AI Workflow Builder

What you will be able to do: The learner can evaluate any repetitive business task against three concrete criteria and correctly sort it into "automate fully," "automate with a human checkpoint," or "leave alone."

Most automation projects fail before the first Zap or scenario is built. Not because the tool was wrong, but because someone automated the wrong task — usually one that felt tedious but actually needed human judgment every time it ran. If you sell automation as a service, this is the single most expensive mistake you can make: you deliver something that works in the demo and breaks in week two, and the client stops trusting you.

This lesson gives you a fast filter to run before you touch any tool. It applies whether you're using Zapier, Make, n8n, or writing custom code.

The three questions that decide automatability

Before proposing an automation, ask the business owner (or yourself, if it's your own process) these three things:

1. Does it happen often, in roughly the same shape? A task that runs 50 times a month with the same 4-5 variations is a candidate. A task that happens twice a year and is different every time is not — you'll spend more time building the automation than the task would ever cost manually.
2. Can you write the exact rules for it, right now, on paper? If the person doing the task can describe every decision as "if this, then that," it's automatable. If they say "I just know" or "it depends on the client," that's a judgment call, not a rule, and no workflow tool can replicate it reliably.
3. If the automation gets it wrong, is the mistake cheap and easy to catch? A miscategorized email is cheap — someone re-reads it. A wrongly issued refund, a wrong shipping address, or a wrong price quote sent to a customer is expensive and often invisible until the client complains. High-volume, low-stakes-per-error tasks are ideal. Low-volume, high-stakes tasks need a human checkpoint even if steps 1 and 2 look automatable.

Worked example: an online retailer in Nairobi

Say you're building automations for a small e-commerce business selling through Instagram and a Shopify store. They hand you a list of complaints about their operations. Two of them:

• Order confirmation and delivery-status updates. Every order triggers the same three messages: "order received," "shipped," "out for delivery." The content is templated, the trigger (order status change in Shopify) is unambiguous, and if a message goes out a few minutes late, nobody is harmed. This scores well on all three questions — high volume, clear rules, cheap errors. Build it with Shopify's order webhook feeding into Make or Zapier, which sends WhatsApp or SMS via Twilio or a service like Africa's Talking. This is a same-week win.
• Handling customer complaints about damaged goods. Every case is different: the product, the customer's tone, whether a refund, replacement, or partial credit is fair, whether the damage was the courier's fault. There's no clean rule set, volume is lower, and getting it wrong (wrongly denying a legitimate refund, or approving a fraudulent one) costs real money and reputation. This fails question 2 and 3. The right move isn't to automate the decision — it's to automate the intake (log the complaint, attach photos, notify the right staff member) and leave the judgment to a person. That's a "human-in-the-loop" design, not full automation, and you should sell it as such.

Notice the pattern: the same business can have one task that's a perfect automation candidate sitting right next to one that will damage your reputation if you automate it fully. Your job is to tell them apart before you quote a price.

Tools and what they actually cost

You don't need to pick a tool yet — that's a later lesson — but it helps to know the free-tier ceilings so you can scope realistically:

• Zapier has a free plan capped around 100 tasks a month, and the free tier restricts you to single-step automations — anything with a filter, a branch, or multiple actions needs a paid plan.
• Make (formerly Integromat) gives a free tier with roughly 1,000 operations a month and allows multi-step scenarios even on the free plan, which makes it more forgiving for testing before you sell a paid build.
• n8n is open-source and free to self-host if you're comfortable running it on a server; it also offers a paid cloud version if the client wants something you don't have to maintain yourself.

None of this matters if you've picked the wrong task. A perfectly executed automation of a low-volume, judgment-heavy process is still a bad automation — it just fails more elegantly.

What this unlocks

Once you can sort tasks this way in a client conversation, the next lesson shows you how to run that conversation itself — how to interview a business owner to surface the repetitive tasks they've stopped noticing, because to them it's "just how we do things."

Key points

• Good automation candidates are high-volume, rule-describable, and cheap to get wrong; if any of those is missing, don't fully automate.
• If the person doing the task can't write down the exact rules, the task requires judgment — automate the intake, not the decision.
• Free tiers (Zapier ~100 tasks/month single-step only, Make ~1,000 ops/month multi-step) are fine for testing whether a task is a good fit before you commit to a paid build.

Practice

Pick a real business (yours, a friend's, or a prospective client's) and list 5 recurring tasks they do manually. For each, score it 1-3 on: frequency, rule-clarity, and cheapness-of-error. Done = you have a table of 5 tasks with 3 scores each, and you can name the single best candidate and explain why in one sentence, plus one task you'd explicitly refuse to automate and why.