I came across a clip this week from a guest lecture Steve Jobs gave at MIT Sloan in 1992, and I haven’t been able to stop thinking about it. Jobs gave many memorable talks, but I hadn’t heard this one until I came across it.

In this video, he’s in front of a business school audience covering a seemingly prosaic topic: why some companies get a return on what they spend on technology while others spend the same money and get nothing. The argument is now thirty-four years old, but it describes the AI-related conversations I’m having with clients right now almost exactly.

Steve Jobs at MIT Sloan, 1992. The relevant passage runs from 3:39 to about 5:45; the full talk is 72 minutes. · Watch on YouTube

Same spending, different allocation

Jobs is working from a book by Paul Strassmann, The Business Value of Computers, written before Strassmann became CIO of the Pentagon. The book is built on Strassmann’s years of measuring what companies got back from their spend on computing, drawing on data he assembled from 292 enterprises. The finding Jobs pulls out is this: companies across the entire spectrum, from the not very successful to the really successful, all spent roughly the same share of revenue on information technology (about two percent). The spending didn’t sort them into winners and losers, but where they applied it certainly did.

The less successful companies, Jobs says, put most of that money into management productivity, which in 1992 meant personal computers: spreadsheets, word processing, that is, the tooling that helps people coordinate work and report on it. The more successful companies put most of their money into operational productivity, that is, operational applications built around the specific job of making the product and getting it out the door.

Then he drops the bombshell - he admits the finding was “not very pleasant” to read, because he’d just spent ten years building computers for the wrong bucket. And I believe that same problem - applying technology to the wrong area of the business - still exists today.

Strassmann published a follow-up to this work in a 1997 Datamation article, expanding the sample from those 292 enterprises to 539 corporations. Across the larger study he again found no connection between computer spending and financial results. Spending on its own didn’t produce results; pointing it at the right place was what made the difference.

Thirty-four years later, the same result

McKinsey published its latest State of AI survey on August 25. In that report, only thirty-seven percent of companies report any EBIT impact from AI at all, unchanged from a year ago despite another year of rising investment. But more strikingly, nearly three-quarters of the companies McKinsey classifies as high performers have fundamentally redesigned their workflows around AI. Among everyone else, it’s about one-quarter.

That’s the 1992 finding again, albeit measured with different instruments in a different decade. What separates the two groups isn’t the size of the budget but rather whether the companies used that money to change how they work, or just to make the existing work easier to observe and manage.

Where is your spend aimed?

A chatbot bolted onto the website is management productivity wearing a new coat. So is the meeting summarizer, the writing assistant, and the tool that turns your monthly report into a better-looking monthly report. None of that is useless - I use these tools myself. But they coordinate and assist the work rather than transform it.

Rewiring how a quote gets built, how an order gets checked before it ships, how a warranty claim gets triaged, or how a technician’s notes become the next service call: that’s operational productivity, and it’s where the thirty-four-year-old data and the brand-new data agree the real returns exist.

Fix the workflow, then point the AI at it

Look at the verb in the McKinsey finding: Redesigned, not adopted or rolled out. The high performers changed how the work happens, using AI as a tool to enable that. That order matters, and unfortunately, it’s often the reverse of how many projects get scoped.

This leads to another conclusion: The first project is often not an AI project at all. If nobody can describe a consistent process for how a quote gets built, or if the pricing logic lives in one estimator’s head, or if the order history sits in a system in a form that can’t be queried, then the redesign is the entire job and AI is a later step. Do that work and you’ll sometimes find that a better intake form, one integration, or a documented rule gets you most of the way there. That’s a good outcome, not wasted work.

Jobs had this figured out in 1992 already. The tools may have changed, but the allocation problem remains the same.

Sources