AI can make almost anyone sound like they know what they're talking about. Ask it to write an ad strategy, and it will produce something structured, confident, and reasonably plausible in about ten seconds (especially if you don’t know better!). Ask it to draft an HR policy, and it will hand back something formatted like it came from someone who has done this before. That's the appeal, and it's also the trap. What AI produces this way is thin. It sounds like expertise because it's trained on what expertise sounds like, not because it has any.
We saw this directly while building out our own paid advertising capability this year. AI can build an ad campaign: it can structure the account, write variations of the copy, and set up the targeting in a fraction of the time it used to take. What it cannot do is know what ad strategy is right for a given business, because that comes from having run the tests, watched what converted and what didn't, and built a database of real feedback over time that tells you why. We spent weeks extracting that kind of thinking from someone who has actually lived it, turning specific, tested principles into something we could point AI at with real direction. Without that layer, AI's version of ad strategy is a decent-sounding guess dressed up as a plan.
The same gap shows up in a pattern we keep running into with clients. A company decides to replace part of an internal function, HR or funnel management are good examples, with a bolt-on AI tool. The pitch is straightforward: the tool handles the routine work, headcount comes down, the function runs itself. A few months in, the tool is live and the return isn't there. The tool doesn't know how HR or sales works inside that specific organization. It doesn't understand the judgment calls a real HR function makes daily, the nuance in a specific policy exception, the read on a situation that isn't in any handbook. That knowledge lives in subject matter experts who have done that job, and it never made it into the tool.
What changes the outcome isn't a different tool. It's putting someone who runs HR in the room and having them teach the AI the specific exceptions, the judgment calls, the read on a situation that isn't written down anywhere, the same way they'd bring a new hire up to speed. The AI can easily handle the routine volume, intake, first-draft responses, scheduling, the paperwork that used to eat a full day. But it's now operating on real institutional knowledge instead of guessing at it, because someone with the expertise built that knowledge into how the AI works.
The outcome for the client isn't a smaller HR team running the same process faster. It's an HR function where the routine work finally gets out of the way of the judgment calls, and where the expertise that used to live in one or two people's heads is now written into a system the whole team can use, refine, and hand off to whoever does the job next. That's the difference between buying a function-replacement and building a function that gets better every time someone corrects it. To work successfully, AI requires the subject matter expertise required to run it well be transferred into it.
The vast majority of AI SaaS tools don’t have domain experts on staff. If you are considering a technology bolt on, we strongly recommend that you explore these criteria extensively, as well as how they train their model on your exact business, and how they teach the model all your use cases. SaaS is very different from industrial businesses, and we are rarely finding clients who are happy with the bolt-on tools they have “installed”. Building with AI well requires someone who actually knows the domain: what good looks like, where the real risk sits, what the last several years of getting it wrong actually taught. Build a pre-call briefing skill without a real sales methodology behind it, and it will produce a plan that looks complete and misses everything that matters. The tool didn't fail. The expertise directing it was never there.
It raises the real question about how most companies still approach AI adoption: as something you buy, not something you build. That's what we want to dig into next.
If this was useful, the rest of the Commercial Intelligence Report is one click away. Subscribe to the Commercial Intelligence Report, a weekly analysis on how industrial commercial teams are turning AI into measurable ROI, and the discipline that makes it stick.