Choosing an automation platform

0:00 Choosing an automation platform

Zapier, Make, n8n, Power Automate, or the automation already built into your CRM? People argue about this online as if there's one right answer. There isn't. There's a right answer for your apps, your volumes, your data and your team. In this lesson you'll compare the main platforms available to small businesses, learn six decision criteria, see how three different businesses chose, and run a one-hour bake-off that settles it with evidence.

0:31 Analogy: choosing a car

Here's an analogy. Choosing an automation platform is like choosing a car. A city runabout, a family estate and a pick-up truck can all get you to work. The right one depends on what you carry, how far you drive, who else drives it, and whether you're happy doing your own servicing. Nobody sensible buys the car with the longest feature list without a test drive.

1:00 The landscape

Here's the landscape. Zapier has a huge app catalogue, is easy for non-technical teams, and now includes AI steps, agents and a way to connect AI assistants through MCP. It charges per task, meaning each successful action step. Make offers visual scenarios with powerful branching and data handling, plus AI modules and agents, and since August twenty twenty-five it bills in credits, where AI modules can use more than one credit per run. n8n is developer-friendly, with AI agent, classifier and extractor nodes, MCP support and human review for AI tool calls, and it can be self-hosted. Power Automate, with Copilot Studio for agents, shines inside Microsoft three six five. And built-in automations in your CRM or help desk are the simplest of all, but limited to that tool.

1:56 Six criteria

Now the six criteria. Integrations: are your core apps supported with the exact triggers and actions you need, plus a generic web request step for the rest? AI capabilities: AI steps with structured output, agent building, choice of model provider, and MCP support. Cost at your volume. Data and compliance: where data is processed and stored, and how long execution logs, which often contain customer data, are kept. Team and ownership: business accounts, roles and version history. And reliability: error handling, retries, alerts, logs and replaying failed runs.

2:34 Do the arithmetic

Cost deserves a closer look, because pricing units differ so much. Take a six-step workflow that runs three thousand times a month. On a per-step model, that's around eighteen thousand units. On a per-execution model, it's three thousand. AI steps may cost extra on top. Neither is always cheaper. It depends on how many steps your flows have and how often they run. So estimate your runs and steps, read each vendor's current pricing page, and do the arithmetic before you commit.

3:10 Simple example: salon bookings

A simple example of the arithmetic. A salon sends a booking confirmation, adds the client to a sheet and posts in team chat: three steps, about four hundred bookings a month. On a per-step model, that's around twelve hundred units a month. On a per-run model, four hundred. Either might fit a cheap plan. Now imagine eight steps and five thousand runs: the difference becomes the deciding factor.

3:40 Three choices

Here's how three businesses chose. A solo creator in Lahore, handling brand enquiries and repurposing, picks Zapier: every app she uses is supported, volume is low, and she needs no code. A twelve-person performance agency in Dubai, with complex client reporting and lots of branching, picks Make for its visual data handling, owned by an operations lead. And a UK e-commerce brand with a developer and strict data requirements picks self-hosted n8n in its own cloud account in a UK region, so execution logs never leave its environment, using the AI agent node with human review on any tool that emails customers.

4:24 Business example (illustrative)

A deeper business example, illustrative. The Dubai performance agency estimated its twelve automations at about forty thousand step executions a month. On a per-step plan that meant a high tier; on a credit-based plan with efficient scenarios it fitted a mid tier; self-hosted, it would cost a small server plus maintenance time they didn't have. They chose the credit-based platform, and a year later their bill was within ten percent of the estimate.

4:56 Self-hosting trade-off

Self-hosting deserves a warning. It gives you data control and can be cost-effective at volume, but you own updates, backups, security patches and uptime. If nobody on your team is comfortable with that, a managed cloud plan, possibly with regional hosting, is usually the safer choice. And whatever you choose, make sure the business, not an individual's personal account, owns the workflows and credentials.

5:24 Hands-on: one-hour bake-off

In the hands-on section, you'll run a one-hour bake-off. Build the same small workflow in your top two platforms: a new row in a leads sheet, an AI step that classifies the message and returns JSON, a team-chat alert for new clients, a log back to the sheet, and one deliberate failure with bad JSON to test error handling. Score each on time to a working version, clarity, AI step quality, error handling, execution logs, cost at your real volume and account ownership.

6:00 Common mistakes

Common mistakes. Choosing the platform your favourite influencer uses. Ignoring where execution logs are stored. Building business-critical flows in a personal free account. Self-hosting without anyone to maintain it. Forgetting to check whether the platform supports your model provider or your own API key. And committing to an annual plan before a bake-off.

6:23 How you'll know you chose well

How will you know you chose well? Six months later, your team can build and change workflows without calling the one person who set it up. Your monthly bill matches your forecast. Failures are visible and easy to replay. And you haven't needed a workaround for any of your core apps. If you have, note it at renewal and rerun a bake-off.

6:50 Watch me do it: the bake-off

Watch me do it. I run the bake-off in the first platform. Step one, a trigger on a new row in my Leads sheet. Step two, an AI step with the classification prompt, returning JSON. Step three, a filter for new clients. Step four, a chat message. Step five, write the category back to the sheet. Then I paste a lead whose message produces invalid JSON on purpose. I check: did the error route catch it and alert me? It did. I time it: forty-two minutes. I open the execution log and see every step's input and output, including the customer's message, which reminds me to check how long logs are kept. I note the log retention setting. Then I repeat the build in the second platform. It took fifty-five minutes, the error wasn't caught until I added an error handler, and pricing at our volume was lower. I score both on the sheet and total them.

7:58 Recap

To recap: there's no universal best platform. Choose on integrations, AI capabilities, cost at your volume, data and compliance, ownership and reliability. Do the pricing arithmetic, think carefully before self-hosting, and let a bake-off decide. Your next step: estimate monthly runs for three planned automations, price them on two platforms, and run the bake-off. Next, you'll build your first complete AI workflow, step by step.

8:26 Try this now

Try this now. Estimate monthly runs and steps for your top three planned automations and calculate the cost on your two shortlisted platforms using their current pricing pages. Then book two one-hour slots this week for the bake-off, build the same small workflow in each, including the deliberate bad-JSON failure, and score them. Decide by Friday.