Companies are spending heavily on AI while many still cannot show how it has improved profit, productivity, or the products they sell. Technical capability alone does not create a return. A return depends on whether the technology solves a real business problem at a cost that makes the investment worthwhile.

Irving Wladawsky-Berger is a research affiliate at MIT and a former IBM executive who spent nearly four decades helping businesses adopt major technologies, including leading IBM’s Internet Division. He saw firsthand that computers and the internet created substantial business value only after companies moved beyond experimenting with the technology and changed how they operated. That experience leads him to believe that AI investment will follow the same pattern.

Computers had been used in business for decades when economist Robert Solow observed that their effect could not be seen in productivity figures. Companies had computerized individual processes, but finance, manufacturing, inventory, and other functions still operated in isolation. It wasn’t until those functions were connected through shared systems that an increase in productivity was observed. Productivity increased because the connection of the systems changed the way information moved across the organization.

The internet extended that change by allowing information to move beyond the organization and reach customers directly. Package tracking previously required a customer to call an operator, who searched the company’s database and reported the result. Giving customers access to that same database through the internet removed the operator from a routine inquiry, reduced the cost of providing the service, and allowed customers to retrieve the information whenever they needed it. The value came from solving a defined business problem with a result the company could see in both its costs and customer satisfaction.

At IBM, Irving Wladawsky-Berger saw businesses begin with applications such as package tracking, which improved one defined process, before using the internet across multiple systems and business functions. He believes many businesses are skipping that progression with AI.

Instead of looking for the simplest business applications to implement, they look for the most complicated ones to show the power of AI.

Agentic AI is a very good example of the approach Irving Wladawsky-Berger describes. Instead of using one tool to solve a defined business problem, companies implement several AI systems that divide up the task and act with limited human direction. This adds cost and risk while making it harder to identify which part of the system is underperforming or causing a failure. The complicated system may show the power of AI, but it has made the solution harder to manage and may not create additional business value.

Across the computer, internet, and AI eras, Irving Wladawsky-Berger’s argument remains consistent: technical capability becomes commercially useful only when it improves the business enough to justify the investment.

Melissa’s Take

I use AI in my own writing because I can spend far too much time rereading a piece, correcting punctuation, and changing it again. When the tool reduces that work to one or two passes, I can measure the benefit in time returned to me. I still make the decisions about the message, and the tool removes work that was consuming time without improving the underlying idea.

The same level of clarity should be required when a company commits money to AI. Leaders should be able to name the problem, explain how the company handles it now, and identify the result they expect to change. Without those details, the team may become very busy implementing AI without knowing whether the investment improved the business.

My gateway action is to take one AI proposal currently under consideration and write down the business problem, its current cost or performance measure, the expected improvement, and the full cost of deployment. If the people proposing the project cannot provide those four pieces of information, send it back for more work before approving the investment.

This is part of a fuller conversation with Irving Wladawsky-Berger about why major technologies take time to spread through the economy, what the computer and internet eras teach leaders about AI deployment, the economics of open and proprietary models, the applications most likely to succeed in the near term, and the kind of leadership required to turn technical capability into business value.

The Full Executive Brief goes deeper into every conversation on the podcast, with the context and the follow-up you won’t get from the audio alone. Subscribers also get access to the Maverick Test, a short diagnostic built to show you where your own leadership has gaps you haven’t named yet.