What Zapier and Make Actually Are
Zapier and Make are no-code automation platforms that connect different software services through pre-built connectors, allowing you to trigger actions in one service based on events in another without writing code. Zapier was first to market and has the largest library of app integrations (over 6,000 apps), the most accessible interface, and the strongest brand recognition among non-technical users. Its execution model is sequential: a trigger fires, then steps execute in order, with conditional branching available on higher plans. Make (formerly Integromat) has a visual scenario builder that is more powerful than Zapier's for complex flows: parallel branches, iterators, aggregators, and more sophisticated data manipulation. Make is generally cheaper than Zapier at equivalent automation volumes and handles more complex workflow logic without requiring workarounds. Both platforms offer managed execution infrastructure: you do not need to deploy or maintain servers, handle retries, or manage execution state. Both platforms have limits on monthly task/operation volumes, execution frequency (most Zapier plans check for triggers every 15 minutes, not instantly), data payload size, and API request complexity that matter at scale.
What Custom Code Automation Actually Is
Custom code automation means writing and deploying your own automation logic as code: a Node.js or Python service running on Railway or Fly.io, a set of serverless functions on AWS Lambda or Vercel, a background job system using tools like BullMQ or Inngest, or a full workflow orchestration platform like Temporal. The spectrum is wide. At the simple end, a webhook handler that receives an event, transforms the data, and calls an API is custom automation. At the complex end, a durable workflow engine that handles multi-step AI processing tasks with retries, human-in-the-loop steps, and compensation logic on failure is also custom automation. The defining characteristic is that you write and own the logic, you deploy and operate the infrastructure, and you are responsible for reliability, error handling, and maintenance. The trade-off is capability ceiling: custom automation has none, outside the limits of available APIs and your engineering team's time. You can implement any logic, handle any data structure, integrate with any service (not just those with pre-built connectors), and optimise for any performance or cost requirement. The cost is that it takes engineering time to build and ongoing time to maintain.
Cost at Scale and the Tipping Point
No-code automation platforms charge based on task or operation volume. Zapier's pricing starts from GBP 18/month for 750 tasks per month and scales steeply: 50,000 tasks per month costs GBP 400-700/month depending on plan features. Make is cheaper per operation but follows the same scaling curve. For low-volume automation (a few hundred operations per month connecting internal tools), Zapier or Make cost is negligible and the time savings from no-code are well worth it. The tipping point comes when automation volume drives platform costs above the cost of maintaining equivalent custom code. A SaaS product running 100,000 automation operations per month is paying GBP 1,000-3,000/month on Zapier at scale. A custom Node.js worker on Railway handling equivalent workloads costs GBP 50-200/month in infrastructure. The engineering time to build and maintain that custom worker is real, but for any team with an existing engineering function, the cost crossover happens at volumes that many SaaS products reach within 12-18 months of launch. Calculating this crossover for your specific automation volume and engineering cost is an important business decision.
Reliability, Execution Guarantees, and Error Handling
Production automation reliability is where no-code platforms and custom code diverge most significantly. Zapier and Make provide managed reliability, but with platform-level constraints. Zapier checks for new trigger events on a polling interval (not instantly on standard plans), which means a trigger event may not execute for up to 15 minutes. Execution failures are logged in the platform but retry behaviour has limits. Payload size caps and execution time limits can cause silent failures on certain data types. When Zapier or Make experience platform outages (which do happen, typically quarterly), all your automations stop until the platform recovers, which is outside your control. Custom automation gives you complete control over execution guarantees. A properly implemented background job system with BullMQ or Inngest supports dead letter queues for failed jobs, configurable retry strategies with exponential backoff, webhook delivery with confirmation, and exactly-once execution semantics where needed. If your custom automation service goes down, you control the recovery. For mission-critical automations in a SaaS product (billing events, user onboarding flows, compliance-required notifications), the execution guarantees and error visibility of custom code outperform no-code platforms.
AI Integration Depth and LLM Workflow Support
Zapier and Make have added AI capabilities to their platforms: Zapier has OpenAI integration, Make has AI modules, and both allow you to call LLM APIs within automation flows. For simple AI automation (summarise this email, classify this support ticket, extract data from this document), no-code AI integration is adequate. For anything more sophisticated, the limitations become apparent quickly. No-code platforms do not support streaming LLM responses, which means AI-generated content is blocked until the full completion arrives. They do not support complex RAG pipelines with vector database queries. They cannot dynamically construct prompts based on retrieved context. They cannot implement multi-agent workflows where AI agents call tools and make decisions across multiple steps. Custom code automation is the only viable path for AI workflows of any real complexity. SpeedMVPs builds custom automation infrastructure for AI SaaS products using Inngest for durable workflow orchestration or BullMQ for job queuing, with full LLM SDK integration (streaming, retry, cost tracking) that no-code platforms simply cannot replicate.
When Zapier or Make Is the Right Choice
Zapier and Make earn their place for specific use cases. Internal business operations that are genuinely simple and low-volume: sending a Slack notification when a new customer signs up, adding a row to a Google Sheet when a payment is received, sending a welcome email sequence through Mailchimp when a trial starts. These are tasks that a non-technical founder can set up in an afternoon without engineering involvement, and the cost is trivial relative to the time saved. Connecting third-party SaaS tools that your product integrates with but does not deeply own: syncing customer data to HubSpot, sending support tickets to Intercom, triggering Stripe invoice creation. If the automation is simple, the volume is low, and the data is not sensitive, no-code is the right tool. Pre-product-market fit, where you want to test multiple integrations quickly without committing engineering time to each one. No-code lets you validate whether a particular integration creates value before investing in a custom implementation.
When Custom Automation Is the Right Choice
Custom automation becomes necessary when any of the following are true: the automation volume makes platform costs material; the automation logic is complex enough that no-code workarounds are unreliable or fragile; the automation involves sensitive personal data that should not transit a third-party platform (GDPR considerations mean that customer PII flowing through Zapier's US-hosted infrastructure needs careful legal assessment); the automation includes AI inference that requires streaming, multi-step reasoning, or vector retrieval; execution reliability is mission-critical and you need full control over retry strategies and failure modes; or the automation must integrate with internal systems or databases that do not have Zapier/Make connectors. GDPR is a particular consideration for UK and EU SaaS products: if your Zapier automation processes personal data of EU residents, Zapier is a data processor and you need to review their DPA. For sensitive data, custom automation within your own GDPR-compliant infrastructure eliminates the third-party processor dependency entirely.
Verdict
The right automation strategy for most SaaS products is a deliberate combination: no-code platforms for simple, low-volume, non-sensitive operational glue, and custom code for anything involving AI inference, high volume, sensitive data, or mission-critical execution guarantees. The mistake is picking one approach and trying to apply it universally. Using Zapier for complex AI workflows leads to fragile workarounds and capability ceilings. Building custom automation for every simple notification and CRM sync wastes engineering time. SpeedMVPs builds custom automation infrastructure for AI SaaS products and uses no-code tools pragmatically for simple operational automations. The architecture decision is made at the scoping stage based on volume projections, data sensitivity, and the complexity of the automation logic required.