Santi: I was sitting in this beachside café in Koh Phangan — terrible wifi, but the view was incredible — when my phone started buzzing. Three messages in a row. All from the same client.
Kira: Eight thousand a month client.
Santi: Eight thousand a month. My biggest one. And the messages were basically — "Why is nothing working? Why are my leads not getting routed? What happened to the automation?"
Kira: This was the Make.com API update.
Santi: The Make.com API update that I didn't know about because I was island-hopping for two weeks without checking their changelog. And here's what actually happened — they deprecated one webhook format, introduced a new one, and my entire lead routing system just... stopped. No errors in the dashboard. No alerts. Just silence.
Kira: How long had it been broken?
Santi: Four days. Four days of leads hitting a dead endpoint while I was posting sunset photos. By the time I got back online and patched it, the client had already started shopping for alternatives.
Kira: And this is the important part — it wasn't the API change that killed the relationship. It was the four days of silence.
Santi: Look, APIs change. Services go down. That's the reality of building on other people's infrastructure. But if you're running a location-independent business — if you're literally on the other side of the world from your automations — you need monitoring that works when you're offline. You need retries that don't need you. You need, what, seven core systems that can run themselves.
Kira: Seven systems. And we're going to map exactly what those are, what they cost at different revenue levels, and — this is critical — how to make them bulletproof.
Santi: By the end of this episode, you'll have a complete automation stack mapped out — seven systems, specific tools at each revenue level, and the exact monitoring setup that would have saved my eight-thousand-dollar client.
Kira: Plus the calculator to figure out whether Zapier, Make, n8n, or Pipedream actually makes sense for your specific workflows. Because the unit economics are wildly different.
Kira: Alright, so seven systems. And I want to be really clear about why it's these seven specifically.
Santi: Because these are the seven things that break when you're in a different time zone. Every single one.
Kira: Every single one. Okay, here's the stack: Intake — that's forms, lead capture, initial data collection. CRM and Lead Nurture — where those leads go and how they get warmed up. Project and Client Ops — task management, deliverables, the actual work. Knowledge and Docs — your SOPs, templates, the brain of your business. Content Engine — everything that publishes on your behalf. Finance Ops — invoicing, payments, forecasting. And then Monitoring and Alerts — the thing that tells you when any of the other six break.
Santi: Plus what we're calling the Agentic Experiments lane. Which is where you test the AI agent features that all these platforms are rolling out. But — and this is important — you add agents after the deterministic stuff works, not before.
Kira: Right. Automation first, agents second.
Santi: Let's start with the economics though. Because if you pick the wrong platform for your workflow type, you can literally ten-x your costs for no reason.
Kira: Walk me through the billing units again. Because this is where people mess up.
Santi: So Zapier charges per task — that's every successful action. Not triggers, just actions. If your Zap has five action steps, that's five tasks every time it runs. Make charges per credit, which is basically per operation, but triggers and searches also cost credits even when they return nothing. N8n charges per workflow execution regardless of how many steps are inside. And Pipedream charges per compute time — one credit per thirty seconds at their base memory tier.
Kira: Which means?
Santi: Which means if you're running a newsletter automation that has, let's say, ten steps — pull content, format, personalize, segment, send — that's ten tasks in Zapier every single run. But in n8n, it's one execution no matter how complex. So we're talking about, what, like three dollars versus thirty cents at scale.
Kira: Okay but here's what you're not considering — Zapier has all these built-in steps that don't count as tasks. Filters, formatters, paths, delays. So if you design your Zaps to use those heavily—
Santi: You can cut your task usage in half. Exactly. Same with Make — if you understand that iterators multiply your credit usage per bundle, you design differently. You consolidate, you batch, you avoid unnecessary fan-outs.
Kira: This is why we built the calculator. Because you need to count your actual steps, your run frequency, your error rate, and then model it across platforms.
Santi: Alright, System One: Intake. This is your front door. Forms, landing pages, anywhere leads enter your system.
Kira: MVP setup?
Santi: Typeform or Tally — both have solid free tiers — into Google Sheets or Airtable. One Zapier task or Make credit per submission. Dead simple. But here's where it breaks: Google Sheets has API quotas. You hit more than sixty requests per minute, you get a four-twenty-nine error, and your forms just stop working.
Kira: We had this happen during a Product Hunt launch. Three hundred signups in an hour, and suddenly nothing was getting through.
Santi: Right. So at ten-k MRR, you upgrade. You add a proper database — Supabase or Postgres — and you implement retry logic with exponential backoff. When you get that four-twenty-nine error with a Retry-After header, you actually wait those seconds before trying again.
Kira: And idempotency keys.
Santi: Yes! This is straight from Stripe's playbook but it applies everywhere. Every form submission gets a unique key, so if your automation retries, you don't create duplicate records. This alone would save most people hours of manual cleanup every month.
Kira: System Two: CRM and Lead Nurture. This is where the economics really diverge between platforms.
