.png)
In brief: Companies keep spending AI budgets on visible, customer-facing tasks like marketing and translation, where returns are hardest to prove. The real value — roughly 70% of it — sits in structured back-office processes like procurement, which are measurable, repetitive, and tied to costly external spending. Hungarian firms have barely started deliberately, which is their advantage: with nothing to shut down, they can begin straight away where the evidence points, not where it's most comfortable.
In an earlier article, we argued that the AI gap facing Hungarian companies isn't a technological problem but an organizational one: people use AI confidently, yet companies fail to build it into how they operate, and what we take for strategy is often just uncontrolled shadow usage. The lesson was that we don't need more individual use—we need a framework.
But a framework on its own is empty. The moment a company leader decides to start using AI deliberately, at the organizational level, they run straight into the next question—one they rarely get answered: which process should they start with? Where should the first well-considered AI investment go so that it delivers real results, rather than ending up like the many attempts that fizzle out?
International data gives a surprisingly consistent answer to this. Most companies, it turns out, pick precisely the area where AI pays off the least.
Let's start with an uncomfortable observation: most companies—Hungarian and American alike—regularly look for AI's value in the wrong place.
MIT's much-cited 2025 study, The GenAI Divide, found that more than half of corporate AI budgets go to visible, customer-facing areas—sales and marketing—while the greater value lies in automating back-office processes. This distortion is even sharper in Hungary: according to domestic surveys, AI shows up most often in translation, writing emails and summaries, and marketing communications—that is, precisely in the surface-level, easily accessible tasks that offer the least demonstrable business value.
This is understandable. These tasks are immediately visible, require no system integration, and anyone can try them out with a free tool in five minutes. But that's exactly the problem: what is easiest and most eye-catching is rarely the most valuable.
So where is the real value? Boston Consulting Group's 2025 report, The Widening AI Value Gap, puts numbers on it: roughly 70 percent of AI's potential value is concentrated in core business areas, and the best-performing companies generate value not through flashy customer-experience demos but deep within their processes. According to the report, they achieve 1.7 times the revenue growth and substantially better results than the laggards—not because they buy more AI, but because they deploy it elsewhere.
One detail sheds direct light on where the value comes from. According to MIT, in successful rollouts the savings came not primarily from headcount reductions but from cutting external spending: terminating outsourced contracts, replacing agency and vendor fees. Real returns, then, arise where structured, repetitive processes run—processes tied today to external providers or manual administration.
This is precisely the domain of back-office operations. And one of its clearest examples is procurement.
It's worth pulling out a framework built for exactly this choice: the Gartner AI initiative prioritization funnel. It's a seven-step filter for ranking AI initiatives—every idea passes through a sieve of multiple criteria, and only those that clear all of them remain. Of the seven criteria, three matter most here: is the initiative tied to a business metric that leadership understands and considers important; is there prior data with which returns can be credibly demonstrated; and how much work does implementation require.
Run typical Hungarian AI usage through this filter and the problem stands out. Translation or marketing copy has no clear business metric and no basis for comparison—it "feels faster," but you can't say by how much, and you can't defend it in a leadership meeting. Structured back-office processes, by contrast, satisfy the conditions almost without exception.
Take procurement. It has metrics everyone understands: cycle time, savings achieved, contract compliance. It has prior data, so returns aren't a feeling but a number. It's repetitive and predictable, so implementation doesn't require a major organizational overhaul. And it typically consists of well-defined, sequential steps that give AI the right framework to work reliably.
The same holds for the rest of back-office operations: financial reconciliations, invoice processing, certain HR administrative workflows. What they share is that they are measurable, predictable, and today typically tie up expensive work hours or an external provider.
From here the crux becomes clear. While the company uses AI for translation, where there's nothing to measure, a well-paying, measurable, easily implementable area sits untouched in its back-office processes. The attention goes to one place, while the value would be in the other.
This is where the Hungarian angle comes back in—and where a surprising advantage emerges.
The large American corporation's problem is often that it has already spent too much in the wrong place—it's full of half-finished, flashy sales experiments it now has to shut down at a loss. The Hungarian mid-size and large enterprise is in a different spot: its problem isn't bad decisions, but that a deliberate decision has barely been made yet.
This is both bad news and good. Bad, because organizational-level AI use shows a real gap, explained mainly by a lack of expertise. Good, because there's nothing to shut down. The Hungarian company isn't stuck in a botched AI strategy—it simply doesn't have one yet. And whoever starts with a clean slate can begin right away in the area with good returns, instead of retracing others' dead ends: the full array of flashy but useless experiments.
The Hungarian company leader, then, doesn't have to rack their brains over how to trim their list of projects, the way their American counterpart does. They have just one job: when they finally make a deliberate choice, start where the evidence points as the best place—not where it's most comfortable.
Three principles mark out a good starting point, and none of them requires a large AI investment.
A basis for comparison is worth more than spectacle. A good starting point isn't where AI is most impressive, but where there's something to measure against. Where you know exactly how much time, money, and error a process costs today, returns will be credible—and that's almost always back-office operations.
The trail of external spending shows where the value is. International data indicates the cleanest returns come from replacing external spending. Where money flows out to providers and agencies today, or where manual administration ties up expensive internal staff for repetitive tasks, that's where the real opportunity lies.
Start small, but measurably. A single well-defined process—a contract-renewal cycle, a supplier data check, an invoice reconciliation—where returns can be drawn out in a few simple steps, is a better start than an ambitious but uncertain program. First proof, then measurement, finally expansion. This is also the closing thought of the Gartner framework: the right entry point isn't the biggest dream but the fastest, measurable gain.
Back-office operations—and procurement within them—aren't the most exciting AI terrain. Nor the flashiest. That's exactly why they remain untouched at so many companies, even though this is where the real value would be easiest to demonstrate.
The AI advantage ultimately comes down to a single decision: not whether the team uses ChatGPT, but whether the company starts in the most comfortable area or the one with the best returns. Most Hungarian companies haven't even made this decision yet—which is precisely what means they're free to head in the right direction.
.png)
.png)
.png)
.png)
.png)