The essay comes from an IBM executive, and its central claim is simple: companies keep measuring AI by how much faster one person can finish a task, when the real money sits somewhere else entirely. Value shows up when AI moves work across an entire process, so a purchase order can travel from procurement to finance to compliance without a person retyping it at every stop.
That argument is not new, but two studies published this year explain why it needed saying again.
The first is a widely cited study out of MIT. It found that 95% of corporate AI pilot projects fail to produce a measurable financial return. The more useful detail sits underneath that number: general tools like ChatGPT get used successfully in production roughly 40% of the time, while custom AI tools that companies build in house for one specific task succeed only about 5% of the time.
That gap is basically the IBM argument showing up as data. Handing someone a chatbot for one task is cheap and easy, and it rarely moves the needle on the business. Rebuilding an entire process so AI can operate across several teams and systems is much harder, and most companies that have tried it so far have not pulled it off.
Separately, McKinsey estimated in November that 57% of paid work hours in the United States could technically be automated using AI tools that already exist. That number describes what is technically possible, not what will actually happen. McKinsey's own researchers are careful to note it measures potential, since cost, complexity, and plain organizational inertia slow down how much of that potential companies actually use.
McKinsey has also estimated, in earlier research, that letting AI operate broadly across a company's daily operations could add between $2.6 trillion and $4.4 trillion a year to global corporate profit. That range is wide because it depends entirely on how much of the harder, workflow-level work companies are willing to do, rather than stopping at individual tools.
IBM's own contribution to this debate is a claim that it saved $4.5 billion over three years by using its own AI products on its own internal finance, HR, and procurement work before selling that software to customers. It is a solid case study, but it is also a sales pitch: IBM is grading its own homework with a product it profits from.
The useful takeaway for a business leader has nothing to do with IBM's specific number. It is the pattern underneath it. Buying individual AI tools for staff is the low-risk step almost every company has already taken. The harder step, and the one actually tied to savings, is redesigning who approves what and which systems talk to each other, so AI can run across a full process rather than sitting inside one job.
Gartner expects that shift to move fast. Today, fewer than 5% of the finance, HR, and procurement software companies already use has this kind of built-in automation. Gartner expects that share to reach roughly 40% within about a year, and businesses that wait for vendors to add it will fall behind the ones redesigning their own processes now.