grained. n8n Cloud charges per workflow execution; self-hosted n8n is free software and you pay for the server. For a pipeline with eight steps that handles 2,000 leads a month, that is 16,000 Zapier tasks, roughly the same in Make operations, and 2,000 n8n executions. Volume is the first thing that moves the answer. Check current pricing pages; the tiers change often.
Branching and data shaping. Make's visual router and array handling are the best of the three for "if the lead is in area X and the service is Y, do Z". Zapier's paths work but get awkward past three branches. n8n handles anything, because you can drop into JavaScript when the visual nodes run out.
AI steps. All three can call OpenAI or Anthropic. n8n has the most natural fit: a model call is just another node, and you can keep the prompt, the parsing and the fallback in one place. On Zapier and Make, AI steps work but every call costs a task or operation, and retries multiply that.
What happens when something fails. This is the one that matters at 2 a.m. Zapier retries and emails you. Make lets you build explicit error routes. n8n gives you full error workflows, retries per node, and the execution log is yours to query. If a lead is worth real money, you want the pipeline to tell you it dropped one, not to find out from the salesperson a week later.
Who maintains it. Zapier is the easiest for a non-technical owner to open and understand. Make is a step up. n8n self-hosted means someone has to patch a server; n8n Cloud removes that but you are back to paying per execution.
Where the data lives. Self-hosted n8n keeps every lead on your own infrastructure. For businesses with client data obligations (recruitment, healthcare adjacent, anything with a GDPR exposure) that alone can decide it.
Our decision rules
- Under 500 leads a month, standard tools, nobody technical in-house: Zapier. It will cost more per lead than the others, but the total is small and the owner can fix it themselves.
- High branching, several locations or service lines, GoHighLevel or similar in the stack: Make. That is why the cleaning company runs on it.
- Volume above a couple of thousand events a month, an AI classification step, or data that should not leave your servers: n8n, self-hosted. That is the revops pipeline.
- An exception to all three: if the pipeline is the business (an outbound engine, a marketplace, anything where latency and reliability are the product), it should be code, not a workflow tool.
What a build costs
A lead-routing pipeline like the two above is a fixed-price build for us, delivered in one to two weeks. The price moves on how many systems are involved, whether an AI step needs tuning on your real leads, and whether we are also cleaning up the CRM pipeline stages the automation feeds into. Tool subscriptions are on top and are yours.
The thing people underestimate is not the build but the month after. We leave every pipeline with a failure alert, a weekly summary of what it processed, and a one-page runbook for the owner. A pipeline nobody can see is a pipeline that stops quietly.
Where to start
If you are not sure which tool fits, write down three numbers: leads per month, systems the lead has to reach, and how many minutes a lost lead costs you. Send us those and the form you use today, and we will tell you which of the three to build on and what it would cost. If it turns out an entry-level Zapier plan is the answer, we will say so.
