Op-Ed: Getting tough on AI ROI as the real cost benefits hit balance sheets

Return on investments of AI and AI agents is getting much more pragmatic and far less idealistic than it was a year or so ago. The demand is for clear, proven cost benefits. That’s not negotiable, particularly in view of the capital outlay on AI both directly and indirectly.

The ongoing mutterings about AI ROI are pretty much continuous in the news as new AI models and modes of operation take shape. Not everybody is or was convinced by the hype, and the demand for clear information has simply increased. The constant drip of AI ROI related information comes from everywhere from Forbes to just about every possible known peer group input for business operations.

For the sake of brevity, we’ll leave out the costs of awkwardness, failures, and simply getting these systems to run properly. Let’s stick to the fully functional AI event horizon for navigation.

These are legitimate major issues for all sectors and all types of businesses. Given the huge outlay and the gigantic “existential gamble” that AI represents, ROI is the only show in town now.

Saving on human costs doesn’t count now

There are real cost benefits, previously rather cynically and inaccurately qualified by so-called savings on human costs that were selling points way back in prehistoric 2024 when majestic and mystic Pteranodons ruled the business news.

These supposed savings seem to have been left out of the cost of layoffs and other payouts. It’s the wrong metric, and it always was. The cost of lost expertise and experience doesn’t fit on balance sheets too well.  

Even Gemini, when asked to define return on investments of AI and AI agents, says that “while 80% of organizations cut headcount during AI rollouts  workforce reductions do not improve ROI”.That’s pretty unequivocal, and Gemini backs it up with a short summary of why that’s the case.

Note: Gemini is used as a benchmark because it’s easily accessible and readers can ask their own questions directly.

You need to see the Gemini response link above in context with the broader market issues of returns on investment and very basic business best practice. “Best practice” is not a euphemism. It means “Everything even an underachieving idiot shouldn’t need to be told about good business operations, like credible accounts, staying out of jail, etc.”

The short history so far is that, as expected, AI processing costs are down by comparison and that AI vs human interaction costs show much lower costs for AI.

It is history. That’s not working anymore as the costs of the now non-existent staff are hardly a realistic assessment base. It’s the conventional expedient rationale of 2024, not the reality of 2026.

Interestingly, Gemini obviously isn’t too sold on the standard AI costs and revenue on balance sheet mantra itself. Responding to a question about that issue, it cites the dreaded and rightly-much-mistrusted “circular funding and vendor financing” equation in the wider sense of AI ROI, saying it “can distort valuation and cash realization”.

Nice to know Gemini and the entirety of the global market are on the same page, isn’t it? Nobody has ever trusted those circular funding numbers, and nobody ever will. They’re far too much like the notorious in-house payments to yourself that typically crash complex corporate financing.  

This is just the background to how AI ROI is assessed and perceived. The perception translates directly into market valuations, which makes it essential for ROI to prove itself in practice.

The professional accountancy perspective

Meanwhile back on the balance sheets, AI agents are now getting a pretty stern, terse, and demanding environment to work with. Mythology isn’t all that popular on real accounts and nobody gives a damn about ideology. The need is to come up with verifiable metrics, not astrology.

According to Corporate Finance Institute (CFI), measuring AI agent values is now pretty much a strict formal process, with no ambiguities and absolutely no demand for AI hype of any kind.  There’s a reason for that. Outlay on systems has to make financial sense.

In the section “What Determines Whether ROI Is Actually Realized,” CFI makes the critical point that practical hands-on management of AI is essential to the delivery of ROI. They then cite the classic reasons for AI ROI falling short and the need for AI ROI to be “an ongoing management process”.

This wary but necessary outcome is a long way from the hallucination of AI being an automated moneymaker. The management workload has simply shifted to AI operations.

Now, envision if you will, a solitary manager confronted with an incoming asteroid in the inevitable forms of unexpected AI revenue figures or AI costs. This isn’t just a thankless task. It’s dangerous. Let’s not undervalue what “an ongoing management process” can become.

The Age of Cute is over for AI ROI

The novelty of AI has well and truly worn off for real business. Business is well within its rights to demand performance and trustworthy metrics for AI operations at all levels.  “Cute” doesn’t quite cut it.

The good news is that AI implementation, clumsy and incredibly myopic as it has been, has delivered what it was expected to deliver to a reasonable level. Bear in mind that the delivery so far is often assessed on static “before and after” metrics, and that future projections of vast wealth generation through ROI seem rather tepid.

Also bear in mind that this is a very fluid environment. AI has a very limited shelf life. What works now has to be made to work in future in the face of new tech and obsolescence.

AI isn’t the new Messiah of wealth. It’s the caterer.