A recent Goldman Sachs analysis argues that the AI story has reached a turning point. A few years ago, the big question was who could keep up with the staggering demand for the technology — who would make the chips, the servers, the data centers. Now the focus has shifted: the question is whether companies can actually turn their AI investments into real, measurable value.
The analysis is written with the stock market in mind, but it's just as instructive from a leader's perspective. Because the insight behind it applies to anyone using AI: access alone no longer confers an advantage. You can reach the same powerful model your competitor can, and both of you are a few clicks away from it. And once everyone is working from the same toolbox, the question changes: it's no longer about who has AI, but about who can actually show results with it.
Generative AI reached roughly 53% adoption in about three years — faster than the personal computer or the internet did in their day. And when a tool spreads that widely, access becomes a given: anyone can buy a serious model, and it's equally available to the competition.
What sets companies apart is what each one does with it inside its own operations.
One data point makes this strikingly clear. According to last year's MIT NANDA study, the vast majority of enterprise generative AI projects produced no measurable financial return. And here's the crux: the researchers found the problem wasn't the models' capabilities, but that companies couldn't find a way to embed the technology into how their organizations actually run. Initiatives stalled in the pilot phase, never connected to existing systems, or tried to create value where there was little chance of it in the first place.
A familiar pattern. The tool didn't fail — the trouble was that it never made it to where it mattered, into the actual flow of work.
Goldman essentially arrives at the same conclusion: the success of enterprise AI hinges on whether a company can get its own data and workflows in order. Put simply, subscribing to a powerful model isn't enough. You have to clean up your data, connect your internal systems, and decide which tasks to hand to which AI.
That last point carries more weight than it first appears, because not every job calls for the most expensive, most powerful model. A simpler, low-risk, high-volume customer-service or administrative task can be handled by a cheaper, smaller model — faster and at a better cost, too. A complex financial calculation or a high-stakes decision, on the other hand, can well justify the more advanced, pricier option. And it's precisely this sorting that matters most when it comes to return on investment.
The realization, then, is that getting your data in order, connecting your systems, and sorting your tasks aren't three separate chores but facets of the same question: how should AI be built into the way the business operates. And that calls for a layer that ties it all together — one that routes each task to the right model, keeps costs in check, and ensures AI isn't a standalone gadget but an integral part of operations. We call that layer workflow, and it's essentially where everything is decided.
Let's look at what this means in practice. At most companies, AI today looks something like this: there are a few enthusiastic employees who paste some text into a chat window and then manually carry the result back into their work. Useful, but an isolated fix. It isn't measurable, it's hard to scale, and if that person goes on vacation, whatever they were working with — and however they did it — likely becomes inaccessible.
A workflow, by contrast, means AI is built into where the work actually happens. It receives the data it needs automatically and delivers the result right where the next step picks up — the process is repeatable and traceable, not living in a single person's head. This is essentially what separates "we tried AI" from "AI is paying off": one is a demo, the other a working system.
What's more, this is exactly where smaller, purpose-built models gain importance, since they're faster, cheaper, and easier to tune to a specific task. But that advantage only emerges if someone actually breaks the operation down into sub-tasks and deliberately decides what goes where. Without a workflow, that decision never even gets made.
So the question isn't whether you have access to the best model, because soon enough everyone will. It's whether your data is in good enough shape for AI to genuinely work with it. Whether your systems are connected, or each one stands on its own island. Whether anyone at the company knows which task is handled by which tool — and whether anyone is making that call deliberately at all. And whether AI has become part of how the business runs, or remains the private hobby of a few colleagues.
Where there are no good answers to these, even the most expensive model won't pay off. Where there are, even modest tools can deliver serious results. As Goldman puts it, the next big test won't belong to the chipmakers either, but to the companies that can embed the technology deeply enough into their operations to produce tangible results. Because AI isn't decided at the model — it's decided at the process.
Source: James Covello (Goldman Sachs Research): Will the Corporate Investment in AI Pay Off?
We built Fluenta One for exactly this purpose: we don't hand you yet another AI — we help you map and automate your processes so that AI actually creates value. If you'd like to see how Fluenta One could fit into your workflows, let's talk.
.png)
