Choosing the Platform: The Decision That Outlives Every Automation

You’ve got your ranked list of what to automate. Before building anything, one decision deserves real diligence, because it outlives every automation you’ll ever build: what platform does it all run on?

This is the least glamorous choice in the whole journey and one of the most consequential. Pick well and every future build gets cheaper. Pick badly and you’ll discover it eighteen months in, when migrating twenty working automations off the wrong platform costs more than building them did.

The four categories you’re choosing between

Point tools. The “AI for X” apps: an AI collections app, an AI scheduling app. Fastest to start, and a trap at scale, because ten point tools means ten subscriptions, ten data silos, and zero shared wiring. You rebuild the doorways for every one.

App-native automation. The automation built into software you already own: Power Automate, your CRM’s workflow builder, and the like. Fine for simple triggers. They strain when logic gets real: multi-step decisions, custom business rules, and the human approval gates we build into everything.

A workflow engine. A dedicated orchestration layer that connects to everything through APIs and MCP, holds your business logic in one place, and treats AI as a component you call, not a product you’re locked into. This is the category we run our own business on (our engine of choice is n8n), and the one we deploy for clients.

Custom code. Maximum control, and now you own a software product: hosting, updates, the developer who wrote it. Right for edge cases. Wrong as a default for a business whose product isn’t software.

Six criteria that actually matter

Connectivity breadth. Can it reach your specific systems, your accounting, your PSA or job software, Microsoft 365, through APIs and MCP? A platform that covers nine of your ten systems leaves your automations blind in one eye.

Pricing model. This is the trap. Per-task and per-execution pricing punishes success: the more your automations run, the more you pay, forever, and a busy automation can quietly cost more than the labor it replaced. Flat or capacity-based pricing keeps the ROI math you did in identification true at scale.

Model independence. AI models leapfrog each other every few months. A platform welded to one vendor’s model rides that vendor’s fortunes. You want to swap the AI component the way you’d swap a part, because the best model for the job next year probably doesn’t exist yet.

Human-in-the-loop primitives. Every automation worth trusting has approval gates. If the platform can’t route a decision to a person in Teams or email and wait for the tap, you’ll bolt that on badly or skip it dangerously.

Data control. Where does it run and who can see what flows through it? For firms with client confidentiality or compliance exposure, legal, healthcare, financial services, a self-hostable option that keeps data on infrastructure you control isn’t a preference, it’s a requirement.

Exit cost. Can you export your workflows in a readable format? Ask this on day one, because the platforms that make leaving hard are the ones counting on it.

What we chose: n8n

We run our automations on n8n’s cloud platform: model-agnostic, with every workflow exportable, and self-hostable if a client’s data ever needs to stay on their own infrastructure. We got here the useful way: by evaluating the wrong options first on our own dime, so our clients don’t repeat the tour.

That’s the platform layer of our AI Managed Services: we build on it, manage it, and stand behind it, and your automations are portable if we ever part ways. Confidence in your own exit costs is a feature worth noticing in any vendor.

Platform chosen, list ranked: now the fun part. The next three posts are implementations, with real numbers, from our own books. Or jump straight to the AI Business Assessment. Early Access, limited spots per quarter.