MESH FLOW
Menu
·

AI Strategy

Which AI Automation Platform Should You Build On? n8n, Zapier, Make, or Code

Zapier, Make, n8n, or custom code — the platform matters far less than three things nobody budgets for: who maintains the workflow, where your credentials live, and whether the process was worth automating at all.

Which AI Automation Platform Should You Build On? n8n, Zapier, Make, or Code

The short answer

Pick the cheapest platform your hardest workflow can survive on. Zapier and Make win on breadth — Zapier carries more than 9,000 integrations. n8n wins when you need self-hosting, data residency, or real code inside a step. Custom code wins when the logic is your product. But platform choice is a third-order decision: McKinsey found AI's contribution to EBIT flat at 37% in 2026, whatever the tooling.

The average organization now runs 957 applications and has connected just 27% of them, according to Salesforce's 2026 Connectivity Benchmark with Vanson Bourne and Deloitte Digital (via CIO Dive, February 2026). That is the real constraint on your automation program. Not whether you picked Make over n8n.

Yet platform choice is where most mid-market teams spend their deliberation. They compare integration counts, argue about per-task pricing, run a two-week bake-off — and then eighteen months later the thing that breaks isn't the platform. It's that the person who built the workflow left, nobody documented the credential it holds, and the process it automated was never worth automating.

If you're the COO or head of operations holding the budget, here's the decision framework we actually use, and the order the constraints come in.

What actually decides which automation platform you should use?

Three constraints, in this order: where your data is legally allowed to live, who will maintain the workflow twelve months from now, and how much branching logic the process genuinely has. Integration counts and pricing pages are tiebreakers, not criteria.

Data residency and credential custody. An automation platform holds the keys to every system it touches. If your workflow moves regulated data — health records, EU personal data, client financials — the question of whose infrastructure that runs on gets decided before anything else. This is the single most common reason teams move off closed SaaS platforms, and it's a compliance answer, not a preference.

Maintenance ownership. Every workflow is a liability with a maintenance schedule attached. Retool's 2026 survey of 817 builders found 26% name maintenance burden as a blocker to further automation and 21% point to fragile existing automations (Retool, February 2026). If no named person owns the workflow, the platform is irrelevant — it will rot either way.

Logic complexity. Linear "when X happens, do Y" chains run fine anywhere. Workflows with real branching, retries, state, and conditional error handling need a platform that treats code as a first-class citizen rather than an escape hatch. Be honest about which one you have — and map the workflow before you automate it so you actually know.

When is a closed SaaS platform like Zapier or Make the right answer?

When your workflows are linear, cross many SaaS apps, touch no regulated data, and will be maintained by someone in operations rather than engineering. That describes most of a mid-market company's first twenty automations, and paying for breadth is the right call.

Breadth is the genuine advantage. Zapier offers more than 9,000 integrations and Make more than 3,000, compared with roughly 1,000 for n8n (Sacra, 2026). If your stack includes a niche vertical tool, the pre-built connector is worth more than any architectural elegance.

Watch the pricing shape, not the headline price. Zapier charges per task and Make per credit, which means a ten-step workflow costs ten times a one-step workflow every time it runs. n8n charges per complete execution with unlimited steps. On simple automations this is noise; on multi-step processes running thousands of times a month it inverts the cost comparison entirely.

The honest ceiling. These platforms are excellent until you need a private network connection, a specific data region, or thirty lines of logic that don't fit the node library. Then you're building increasingly baroque workarounds, and that's the signal to move — not a reason to have started somewhere else. We've written separately about when off-the-shelf tools are enough.

When should you move to n8n or another code-capable platform?

When at least one of three things is true: the workflow must reach systems inside your network, your data cannot leave a specific jurisdiction, or the logic needs real code in a step. Any one of those makes the breadth advantage of closed SaaS irrelevant.

n8n is the serious option in this tier, and the market has validated it hard. SAP invested at a $5.2 billion valuation in May 2026, roughly double the prior round, and signed a multi-year deal to embed n8n inside Joule Studio (TechFundingNews, May 2026). Sacra estimates it reached $100M ARR in April 2026, serving more than 1,400 enterprise customers and 1.7 million monthly active builders, with over 80% of platform workflows involving AI agents by late 2025.

What you're actually buying is control. Self-hosting gives you network access to private systems, control over data residency, and control over upgrade timing. For a regulated mid-market company, that last one matters more than people expect — a vendor-forced upgrade that breaks a billing workflow is a bad week.

What you're buying is not simplicity. The same report notes that self-hosting "may require more internal platform engineering and governance design." That is the trade, stated plainly. Which brings us to the part most comparison posts skip.

What does self-hosting actually cost you?

It costs you a patching obligation, and that obligation is not theoretical. In January 2026, n8n disclosed CVE-2026-21877, rated CVSS 10.0, allowing authenticated remote code execution and, in n8n's own words, "full compromise of the affected instance" — affecting both self-hosted and cloud deployments (The Hacker News, January 2026).

It got sharper in March. CVE-2026-27493 (CVSS 9.5) allowed unauthenticated code execution through n8n's public Form nodes — researchers at Pillar Security demonstrated running arbitrary shell commands by submitting a payload into the Name field of a "Contact Us" form. Three further flaws rated 9.4 or above landed in the same release, and Pillar showed an attacker could read the instance encryption key and decrypt every stored credential: AWS keys, database passwords, OAuth tokens, API keys (The Hacker News, March 2026).

