Simplification is now the funding mechanism for AI

Here is a finding that should change how your next budget conversation goes. Enterprise Technology Research, which surveys around 9,000 technology leaders and sells its work to investors rather than to IT buyers, tracked where AI money is actually coming from. The answer: it is coming out of the budget you already had.

That reframes the whole simplification question. Portfolio rationalization used to be hygiene — worthy, deferrable, the thing you did when the CFO applied pressure. It is now the funding mechanism for the AI work your board is already asking about. You are going to pay for AI out of your existing estate whether or not you do it deliberately. Doing it deliberately is strictly better.

But be careful which numbers you carry into the room

The SaaS sprawl statistics circulating right now are mostly published by companies that sell SaaS management software, and the methodologies diverge wildly.

Similarly, I'd avoid the widely repeated "IT should be about 5.5% of revenue" benchmark. I could not trace it to a primary source. The defensible figure is Avasant's composite median of 2.9% of revenue — and Avasant themselves warn against using the composite for individual benchmarking, because industry dominates everything else. Their data shows financial services running 4.4% to 11.4% while discrete manufacturing runs 1.4% to 3.2%. That spread is wider than any effect of company size. As they put it: there is no factor more important than industry — not size, not geography.

What the honest savings number looks like

Vendors imply 30–50%. The most specific published evidence I can find says otherwise, and it comes with an uncomfortable companion statistic.

Bain reports that while roughly 90% of companies have run technology cost programs in recent years, three out of four missed their targets — and nearly half of those missed by more than 50%. Their documented successes land in a 15–25% range over two to three years: an energy company at 20% over two years through managed-services cuts plus systematic application retirement, a life sciences company at about 15% over three.

Fifteen to twenty percent over two to three years is the number I'd put in front of a board. It is less than the pitch and more than nothing, and it has the considerable advantage of being achievable. Bain also notes the failure mode that anyone who has done this recognizes: the costs creep back, and you are at the well again in a year or two. Rationalization without governance is a diet, not a metabolism.

The concentration finding is the one that changes your work plan

McKinsey's research on technical debt contains a detail that gets less attention than it deserves: in a typical enterprise, 10 to 15 assets account for the majority of the debt.

Sprawl is Pareto-distributed. Which means the target list is short, and the instinct to inventory everything before acting is the wrong instinct. I have watched organizations spend two quarters building a complete application catalogue and never get to a decision. The catalogue was the deliverable, and the deliverable was not the point.

When I led application portfolio rationalization for two large packaging and paper manufacturers, the useful work was not enumerating everything. It was building a capability-based map — what business capability does each system serve? — and then finding where genuine functional overlap existed rather than superficial similarity. Two plants running different systems that do the same thing is consolidation. Two plants running different systems because their processes genuinely differ is not, and forcing it produces a failed program and a lot of angry plant managers.

The real cost of complexity isn't licence waste

Unused licences are the visible cost, which is why they get the attention. They're also usually the smaller number. Two costs matter more:

Cancelled modernizations. McKinsey found organizations in the worst quintile for technical debt are 40% more likely to have incomplete or cancelled IT modernizations. Complexity doesn't just cost you run-rate, it costs you the ability to change — and that cost shows up as a strategic failure, not a line item.

Instability under AI. Google's 2025 DORA research found AI adoption is now positively associated with throughput but negatively associated with delivery stability. A sprawling, poorly understood estate does not get better under AI acceleration. It gets worse faster. Which means the window in which simplification is optional is closing.

The uncomfortable root cause

Gartner's 2026 CIO survey found 52% of technology executives expect cost reduction to grow in importance — but only 33% consistently pursue financial outcomes from technology. Those who do are 25% more likely to be top performers.

That gap is the actual problem. It is not that IT costs too much. It is that two-thirds of technology organizations never connect what they build to a financial outcome at all, which makes every cost conversation a negotiation over invoices instead of a discussion about value. Gartner's own framing is the right one: harvested savings are not the outcome, they are the fuel.

One caution before you start. Bain's data on programs that miss their targets should be read as a warning about ambition, not about effort. The programs that fail are usually the ones that promised a number nobody had tested against the estate. Size the prize after you've looked, not before.