Do you need AI, or just automation? How to decide before you pay for it

I’ve used AI every day since ChatGPT launched in 2022 and built nearly 100 AI tools, most of them for other businesses. One of the most useful things I’ve learned is when not to use it.

AI is powerful, but it isn’t free. Most AI tools charge by usage, so every message, image or document they process costs a little money. On a busy system those costs add up. And for some jobs, AI is simply the wrong tool: slower, less predictable and more expensive than a plain automation that does the job perfectly.

Here’s how I decide.

Automation vs AI: what’s the difference?

Automation follows rules you define. “When a form is submitted, add it to the CRM and send this email.” It does exactly the same thing every time, instantly, and usually costs next to nothing to run.

AI makes judgment calls. It reads messy information, understands what someone means, writes something new, or decides what should happen next. That flexibility is what you pay for.

A lot of what gets sold as “AI” is really automation, and that’s fine. The goal is the cheapest, most reliable tool that solves the problem.

You probably don’t need AI when…

• The rules are clear. If you can write the logic as “if this, then that,” plain automation will do it faster, more reliably and for a fraction of the cost.
• The data is already structured. Moving an order from a form into your inventory system doesn’t need understanding. It needs a connection.
• There’s one correct answer. Calculating a price, a tax, or the shortest route between stops is maths, not judgment.
• You need the exact same output every time. Invoices, confirmations and legal notices should be predictable. AI is designed to vary.
• It runs thousands of times a day. Small per-use AI costs multiply fast at high volume.

Real example: we built a pickup booking and route planning portal for Habitat for Humanity. The route optimization uses a routing engine, not AI, because finding the shortest driving order is a well-solved problem with an exact answer. It’s faster, free to run and never “gets creative.” We also built a dealer quoting portal with no AI at all. The win came from cleaning up the workflow so quotes became orders without retyping.

AI is worth it when…

• The input is messy or unstructured. Emails, PDFs, free-text messages, photos and conversations all need understanding before anything can happen.
• The task needs reasoning. Weighing several pieces of information, spotting what doesn’t add up, or deciding what to do next.
• The output must be written or created. Replies in your voice, summaries, reports, descriptions and images.
• The situation varies every time. Customer questions, edge cases and anything you couldn’t fully write rules for.
• A person is currently doing it by reading and thinking. That’s the work AI replaces best.

Real example: for Coins For Anything, the Coin Nerd is a Claude assistant that reads email threads, cross-checks payments against the books, and decides what each order needs next. No set of rules could cover every customer email or messy situation, so reasoning is the whole point, and it saves the team far more than it costs.

Their AI Coin Builder generates a custom coin image from a description. Every image costs money to create, so we built in limits: per-visitor caps, a daily site-wide budget with alerts, and free repeats of the same design. AI was the right tool, but only with guardrails.

A quick decision checklist

Ask these five questions about any task you want to automate:

1. Could I write down the exact rules? If yes, start with plain automation.
2. Does it need to understand language, images or messy data? If yes, AI is likely worth it.
3. Does it need judgment or reasoning? If yes, AI. If it’s a calculation, no.
4. How often will it run? High volume means usage costs matter. Estimate them before you build.
5. What happens if it’s wrong? If a mistake is costly, add human review, or don’t use AI for that step.

Most good systems end up as a mix: plain automation for the predictable steps, with AI used only for the parts that need understanding or reasoning. That keeps costs low and reliability high.

How to keep AI costs under control

When AI is the right call, a few habits keep the bill predictable:

• Use AI only for the step that needs it. Let ordinary automation do the routing, saving and sending around it.
• Pick the right model size. Smaller, cheaper models handle simple sorting and extraction well. Save the most capable models for real reasoning.
• Set limits and alerts. Daily budgets, per-user caps and spending alerts stop surprises.
• Reuse results. If the same question or input comes up again, reuse the earlier answer instead of paying twice.
• Run on a schedule when you can. A summary three times a day often works as well as reacting to every single change, at a fraction of the cost.

The bottom line

Don’t start with “how can we use AI?” Start with “what’s slowing us down?” Then use the simplest tool that fixes it. Sometimes that’s AI. Often it’s a clean automation. The best systems use both, in the right places.

That’s exactly what our AI Business Audit is for: we map your workflows and tell you honestly which parts need AI, which just need automation, and what each will cost to run. Book one here.