Why AI investments compound rather than depreciate

One of the more insightful pieces I’ve read on AI strategy recently is from Baba Prasad, a professor of leadership at Brown University’s School of Professional Studies in the US. In a piece for HBR he identifies five types of AI investment, with an analysis of their financial measures and impact.

I won’t parrot Professor Prasad’s findings, which you should read in full, but I do want to share the main insight I gained: that three of his five types of AI approaches result in compounding assets, not depreciating investments.

Conventional enterprise software is worth the most on the day it goes live and declines from there. But what we can build with AI, data flywheels, deep integration into proprietary workflows, and organisational capability, all run the other way. They’re worth more after two years of use than after two months, because the accumulating asset is proprietary data, embedded processes, and human fluency rather than the tool itself.

That reframes the two-to-four-year AI return figure from consultancies like Deloitte. It’s long not because AI is slow to install, but because you’re accruing value. This means the honest board metric during years one and two is the rate of asset accumulation. Is the data moat deepening, is workflow getting genuinely rewired, is fluency spreading, rather than short-term, incremental margin.

Capability building is the sharpest case he makes, being simultaneously the least measurable and the most durable of the five. It also happens to be where other evidence keeps landing. Microsoft’s 2026 Work Trend Index attributes roughly two-thirds of AI impact to organisational factors rather than individual behaviour, Info-Tech found firms with a governed AI strategy three times likelier to report measurable impact, and in aibl’s own study of the UK’s mid-market, mature AI governance had the strongest relationship with positive ROI.

Prasad’s framework gives these findings a place in the capital-allocation conversation instead of leaving it strictly in the change-management arena. Don’t take my word for it. Give it the full read.

Richard