Santi: Here's what actually happened with one of my clients. They had a lead enrichment flow — form submission, enrich with Clearbit, score with their custom model, route to the right salesperson. In Zapier, that's four tasks per lead. In Make, it's five credits because the search operation costs a credit even when Clearbit returns nothing. But in n8n—
Kira: One execution.
Santi: One execution for the entire flow. So at a hundred leads per day, we're talking about twelve thousand tasks versus three thousand executions per month. The cost difference is like seventy percent.
Kira: But — and this is important — n8n requires more technical setup. You need to handle errors yourself, build your own retry logic, manage your own queues.
Santi: Which brings us to the reliability layer. Every CRM update, every lead assignment, needs idempotency. You cannot accidentally create two contacts or assign the same lead twice because your webhook got retried.
Kira: Imagine you're in Guatemala with sketchy internet, trying to debug why your client has duplicate everything in their CRM.
Santi: Don't have to imagine it.
Kira: System Three: Project and Client Ops. This is the one that actually runs your business.
Santi: And the one most people underestimate. They think, "Oh, I'll just use Notion and check it manually." Then they're in Bali, it's midnight, and they forgot to assign that urgent task to their VA.
Kira: So here's what we do. Every project management tool — Notion, Asana, Monday, whatever — becomes a trigger source. New task created? Automation assigns it based on capacity. Task overdue? Automation escalates. Client comments? Automation notifies the right person.
Santi: But here's the thing — you need a dead-letter queue.
Kira: Explain that.
Santi: It's a pattern from AWS but you can implement it anywhere. When an automation fails — and they will fail — instead of just losing that task assignment, it goes to a holding queue. Then you have a separate process that monitors the queue and either retries or alerts you.
Kira: We simulate this in Make using a data store. Failed operations write to the store, and we have a scheduled scenario that checks it every hour.
Santi: In my setup, if the dead-letter queue has more than five items, I get a Telegram message immediately. Even if I'm on a boat in the Gilis with no laptop.
Kira: System Four: Knowledge and Docs. This one's deceptively simple but it's actually the foundation for everything else.
Santi: Every SOP, every template, every prompt — centralized in one place. We use Notion but honestly, any system with an API works.
Kira: The automation angle here is AI-assisted documentation. Every time we finish a client project, we have a Make scenario that pulls the project data, feeds it to Claude or GPT-4, and generates a case study draft.
Santi: But — and I cannot stress this enough — you need version control and access logs. When your VA in the Philippines updates an SOP while you're asleep in Mexico, you need to know what changed.
Kira: Also, watch the Notion API limits. Payload size, rate limits — they're stricter than you think.
Santi: System Five: Content Engine. This is where the unit economics get really interesting.
Kira: Here we go with the math.
Santi: No, but seriously. Let's say you're running a weekly newsletter. Pull content from five sources, aggregate, summarize, personalize for three segments, schedule sending. In Zapier, if each source is a search action, that's five tasks. Aggregation, one task. Three personalizations, three tasks. Scheduling, one task. Ten tasks per run, fifty-two weeks, that's five hundred twenty tasks just for one newsletter workflow.
Kira: Versus?
Santi: Make would be similar in credits, maybe slightly less if you use arrays efficiently. But n8n? One execution. Pipedream? Depends on compute time, but probably under five seconds, so less than one credit.
Kira: This is why you have to model your specific workflows. The calculator we're including has fields for all of this — steps per workflow, runs per month, error rate.
Santi: System Six: Finance Ops. This is the one you cannot get wrong.
Kira: Cannot.
Santi: Every payment, every invoice, every subscription renewal — needs idempotency keys. You accidentally charge a client twice because your automation retried? That's a relationship ender.
Kira: We use Stripe's idempotency pattern everywhere, even for non-Stripe operations. Generate a unique key for every financial operation, store it, check it before processing.
Santi: And audit trails. Every financial automation writes to a log that's separate from your main database. When not if, but when something goes wrong, you can trace exactly what happened.
Kira: System Seven: Monitoring and Alerts. This is what would have saved your Thailand situation.
Santi: Yeah. So here's the checklist. Every external API you depend on — subscribe to their status page. Every critical automation — set up heartbeat monitoring. If it hasn't run in its expected window, you get alerted. Every error — goes to the dead-letter queue we talked about. And every queue — has a threshold that triggers an alert.
Kira: We use Better Uptime for the heartbeats, but honestly, even a simple Zapier that expects a webhook every day works.
Santi: The key is redundancy. Email alerts are useless if you're not checking email. We use Telegram, SMS, and push notifications. At least one will get through.
Kira: Also — and this is something we learned the hard way — monitor your monitoring. We have a weekly automation that tests all our alert channels. Sends a test message to each one, confirms delivery.
Santi: Monitoring the monitoring that monitors the automations. It's turtles all the way down.
Kira: Okay, let's talk about agents. Because every platform is pushing these now, and people are adding them everywhere without thinking about reliability.
