We use AI more aggressively than almost any company you’ll meet — it drafts, codes, triages, and summarizes for us every single day. Which is exactly why we’ll tell you something most vendors won’t: about a third of the “AI for business” pitches you’re hearing right now will waste your money, and a few will actively embarrass you in front of customers.

Here’s the honest map.

Where AI genuinely earns its keep

The first response. When an inquiry arrives at 9pm, AI can acknowledge it, answer the obvious questions, and book the appointment — in your voice, in seconds. Nobody wants to talk to a robot, but everybody prefers a helpful instant reply over silence until Tuesday.

The blank page. Drafting the review response, the follow-up email, the service description, the job posting. AI’s first draft is 80% right in three seconds, and editing is faster than writing. Multiply that by every piece of text your business produces in a year.

The re-keying. Reading an order confirmation and typing it into accounting. Copying a lead from email into the CRM. Pulling the same numbers into the same Monday report. This is the highest-value, lowest-glamour AI there is — work that used to require a human simply because it involved reading.

The triage. Sorting what’s urgent from what’s spam, which review needs a careful human reply versus a warm thank-you, which inbox item is a lead versus an invoice. AI is an excellent first-pass filter when a human makes the final call.

Where it will embarrass you

Unsupervised customer conversations about money, promises, or problems. An AI that misquotes a price, promises a deadline, or handles an angry customer with chipper boilerplate does damage a discount can’t fix. Anything with stakes needs a human in the loop — not as a slogan, but as an actual workflow step.

Published content nobody edited. Readers can smell it, and increasingly, so can Google. AI-drafted, human-edited is a superpower; AI-published is a liability with your name on it.

Anything where “confidently wrong” is expensive. AI doesn’t know when it doesn’t know. Legal language, compliance claims, medical or financial specifics — the model will produce something fluent either way. Fluent and wrong is worse than absent.

The pattern behind the pattern

Notice what separates the two lists. AI excels when it accelerates a human — drafting, sorting, moving data, answering the questions with known answers. It fails when it replaces a human at the exact moments a customer needs judgment, accountability, or empathy.

That’s why every automation we build ships with the same architecture: AI does the repetitive middle, humans own the edges — the approval, the exception, the relationship. Our own inbox runs this way. So do the systems we build for clients.

If someone pitches you AI without ever mentioning where the human sits in the loop, they’re selling you the embarrassing half of this article.