There's a quiet shift almost no one is naming. For years, the bottleneck in any artificial intelligence project was technical: you needed to code, train models, connect systems. Whoever mastered that part held the advantage.
That advantage has shifted.
Today, building an AI agent that answers on WhatsApp, drafts emails or processes requests is within reach of many profiles, non-technical ones included, with a weekend and some curiosity. The tools exist, they're cheap, and they get better every month. But it's worth being precise: anyone can build a prototype that replies; very few can assemble processes, data and brand tone into something that's still working six months later, with real clients rubbing against it every day. That second part is still hard. What's been democratized is the "it replies," not the "it holds up."
Even so, the basics have been democratized. And when something gets democratized, it stops being a competitive edge.
So where does the difference now lie between an AI project that transforms a company's business and one that ends up an expensive toy nobody uses?
Not in the code. It lies in a far more uncomfortable question: which problem is worth solving, and for whom?
The project that works but doesn't help
There's a very specific kind of failure in AI applied to small businesses, and it's rarely a technical fault. It's the project that works perfectly —the bot replies, the automation runs, the data gets processed— but that doesn't move the business needle. Technically flawless. Commercially irrelevant.
It happens because it started from the wrong question. It started from "what can we build with this?" instead of "what is the client actually trying to achieve, and where are they losing money or time right now?"
Those are opposite starting points. The first produces solutions in search of a problem. The second produces solutions the client was missing without being able to name them.
Anyone who has spent years selling, negotiating and understanding real clients learns this the hard way long before touching a single line of code: a technically brilliant product that doesn't solve a real pain doesn't get sold, doesn't get used, and doesn't get paid for. Marketing —the real kind, not the advertising kind— is the discipline of always starting from someone else's problem.
Three things commercial judgment brings to AI
This isn't marketing versus technology. It's about what kind of judgment comes into play before the technology does anything. There are three concrete contributions a commercial background puts on the table that pure technical skill doesn't guarantee.
1. Start from the job the client is trying to do, not from the feature.
In marketing there's a framework —Jobs to be Done— built on a simple idea: the client doesn't want a drill, they want a hole in the wall; and deep down, they don't even want the hole, they want to hang the picture. Applied to AI, the question isn't "what can this model automate?" but "what specific task is stealing hours and money from this business, and how do we take that weight off them?" That reframing completely changes what gets built, and in what order.
It's not that every technical profile lacks this judgment —there are more and more engineers with strong business instincts—. But the path of sales, marketing or commercial leadership forces you to learn it the hard way much sooner.
2. Knowing what NOT to automate.
This is possibly the least intuitive and most valuable contribution. A purely technical approach tends to automate everything automatable, because it can. But there are touchpoints where the human relationship is the value, not a cost to eliminate.
And this isn't a hunch: it's an observable pattern. When a delicate interaction is automated badly —the sensitive call, the complaint, the first contact that defines a relationship— the effect isn't savings but lost trust: referrals drop, complaints rise, the client cools off. Commercial judgment knows how to tell apart the process you should get out of the way from the moment you should protect.
The rule, stated precisely, isn't "never automate this." It's: experiment carefully at those points and measure the impact on satisfaction and retention before scaling. Diagnosing before automating isn't decorative caution: it's what separates a solution from a mess.
3. Turning data into a decision that moves money.
AI produces outputs: classifications, summaries, lists, predictions. But an output isn't a result. And the leap between the two is almost never automatic —not even with AI—: it's process design and commercial judgment.
One example makes it clear. A market-leading dental management software contains, in its quote listings, everything needed to know which treatments were quoted, which were accepted, which were left pending and why others were rejected. The data exists. It's right there, complete, waiting. And yet, in many settings, treatment opportunities are systematically lost —not over price, but because nobody followed up—.
The machine can flag the quote that's gone three weeks without a reply. But what to do with that signal —whether to call, when, in what tone, what to say and what to leave out— is a matter of commercial judgment, not software. Data without the right human action closes nothing. And only someone who has lived through a sales cycle knows how to design the automation so it ends in a conversation the client appreciates, not in a pestering that drives them away.
The base technology is the easy part
This isn't a case against the technical. Execution matters, a great deal. But the nuance is worth sharpening: what's become cheap and accessible is the base technology —standing up the system that replies—. What's hard is no longer answering, but answering well: reliably, securely, integrated with real data and legacy systems, and aligned with a regulatory framework that in Europe is increasingly demanding. That isn't "solved" for anyone with a weekend.
The question a business should ask whoever is going to implement AI for them isn't "can you build this?" —more and more people can—. The question is: "how do you decide what's worth building, and how do you know it will solve my problem and not the one you imagine I have?"
What's truly in short supply
Here's the heart of the matter, and it isn't what it seems. The clash isn't "technical people versus commercial people." What's scarce is neither —there's plenty of both—. What's scarce is the hybrid profile: people who have lived through a full sales cycle, who understand why people buy or don't buy, and who also know what AI can and can't actually do today.
That profile is rare because the two paths that form it almost never cross. The one who comes from technology has rarely had to close a sale with a wary client. The one who comes from business rarely knows where a model's real limits lie. And AI applied to a small business lives precisely at that intersection: in the person able to look at a process, understand the business pain, and decide with judgment what gets automated, what gets protected, and what doesn't get touched at all.
Twenty years spent understanding clients and markets are not a decorative add-on to an AI project. Combined with knowing what the technology actually does, they are increasingly the only thing that separates a project that moves the needle from the pile of technically correct solutions nobody asked for.
AI makes the answer cheap. The right question is still expensive.