Santi: Here's my position: agents are for judgment calls, not for deterministic operations. If you can write a rule for it, don't use an agent.
Kira: But there are places where they make sense. Support ticket classification, lead qualification, content summarization — anywhere you need human-like judgment.
Santi: Right, but with constraints. Short contexts, specific tools, clear boundaries. Zapier Agents are good for this — they're sandboxed, they have activity limits, they integrate with your existing Zaps.
Kira: Make's new AI Agents are interesting because you can see the reasoning. Like, it actually shows you why it made each decision.
Santi: Which is critical for debugging. N8n's AI Agent node is similar — full visibility into the chain of thought.
Kira: But here's what actually happens when you're on the ground — agents fail in ways that are hard to predict. They'll work perfectly for weeks, then suddenly start hallucinating because something in the context changed.
Santi: So you need human-in-the-loop for anything critical. Agent makes a classification? Human reviews it. Agent drafts a response? Human approves it.
Kira: At least until you have enough data to trust the error rate.
Santi: Let me give you real numbers from our setups. My consultancy runs about two hundred automations across Make and n8n. Monthly cost: around three hundred dollars. That includes ten thousand Make credits, five thousand n8n executions, and various API costs.
Kira: My agency — thirty-one-k MRR — we're heavier on Zapier because of client requirements. About five hundred a month for twenty thousand tasks, plus another two hundred for auxiliary tools.
Santi: The key is knowing when to upgrade. Under ten-k MRR? Stick with free tiers and basic plans. Ten to twenty-five-k? Invest in reliability — proper databases, monitoring, redundancy. Over twenty-five-k? You can afford the premium tiers that include better support and higher limits.
Kira: Also, some quick wins from real implementations. Slate Magazine — they used Zapier Agents to generate two thousand qualified leads in a month. Automated the enrichment and qualification that would have taken a human weeks.
Santi: Laudable — team of four — saved two hundred forty thousand in equivalent salaries by automating their entire onboarding flow. Two hundred automations, but each one simple and reliable.
Kira: System AI cut a task from four minutes to twenty seconds using n8n. Not fancy, just well-designed workflows that handle the repetitive stuff.
Santi: Alright, the counterargument. Because I know what some of you are thinking.
Kira: Which is?
Santi: "This is overengineered. I don't need seven systems and monitoring and dead-letter queues. I just need something that works."
Kira: Fair point, actually.
Santi: And for some businesses, that's true. If you're solo, doing five-k MRR, selling info products — you probably don't need all of this. But the moment you have clients depending on your automations, or team members in different time zones, or any kind of operational complexity—
Kira: You need resilience.
Santi: Exactly. It's not about the tools, it's about the patterns. Idempotency, retries, monitoring — these work whether you're using Zapier or writing custom code.
Kira: And here's the thing — you can implement these gradually. Start with the basics, add monitoring when something breaks, add retries when something fails repeatedly. You don't have to build everything at once.
Santi: But you do have to build it eventually. Because the alternative is losing eight-thousand-dollar clients while you're trying to enjoy a beach in Thailand.
Kira: So here's your thirty-day challenge. Map your current stack against these seven systems. Identify the gaps. Pick one gap — probably monitoring — and fix it this week.
Santi: Use the calculator to model your costs across platforms. You might find you're overpaying by fifty percent just because you picked the wrong billing model.
Kira: Document everything in the template we're providing. Seven systems, current tools, upgrade triggers, monitoring setup.
Santi: And here's the accountability piece — screenshot your stack diagram, post it on LinkedIn or Twitter, tag the show. We'll review the most interesting ones on next week's episode.
Kira: Plus, if you find a creative solution to any of these problems — a monitoring setup we haven't seen, a retry pattern that works better — share it. This community is small enough that your solution could help dozens of other nomad founders.
Kira: So that's the stack. Seven systems, each with its own economics, each with its own failure modes, and — most importantly — each with patterns that work whether you're at ten-k MRR or a hundred-k.
Santi: The thing that would have saved me in Thailand wasn't a better automation platform. It was understanding that automations fail, APIs change, and if you're going to run a business from the other side of the world, you need systems that can handle failure without you.
Kira: And that's really what this is about. Not the tools — the patterns. Idempotency, retries, monitoring, dead-letter queues. These aren't fancy concepts. They're just the difference between a business that runs itself and one that breaks the moment you lose wifi.
Santi: Look, start with the calculator. Model your actual workflows. Count your steps, your runs, your error rate. You might discover you're using the most expensive possible platform for your specific use case.
Kira: Then pick one system — probably monitoring — and make it bulletproof this week. Just one. Next week, another one.
Santi: And share what you build. Screenshot your stack, tag us, show the community what's working. Because someone else is dealing with the exact same problem from a coworking space in Lisbon or Canggu or Mexico City.
Kira: I'm Kira.
Santi: I'm Santi.
Kira: Build smart, travel light, and automate everything that doesn't need your genius.
Santi: See you next week when we break down how to price AI services when your costs are completely unpredictable.