Jake Makes AI
Vendor Math

Which Part Is the AI?

A lot of what gets sold as artificial intelligence is a rule a spreadsheet could run. Make the vendor point at the box.

A chrome robot on a showroom pedestal, its open back panel stuffed with paper and an old rubber stamp, while a salesman in a suit gestures at its polished front

When I build an intake or quoting system for a small business, the question I get asked most is some version of, "is this actually AI, or is it just doing stuff?" It's the sharpest thing anyone says in those meetings. It's also the question almost nobody selling AI wants you to ask, because the honest answer shrinks the invoice.

Here's the part they skip. Most of what gets quoted to you as "AI" isn't. It's rules. If the email contains the word "quote," route it to this folder. If the form field says "roof," send template B. If it's after 5pm, fire the after-hours reply. That's not a model making a decision. It's an if-then. Twenty years ago it lived in a spreadsheet macro, and it did the same job for the price of an afternoon.

I'm not knocking rules. Rules are the good part. They're deterministic. They do the exact same thing every time, they don't invent a customer's name, they don't book an appointment that doesn't exist. When I build something that has to survive a Tuesday where the client is slammed and nobody's watching it, the reliable core is almost always boring if-then logic. The AI is a thin layer sitting on top, and more often than not it's the layer that breaks.

The rules are the part that works every time. The AI is the part that works most of the time and costs ten times as much.

So why is all of it sold as AI? Because "AI" is priced differently than "automation," and "automation" is priced differently than "a filter you could set up yourself in software you already pay for." The vendor isn't flat-out lying. There's usually a language model in there somewhere. It drafted a reply, or read the sentiment of a message, or sorted an inbox. But the quote in your hand bundles that one genuinely smart step with six dumb ones and prices the whole thing like it all thinks.

Here's the move. Next time a vendor walks you through their system, stop them at every single step and ask one question: which part is the AI?

Point at the box on their flowchart that reads the email. Is that a model, or is it matching keywords? Point at the one that routes it. Model, or a rule? Point at the one that sends the reply. Did a model write that fresh, or is it dropping the customer's name into a template?

Then watch what happens. On the steps that really are rules, a good vendor will just tell you, and that's fine, that's honest work. On the steps where they go vague, where "the AI handles that" softens into "well, it's intelligent routing," you've found the markup. Intelligent routing is usually an if-then with a thesaurus.

Let me give you the concrete version, because I've built both halves. The genuinely hard, genuinely AI part of an intake system is reading a messy human message and figuring out what the person actually wants. "Hey do you guys do the big windows on the second floor, the ones over the garage" has to become "two-story exterior clean, needs a site visit." A model does that well, and that part is worth paying for. The part where you then email them to schedule? That's a calendar link and a template. That's not intelligence. That's a mail merge from 1998 with a nicer font.

When a vendor prices those two steps the same, they're betting you can't tell them apart. For a lot of AI shops right now, that bet is the entire business model.

So here's what to do with the answer once you have it. For every step that turns out to be a rule, ask the second question: what does this cost if I only buy the rules part? Because the rules are the cheap part, the reliable part, and increasingly the part you can set up yourself in tools you already own. Your scheduler probably does the follow-up. Your email probably does the routing. The boring eighty percent of that quote might be a feature you're already paying for somewhere else.

And the one AI step left standing? Now you can judge it on its own. Is it reading intake well enough to save your front desk an hour a day? Then it might be worth what they're asking for that step. But you'll only see that clearly once you've stripped out the if-then logic they dressed up to pad the bill.

I sell this. I build these systems for money. And I'm telling you that on most of the quotes landing in your inbox, the letters A and I are doing a lot of work an afternoon with your existing tools could do for free. Not all of it. The real part is real, and when it's real it's genuinely good. You just won't find it until you make someone point at it.

Ask which part is the AI. Make them point at the box. If they can't, or won't, you already have your answer about what you're buying.

§
Post-ready for LinkedIn
Half of what gets quoted to you as "AI" is a rule a spreadsheet could run. If the email says "quote," route it here. If the form says "roof," send template B. That's not a model making a decision. It's an if-then. Twenty years ago it lived in a spreadsheet macro. I'm not knocking rules. Rules are the part that works every time. They don't invent a customer's name or book an appointment that doesn't exist. The AI is a thin layer on top, and it's usually the layer that breaks. So why is all of it sold as "AI"? Because AI is priced differently than automation, and automation is priced differently than a filter you could set up in software you already pay for. Here's the move. Next vendor demo, stop them at every step and ask one thing. Which part is the AI? Point at the box that reads the email. A model, or keywords? Point at the one that sends the reply. A model, or a template with the customer's name dropped in? Where they go vague, that's the markup. "Intelligent routing" is usually an if-then with a thesaurus. I build these systems for a living. On most quotes, the letters A and I are doing a lot of heavy lifting. What's the last thing a vendor sold you as "AI" that turned out to be a glorified filter?
← All essays