Not Everything Needs AI. Here's How to Tell.
Every week I talk to small business owners who want to implement AI. And the first question I always ask is the same:
Why?
Not to be difficult, and not to make them feel stupid, but because "implement AI" isn't actually a goal. It's a solution in search of a problem.
More often than not, when we slow down and get specific about what they're actually trying to solve, one of two things happens: either we find that a simple, well-built automation would do the job best, or we find that the foundation isn't solid enough for AI to add value yet anyway.
The rush to integrate AI is real urge these days, but let me walk through why rushing is exactly the wrong approach.
Two types of logic, and why it matters
Before you can make a good decision about when to use AI and when not to, it helps to understand one key distinction: deterministic versus non-deterministic logic.
Deterministic means predictable: same input, same output, every time. No judgment, no interpretation, and most importantly, no surprises. Think of it like a decision tree: if your lead selects Service A on your website form, they get routed to Person X. If they select Service B, they get routed to Person Y. Clear, reliable, repeatable. This is what traditional automation does, and it does it exceptionally well.
Non-deterministic means there's room for judgment. The same input might produce a different output depending on context, nuance, or interpretation. This is where AI not only lives but can thrive. When you're drafting a reply, analyzing a customer's feedback, or researching a lead based on limited information, you want that flexibility. You need something that can think, not just execute.
The mistake I see most right now is people reaching for AI by default, even when a deterministic approach would be faster, cheaper, and more reliable.
When regular automation is the right answer
Here's a real example. A client wanted to use AI to route incoming leads to the right salesperson based on what they'd written in a free-text form field. The thinking was: AI can read the submission and figure out who to send it to.
But the better answer was simpler. We updated the form to include a simple dropdown (Which service are you interested in?) and built a straightforward Zapier automation that routed based on that selection, every time. No judgment required, and no risk of a lead ending up with the wrong person because the AI misread or misinterpreted something.
When done right, deterministic logic is airtight. For anything that has to work the same way every single time (like routing, record creation, notifications, and data syncing) it's almost always the right choice.
When AI actually earns its place
Now here's the flip side. That same lead, once routed correctly, still needs to be researched before the sales call. Someone has to go find out who this person is, what company they work for, what their background looks like, and whether they seem like a strong fit. That's 15 to 30 minutes of manual work per lead.
This is where AI shines. An AI agent can take the lead's name, email, and company, search the web, compile a dossier, assign a confidence score, and drop it into the CRM automatically, before anyone picks up the phone. Evaluating what's relevant, synthesizing disparate information, and making a confidence call are the kinds of judgment AI is built for.
Another great use case is post-project feedback. If a client submits a long written response about how an engagement went, AI can read it, identify the sentiment, extract the key themes, and output a clean summary with a simple rating that your team can act on. AI shines with that type of judgment and analysis.
The foundation problem
Here's what I see most often when a client comes to me wanting to "do more AI": their processes aren't ready for it yet.
AI is a supercharger. It can absolutely take your team and your workflows to another level, but if what you're supercharging is inconsistent data, unclear ownership, and undocumented processes, it's only going to amplify the mess. You'll be moving faster in the wrong direction.
Think of it like adding another floor to a house. It sounds exciting, but if the foundation and the first floor haven't been properly built yet, adding more on top isn't going to end well.
Before AI, you need the basics:
Clean, consistent data
A clear understanding of who does what, when, and why
Tools that are actually being used the way they're supposed to be
Documented processes that people trust.
The things I've been writing about in this newsletter, like data hygiene, tech stack hygiene, process documentation, etc., aren't prerequisites for some imaginary future state. They're prerequisites for AI to work the way you're imagining it will.
It's also important to note: you don't need to be an AI expert to navigate any of this. You don't need to know every model, every platform, or every capability. Automation tools like Zapier make it genuinely accessible to build both deterministic automations and AI-powered workflows, without writing a single line of code
So what should you do first?
Get the lay of the land. Internally, with your team, or with someone who can help guide you through it. Get clear on your processes, your tools, and how everything fits together. You don't need it to be perfect; you just need to actually understand it.
Once you do, start looking at everything through the lens of that deterministic versus non-deterministic question. What's a clear if-this-then-that? Build that with automation; it'll be faster, cheaper, and more reliable.
What actually requires judgment, nuance, or interpretation? That's where AI can fit in.
Before we wrap
Have you added AI to any of your workflows in the last year? And if so, did you start with the foundation, or did you start with the AI? Drop it in the comments. I'd love to hear what you're working with.
If you want help getting the lay of the land before you start building, that's exactly what I do. Book a free discovery call at processpowerup.co/schedule-a-call.
See you next week!
— Andrew