Someone on your GTM team opens a chat window. They type in a question they believe has never been settled. Who’s really making the decision here? Why did last quarter stall? Should we change our positioning?
Thirty seconds later there’s an answer on the screen. It’s clear, confident, well organized. And now the question is closed.
That’s what AI does to go-to-market when it goes wrong. Convincingly, it hands you perceived certainty nobody earned and fast enough that nobody stops to check.
Paula Skaper runs workshops for leadership teams, so she pays attention to how people ask for help. She’s also a GTM leader and AI advisor, and she joined Mark Stouse and me on last week’s Causal GTM Leader to chat about when and why AI actively makes go-to-market worse.
The morning we recorded, Paula sat in on a webinar about AI and neuroscience. The session was good. But the chat running beside it told her more. People kept typing the same questions. When do we get to the how? When do we get to the skills we can download, the thing we plug in and are done with?
I immediately thought of the current HubSpot LinkedIn campaign shown below. You know the drill. Just grab the playbook of a thousand prompts, and you’ll be fine. Of course it’s a lie. It’s just another lead magnet circa 2019 dressed as 2026 help, aimed at people who are already drowning or don’t know they are.
Truth be told, as Paula explained, companies sell easy because buyers keep buying easy, and that pitch has thirty years behind it. So as much as I’d like to demonize companies like HubSpot, we sold the internet to business the same way. We still sell software the same way. “We’ll make it easy. We’ll reduce your workforce.” Businesses bought exactly what we told them to buy, and now we’re annoyed that they believed us. Fair point.
So when someone asks AI to think for them, they’re likely three people short and out of options. That’s why the easy button gets pressed. We can empathize and demonize all we want. It still doesn’t work.
Because GenAI is a pattern-matching tool, it can hand different businesses the same strategic advice if left in the wrong hands unchecked.
But as Mark explained, a thousand people using the tool properly on a thousand different business cases will rarely land on the same recommendation. When they do, something specific went wrong upstream.
When you do get the same kind of response, it’s usually because you have starved the tool for contextual insight.
- Mark Stouse
Researchers at Esade, Sydney and NYU Stern found that leading models consistently recommend whatever aligns with fashionable management thinking rather than the logic of the case in front of them. They called it trendslop and warned against assuming more context alone will fix it.
Paula had spent months building systems she was proud of, checking outputs across multiple tools, verifying sources, automating what she could. She took the certification, went deep, and by her own account thought she had it nailed.
One day my daughter said, “You’re not remembering our conversations. I tell you something and an hour later it’s gone. I think you should get checked.” And then I thought, “What am I doing?” I screwed it up royally. And I really did screw it up royally.
- Paula Skaper
There’s a lesson here for all of us. She realized later how much time she was spending in AI (seven or eight hours a day) and how little she was retaining So she shut it off. For thirty days. No AI. Just one meeting recorder, a notebook, and pen and paper.
It took Paula thirty days to start over, but she’s rebuilt her discipline since. Pen and paper first, then the tool. She’ll sit in a park with a notebook and no signal, think a problem through, and only then open a chat and hand over the context she arrived with.
The person who finally called it out was her daughter. Which is the same gap I’ve written about before, from the other direction. The checks an org chart or a manager would imply have to be built deliberately, or they aren’t there at all.
GenAI’s strongest capability isn’t generating answers. It’s finding the holes in yours, and the reason is structural. You’re pattern-matching against human failure, which has been running consistently since the Bronze Age.
It’s the most consistent pattern that exists anywhere. It’s human failure. Human success is not a pattern.
- Mark Stouse
So if you want to remove a lot of failure quickly, go find what isn’t working first. Knock out the underpinnings of your pet theory and see what’s still standing. Mark has run more than a thousand tests this way, several hundred on his own thinking, and describes it as deeply humbling. It has also saved his ass repeatedly.
There is a thorny catch, however:
You don’t much want to spend time with something that keeps telling you you’re a knucklehead.
- Mark Stouse
My own version is to make the models argue. It’s something I learned from my conversations with Mark over the past two years. I run the same work through two or three AI tools and inspect where they contradict each other. Different mechanics, same principle. The disagreement is where the thinking has to happen. It’s fun to watch.
According to Mark’s own research and experience, AI initiatives fail at roughly the rate large technology implementations have failed for the past fifteen or twenty years. He was being generous.
RAND cites estimates that more than 80% of AI projects fail. That’s about twice the rate of IT projects that don’t involve AI. RAND’s research worth reading. They interviewed sixty-five experienced AI practitioners, and 84% named leadership-driven failure as the primary cause: leaders who misunderstood or miscommunicated the problem, and technical teams left without the business context to fill the gap themselves.
RAND is describing the Reality Gap between what leaders wanted and what their technical people understood, which isn’t quite the same thing as starving a model of input. But the shape is familiar, and go-to-market has been telling itself a version of this story for years.
No one is asking themselves, “Is what we’re doing still effective?” They’re only asking, “How much more efficient can we make this?”
- Mark Stouse
Buried in that second question is the elephant in the room: that we’re effective, and volume is the only variable left. CAC climbs, results flatten, and the answer is always more of the same thing, just harder. A cheaper task can still produce a more expensive business, and AI will help you run the wrong system faster and with more conviction if you allow it to.
This one is thorny. It’s the opposite of every AI pitch you’ve read.
If you haven’t used AI, you’re going to make it worse before you make it better. When you’re using AI well, it’s initially going to slow you down.
- Paula Skaper
Start small. When you download one of those skills or prompt packs, install it and then open the file and read what it says. You’ll find assumptions baked in that don’t match your business and instructions you’d never give. Edit them until they reflect the context in your head.
If you lead a team, the training that matters isn’t prompt writing. You can download 1,000 prompts from HubSpot. Remember? Plus any GenAI tool will help you write the prompts for you if you ask it.
Critical thinking is what’s missing. That’s the required training. How to break a problem into parts, how to prioritize, how to look at something and decide it isn’t good enough yet.
Teach them how to think… and if you teach that, they’ll work out the tool for you.
- Paula Skaper
Pick the last significant thing AI helped you produce. A plan, a segmentation, a positioning doc, a forecast. Open a fresh chat, paste it in, and ask the tool to find everything wrong with it. Not to improve it, to break it.
Then sort what comes back into two piles: objections you already suspected and buried, and objections that hadn’t occurred to you. The first pile is what your team isn’t saying out loud. The second is what your context was missing. If neither pile bothers you, you didn’t give it enough to work with.
Missed the session? Watch it here.
Paula Skaper is the founder of 33Dolphins Growth Strategy and author of Rethink, Realign, Reinvent.
If you like this content, here are some more ways I can help:
Cheers!
This article is AC-A (“Authenticity Commons, Assisted”): AI helped produce it, with me in the lead. Here’s what that means.
Achim’s Razor is also published as a LinkedIn newsletter.
