AI Demand Is Automation Demand

Zapier recently analyzed the top 25% of companies in a panel of 1,500 mid-market and enterprise-level companies, ranked by AI workflow adoption.

The findings were pretty eye-opening, but before we dive in, let's play a quick guessing game: what percentage of those companies' workflow steps do you think were represented by AI?

Would you be surprised to learn it was only 18%?

What about the other 82%? That was comprised of conventional automation: rules, logic, app connections, and moving data.

In a world where it increasingly feels like everyone's telling you the answer to everything is to "build an agent" and "leverage AI more," that number might seem surprising.

But here's the thing: the demand for AI is typically just demand for automation.

When most business owners say they want AI to "just do the work," what they're describing is actually a system that runs automatically and reliably, every time. If "X" happens, do "Y," no judgment or LLM required.

That's not AI, that's automation. Knowing the difference is one of the most underrated skills in this space right now, and it was one of the major recurring themes echoed by experts across dozens of sessions earlier this week at ZapConnect, Zapier's annual virtual conference.

So when do you use which?

The simplest way I've found to think about it: automation handles certainty; AI handles ambiguity.

If the logic is always the same, use automation. If the input varies and requires judgment, that's where AI earns its place.

A concrete example I use with clients all the time: lead routing. If a lead comes in interested in Service A, they always go to Rep 1, whereas Service B always goes to Rep 2. That's deterministic logic: the answer is always the same based on a known input. Build that with automation. It'll be faster, cheaper, and more reliable.

But now say those same leads are coming in through a contact form with an open text field, so every submission is different. Someone writes three sentences about their business challenges, and you need to figure out which service they actually need and who's best equipped to handle them. That input is messy and nuanced. That's where you bring in AI to read it, analyze it, and make a smart routing decision based on context. The AI Workflow Index actually has a name for this pattern: the Analyst role, where AI makes a judgment call that a downstream step then acts on.

Client onboarding is another good one. A lot of people assume it now needs AI for personalization and communication. But unless you're explicitly using AI to analyze something or prepare something personal for a specific client, onboarding rarely needs it. If a new client gets added to your project management system, receives a welcome email from a template, and gets sent a document to sign, all of that is a repeatable sequence. Automate it and save your AI budget for something that actually requires judgment.

NPS and feedback routing works the same way. Promoters get a referral ask, while detractors get flagged for follow-up, and your team gets a Slack ping either way. No AI needed; just a score, a threshold, and a trigger.

Where the two work together

The most powerful builds combine both. Take meeting prep, for example. Pulling a name from your calendar, finding the account record, and retrieving recent emails don't need AI. But synthesizing all of that into a pre-call brief that actually reads like a human wrote it? That's the AI step. One judgment-requiring moment inside an otherwise fully automated workflow.

That's the pattern worth internalizing: AI does the thinking; automation does the doing.

Another one I often build for clients is lead research and enrichment. A new contact fills out a form, which triggers an AI agent to pull company data, surface relevant context, and deliver a pre-call brief to the sales rep, all before anyone picks up the phone. The trigger and delivery? Pure automation. The research and synthesis? That's AI doing what AI is actually good at.

The question to ask before you build anything

Does this step require judgment, or does it just need to happen the same way every time?

If it's the latter, you don't need AI. You need a trigger and an action. Something that runs quietly in the background and never makes you think about it again.

If it requires judgment, such as reading context, handling ambiguity, or making a decision based on nuanced input, that's your AI moment. Use it there, and only there.

The businesses getting the most out of AI right now aren't the ones using it everywhere; they're the ones who've figured out exactly where it belongs.

The report also found that workflows reserving AI only for reasoning steps cost 71% less to run than workflows routing every step through a model. That's not a minor rounding error; that's the difference between a system that scales and one that drains your budget.

Before we wrap

Zapier recently released the full AI Workflow Index and it's worth a read, especially if you're trying to make sense of where AI actually fits in your workflows versus where automation does the job better. You can download it here: https://zapier.com/ai/workflow-index/q2-2026-report

What's a task in your business that you've been assuming needs AI? Drop it in the comments. I'd be willing to bet there's a simpler answer than you think.

And as always, if you want a second set of eyes on your workflows, that's exactly what I'm here for: processpowerup.co/schedule-a-call

See you next week!

— Andrew