I pitched the same project twice in one week: in Budapest and in Zagreb.
In Budapest, a client and I were working out how to turn an AEO pilot into an ongoing operation inside HubSpot. AEO is about how ChatGPT and similar tools mention your company when a buyer asks them for a supplier. We would track the signals. Every week, they would get a report showing where their content has gaps. A smart, responsible arrangement.
In Zagreb, a forward-thinking entrepreneur asked me straight away: “Instead of flagging the signal in a weekly report, why not generate the missing content and send it to the CMO for approval before it goes live?” With one question, he turned a report reader into an editor. Content marketing on AI steroids. We built it for him.
Both clients use AI. But they stand on two different steps. That week came back to me when I was reading Andrej Karpathy.
The funnel
Karpathy, an OpenAI co-founder and former head of AI at Tesla, noticed in April that people increasingly talk past each other about AI. Some tried the free version and now laugh at clips of ChatGPT’s voice mode unable to tell whether to walk or drive to the car wash. Others write code with it and can’t believe their eyes.
In October, he drew it as a funnel. At the top, 6 billion people who have barely touched AI. Below them, 1 to 2 billion occasional users of free versions. Lower still, 20 million who use agents to write code and solve maths problems. At the bottom, 5,000 people inside the labs.
Where are the executives? Karpathy puts them on the second step. People who spend their days in conversations know what AI can do only second-hand. Those are my buyers. And they are the ones who decide which step the company will even see.
Slovenia is above average. Unfortunately.
According to Eurostat, 21.61% of Slovenian companies with at least ten employees use AI. The EU average is 19.95%, and Croatia is at 15.19%. A nice number for LinkedIn.
Then you look at what companies use it for. Most often, to analyse text. That is exactly the step where, according to Karpathy, the models are improving most slowly. Code has a test that tells you whether it works. Text has no such test.
The statistics count who uses AI. They don’t count what work it is given. The US firm Bridge Group, which researches B2B sales teams, measured this too this year: in the third of companies that build AI most deeply into their salespeople’s work, 57% of reps hit full quota; in the third with the shallowest use, 39%. It’s a correlation, not proof of cause, as they point out themselves. But the gap is too big to ignore. So I won’t congratulate you if your team uses AI every day. Daily use for writing and summarising is often a sign that the team is stuck on the second step.
Two salespeople
Karpathy draws his funnel for code and maths. Sales has the same steps. I’ve been selling for twenty years and I still do my own selling, so that’s where I see them first.
The first salesperson uses AI every day and, if you ask, will tell you so with pride. He drops his follow-up into ChatGPT with the instruction “make it more professional”. The client gets an email that sounds like every other one in her inbox. No surprise. The same model wrote them all.
The second walks into the meeting already knowing that the client has a new purchasing manager. He knows what was left open last time. He knows which deal is going cold because his champion has gone quiet. He didn’t skim the website ten minutes before the meeting. His brief was prepared by an agent that sees the CRM, email and transcripts.
The first gives AI tasks. The second has handed it a process. The first uses it where it is weakest and concludes it is overrated. The second doesn’t debate it. Meanwhile, he’s selling.
Once the machine runs, people are the bottleneck
Back to Zagreb. Once the content machine was running, the entrepreneur hit the next wall: new salespeople. He says onboarding time is exactly what holds him back from expanding into new markets. So now we are building him a sales mentor.
He’s not unusual. The same Bridge Group study found that a new salesperson needs 6.2 months to become fully productive, the longest since they started measuring. In 2020, it was 4.3 months. In the same years, ChatGPT arrived in sales, and companies started hiring more experienced people: they now ask for 3.7 years of experience, up from 2.7 in 2022. Even so, onboarding didn’t get shorter. It got almost 50% longer.
If that average holds in our region too, two new salespeople for a new market draw 12.4 monthly salaries between them over 6.2 months of onboarding, before either is fully productive. And again for every new market.
A newcomer doesn’t need more information. He needs the context that, today, only your best salesperson carries in his head. No training course gives him that, and neither does a week of shadowing a colleague. An environment can: one that gets that context out of people’s heads and into the system, and puts the newcomer on the third step from day one. That is exactly what the sales mentor we are building in Zagreb is for.
You can hire a salesperson who knows AI. But the environment where AI knows your customers is something you have to build yourself.
What only leadership can do
The third step isn’t a matter of individual curiosity. It takes three decisions from leadership.
Connect AI to your CRM, email, calendar and meeting transcripts, or it will keep writing nicely formatted nothing. Write down what a good deal, a good meeting and a genuinely agreed next step look like, because code took off precisely thanks to a test that tells you whether it works. In Zagreb, that test is the CMO, who approves the content before it goes out. And stop giving AI tasks. Give it a process: not “write an email”, but “every Monday, review our existing customers, find where we can grow, propose the next step and log it in the CRM”. The salesperson then stops writing. He decides.
At your next sales meeting, don’t ask who on the team uses AI. Ask what work AI did this week without anyone telling it to.
A ticked box is not a step
Budapest didn’t make a mistake. For a company just getting to know AEO, a partner who runs the operation for them is a smart first step. Zagreb went two steps further and is now solving a problem most haven’t even noticed: how to get people onto the same step as the machine.
Eurostat would tick the same box for both. But a ticked box is not a step.
igor.pauletic@frodx.com
P.S. Karpathy himself calls his numbers back-of-the-napkin estimates. Eurostat’s and Bridge Group’s are measured, but each measures something different.
More on how AI agents connected to your CRM take over processes, not just single tasks.
