Map the workflow first. Then automate it.

Step 1 — Map the current workflow

Before you open Make, Zapier, or a single AI prompt, draw the workflow exactly as it runs today. Not how it should run — how it actually does. Every trigger, every action, every handoff between people, and every decision point. If you cannot draw it on one page, you do not understand it well enough to automate it.

For each step, capture four things: what kicks it off (the trigger), what gets done (the action), who or what owns it (the handoff), and where someone has to make a call (the decision). Decisions are where automation either earns trust or loses it, so mark them clearly.

This map is the spec for everything that follows. Don't automate a broken foundation — if the workflow has redundant steps or unclear ownership, fix that on the map first. Automating mess just makes the mess run faster.

Step 2 — Find the automation candidates

Now go through the map and sort each step into one of two buckets: machine work or human judgement. You are not trying to automate the whole workflow. You are trying to automate the parts that are draining your team's hours without needing their brains.

Good automation candidates share three traits — repetitive, rule-based, and high-volume:

• Repetitive: the same steps happen the same way, many times a week
• Rule-based: the logic can be written as clear if-this-then-that conditions
• High-volume: enough throughput that automating it actually returns meaningful time

Leave the rest for people. Anything that needs taste, negotiation, relationship context, or a judgement call on ambiguous data stays human-in-the-loop. Brand voice on a high-stakes email, approving a budget shift, deciding whether a borderline lead is worth a personal call — those belong to a person, sometimes with an AI draft in front of them, never replaced by one.

Step 3 — Choose the layer and the tools

Pick the right tool per step, not one tool for the whole workflow. The most maintainable automations use the simplest layer that does the job and only reach for complexity where it is genuinely needed.

Native features first

If the platform you already use can do it — a CRM automation rule, a scheduled report, a built-in sequence — use that. It is free, it is supported, and it is one less integration to maintain.

Scripts for platform-specific logic

When native rules run out of room, a script (Google Ads Scripts, an Apps Script, a small API job) handles custom logic against one platform's data on a schedule.

Make, Zapier, or n8n for connecting systems

When a step crosses tools — form to CRM to Slack to email — a workflow automation platform is the right layer. Make and n8n give you branching and error handling; Zapier is fast for simpler linear flows.

AI agents for the judgement-shaped steps

Where a step needs to read unstructured input and draft a response — summarise a call, classify an inbound message, write a first-pass email — an AI agent fits, kept inside guardrails and reviewed by a person. Choosing well across all four layers is a skill in itself; see our team tooling guide for how to match tools to steps.

Step 4 — Add guardrails and logging from day one

Guardrails are not a phase-two add-on. They go in with the first automation you build. A guardrail is a rule that constrains what the automation is allowed to do, and a log is the record of what it actually did. Without both, you have an automation you cannot trust and cannot debug.

• Rate limits: cap how many records, messages, or changes the automation can touch in a run
• Approval gates: hold high-stakes actions for a human to confirm before they fire
• Decision logs: record the trigger, the action taken, and the outcome for every run, so any decision is auditable
• Fail-safe defaults: when an input looks wrong, the automation stops and flags it rather than guessing

This is the heart of guardrail-driven automation, and it is covered in depth in our governance guide. The other half of reliable automation is clean inputs: an automation is only as good as the data feeding it. Standardise your inputs — consistent field names, formats, and naming conventions — so the workflow never has to guess. Our standardization guide walks through how.

Step 5 — Build, test, measure, and hand over

Now build it — against the map, one step at a time. Test each step in isolation before chaining them, and run the whole workflow on a small batch of real data before you let it touch everything. Watch the logs on the first live runs the way you would watch a new hire's first week.

Then measure against the baseline you established when you mapped the workflow: hours returned to the team, error rate, throughput, and how often a human had to step in. Those numbers tell you whether the automation earned its place — and where the next candidate is.

Finally, hand it over. The point is that your team owns the system. If your team wants to build and maintain these workflows themselves, our Training covers Make, Zapier, and n8n workflow design hands-on. If you would rather have it built done-with-you, with your team owning the output at the end, that is what an Automation Sprint is for.

Automation Sprints

In two to four weeks we map your workflow, build the automation with guardrails and logging, and hand you a system your team owns outright — no lock-in.

Where to start

You can run this five-step method on any workflow yourself starting today. If you want a clear read on which dimensions of your operation are ready to automate first — democratization, instrumentation, standardization, governance, and tooling — the free Readiness Score scores you across all five in a few minutes and points you at the highest-leverage place to begin.