None of this makes n8n a bad choice — mature, widely deployed software gets scrutinized, and the team shipped fixes. It makes the trade explicit: self-hosting moves risk from the vendor to you. If you self-host an automation platform, someone on your team owns patching it on a schedule, and that person needs to be named before you deploy, not after an advisory lands. Budget it the way you'd budget monitoring agents in production.

When should you skip platforms and just write code?

When the logic is a differentiator, when the workflow is the product, or when the orchestration is complex enough that the platform becomes overhead rather than leverage. Not because a coding agent can scaffold something in an afternoon.

That last caveat is doing real work right now. McKinsey's 2026 State of AI survey found that 32% of respondents decided against buying a software product or feature because they could build it internally with agentic coding tools (McKinsey, August 2026). Retool found 35% of its builders had already replaced at least one SaaS tool with a custom build, and 78% expected to build more in 2026.

The same Retool data shows what that produces: 60% of builders had built something outside IT oversight in the past year, and 25% did so frequently. Build velocity has gone up. Governance hasn't. The thing that was expensive about building was never the first version — it was year three, and agentic coding tools have not changed that. We've put the full numbers in the real cost of building AI agents in-house.

Why platform choice is a third-order decision

Look at what actually kills these projects. Gartner predicts more than 40% of agentic AI projects will be canceled by the end of 2027 — because of escalating costs, unclear business value, or inadequate risk controls (Gartner, June 2025). Not one of those three is a platform problem. Gartner's analysts also estimated that only about 130 of the thousands of vendors claiming agentic AI were building anything genuinely agentic — "agent washing" (via Staffing Industry Analysts, June 2025).

The McKinsey data points the same direction. AI's contribution to EBIT held at 37% in 2026, essentially unchanged year over year, and only 6% of organizations qualify as high performers. What separates them isn't tooling: nearly three-quarters of high performers fundamentally redesigned workflows around AI, versus one-quarter of everyone else. They changed the process. Everyone else inserted AI into the process they already had — on whichever platform.

When we run an automation engagement at Mesh Flow, platform selection is usually a thirty-minute conversation at the end of process mapping. By then the constraints have chosen the platform for us: the data residency requirement, the integration that doesn't exist, the maintenance owner we can realistically name. Teams that start with the platform tend to end up with a tool looking for a job.

And the cost of getting that order wrong is visible in the sprawl. Zapier's own survey of over 500 enterprise leaders found 70% had not moved beyond basic integration for AI tools, 31% were discovering rogue AI tools in their organization every month, and only 35% said their AI tools went through proper approval channels (Zapier, December 2025).

Frequently Asked Questions

Is n8n better than Zapier?

For different jobs. Zapier's 9,000+ integrations beat n8n's ~1,000 for breadth, and its per-task pricing is fine on short workflows. n8n wins on self-hosting, data residency, code-in-a-step, and execution-based pricing that doesn't penalize multi-step processes. Pick by constraint, not by review score.

Is self-hosting an automation platform cheaper?

Usually on license cost, rarely in total. You take on patching, upgrades, monitoring, and security response — and in 2026 n8n shipped fixes for a CVSS 10.0 remote code execution flaw and an unauthenticated flaw exposing stored credentials. Self-hosting is cheaper only if you genuinely staff the operations work.

Should we just build our automations with AI coding tools instead?

Sometimes, and more often than two years ago — 32% of McKinsey's 2026 respondents skipped a software purchase because they could build it internally. But 60% of Retool's builders shipped something outside IT oversight, and 26% cite maintenance burden as their main blocker. Build when the logic is a differentiator, not because generation got cheap.

How many automation platforms should a mid-market company run?

One primary, with a clear rule for exceptions. Running three because different teams picked their own is how you reach the state Salesforce's benchmark describes — 957 applications, 27% connected. Consolidation is worth more than optimization.

Does the platform choice affect whether AI automation delivers ROI?

Barely. EBIT impact from AI held flat at 37% across 2025 and 2026 while tooling improved substantially. Workflow redesign is the variable that moves: high performers redesign, and they're 6% of the market. Platform is a constraint-satisfaction problem, not a growth lever.

The bottom line

  • Order the constraints correctly: data residency, then maintenance ownership, then logic complexity. Integration counts and pricing are tiebreakers.
  • Closed SaaS (Zapier, Make) for linear, multi-app, non-regulated workflows owned by operations. n8n or code-capable when you need private systems, a data region, or real code. Custom code when the logic is a differentiator.
  • Self-hosting is a staffing decision, not a savings decision. Name the patch owner before you deploy.
  • Nothing on this page matters as much as process redesign. Gartner expects 40%+ of agentic projects canceled by 2027 over cost, value, and risk — none of them platform problems.

If you'd rather map the process first and let the constraints pick the platform, that's the engagement we run. Mesh Flow builds and operates these systems in production, which is also why we're blunt about what breaks.

Sources

Founder of Mesh Flow. Builds and ships AI automation systems for mid-market companies and founders.

Related reading