Another true story, this time courtesy of Paula Skaper:
One of her clients was working through everything an AI pilot would touch: their environmental and values commitments, what they owed stakeholders in reporting, and data they weren’t sure was accurate. Then they made the call.
No pilot.
Instead, they wrote a governance policy, agreed on acceptable use cases, and set rules for data sharing. People could experiment with AI in their own work, inside those guardrails. They’d look at it again in a year, once they could see where the legislation and the technology were going.
Eight months later, it was absolutely the right call. Paula said so much had changed that anything they’d started building would have already been undermined.
During our last Causal Revenue jam session, Mark Stouse, Paula, and I got into what should happen before you fund another AI pilot. Here’s the recap.
Only 25% of companies in Deloitte’s State of AI in the Enterprise 2026 have moved 40% or more of their AI pilots into production. But production and business success are two very different things.
Many get dropped right on top of whatever the team already does. Cost per lead goes down and the dashboard looks better. But did qualified pipeline go up? Win rates? Or did you just get cheaper at producing leads nobody wanted?
“Everybody has just been dropping LLM or agentic AI right on top of what they’re already doing. QED, everyone is assuming that what they’re doing is effective… if the more you do that and you’re strapping it on, you’re making something ineffective, more efficient, higher productivity, you’re actually negative faster.”
Mark
Paula has a very technical term for it: scaling to suck.
Figure out what’s working, cut what isn’t, then speed up what’s left. So before a pilot gets a dime, answer three questions.
Taking work off people’s plates is a legitimate business problem. But that answer still needs a number the CFO believes, and that’s where a lot of efficiency pilots come up short.
If you can’t answer the first one, the other two don’t matter. It’s the same discipline as starting with the decision before the data, only more expensive to skip.
We see plenty of efficiency gains land in the admin backlog: data cleanup, data accuracy, the work people kept pushing to the side of the desk. It’s worth doing. Better data probably leads to better decisions down the road. But without a measurable financial benefit, good luck convincing a CFO to spend the money.
Leaders chase efficiency anyway, because it fits an earnings-per-share story. Cut costs, report better EPS, repeat. You can’t ride that forever, and the costs nobody counted keep piling up (we got into those in AI Is Not Free Labor).
“The irony of the whole thing is, is that the greatest efficiency improvements that you can make in a business is identifying what doesn’t work and killing it.”
Mark
Finding what doesn’t work starts with how the work actually gets done, and leadership’s picture of that is rarely right. Paula thinks the process conversations are where pilots break, and where the real learning is. Mark thinks process work hasn’t produced the payback CFOs need before they invest more. We didn’t settle it.
A pilot that’s turning into an expensive efficiency play needs someone with the authority to stop it. That’s usually the officer running that part of the business.
In 2023, Delaware's Court of Chancery ruled that corporate officers, not just directors, owe oversight duties within their areas of responsibility. Canada's federal corporate law also requires directors and officers to exercise reasonable care and diligence. The legal tests differ, but the principle is similar: the courts expect you to know what’s working, what’s not, and act accordingly.
Which officer, though? An AI tool managing your Google ad spend could belong to the CMO, the CTO, or the chief growth officer.
Often it’s more than one. Mark was an expert witness on a CRM data-quality case where the suit named the CIO, the CRO, and the CDO. IT owned the system. The data team owned governance. The CRO owned how well the reps entered data. It’s called the nexus of accountability: each person on the hook for a different piece of the same failure.
In other words, a pilot that touches three functions can easily have three owners, each answerable for a piece. One of them still needs the authority to stop it, and everyone needs to know which one before kickoff. We drew the same line in Automate the Task. Own the Decision.
Even with the right person named, someone still has to tell them the pilot isn’t working.
I’ve seen people raise their hand about a project they couldn’t see helping the business, and get told to zip it and keep going. After that it turns into pass-the-buck. Why are you doing this? Well, so-and-so told me to. Someone can blow the whistle, but somebody has to listen.
Nobody loves being held accountable. But in a lot of companies, accountability has come to mean blame, and once it means blame, people stop telling you things.
When Paula ran her agency, she had a conversation with every new hire in their first week that went sort of like this:
“At some point, you’re going to screw up. Unless you deliberately do something really bad, you’re not going to get fired for screwing up. You will get fired for hiding it, and you will get fired for doing it twice.”
Paula
The fix for a mistake comes from the person doing the job, because they know what they’d do differently. But they’ll only tell if it’s safe to.
Letting someone off the hook doesn’t help either. When they don’t do their job, someone else has to pick up the slack. Mark calls it a tax on the rest of the team.
Missed the session? Watch it here.
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.
