BLOG

Introduction

Dear Evangelists of AI implementation — must you still insist on striking yourselves with your own fist, proclaiming Return on Investment as the measure of all things?

ROI was born in an era when the economic world was simple, assets tangible and the relationship between cause and effect linear. Its intellectual origins trace back to Donaldson Brown, an American engineer–financier who, in 1914, working for DuPont, devised what became known as the DuPont Formula - the first systematic framework for assessing the profitability of industrial capital.

Brown decomposed return on investment into profit margin and asset turnover, allowing factories to evaluate efficiency with the precision of a slide rule and the certainty of arithmetic. A decade later, when DuPont became the principal shareholder of General Motors, Brown transferred his method there, transforming ROI into the very grammar of corporate decision-making.

By the 1950s the formula had become gospel - a universal metric for an age of factories, inventories and predictable markets. It was the perfect instrument for a world of visible assets and stable assumptions. That world, however, has vanished.

From Industry to Complexity – The End of Numerical Certainty

In the age of artificial intelligence, ROI has lost its epistemic authority. AI enters domains that defy quantification: judgement, learning, culture, perception, trust.

It is not an investment in a discrete technology, but an intervention in a complex, adaptive system - fluid, nonlinear, laden with feedback loops. The classical formula rests on a comforting fiction: that one can clearly define a relationship between net gain and the capital expended to achieve it. Yet in AI both profit and cost are dispersed - across functions, over time, and through the organisation’s social fabric. ROI, when applied to such phenomena, is not a measure of precision but an artefact of nostalgia: a relic of arithmetic certainty in a probabilistic world.

Why End-to-End Pilots Mislead the Mind

Many contemporary zealots of AI continue to advocate the end-to-end pilot - a neatly bounded experiment designed to 'prove the business case.' It is an elegant delusion. An organisation is not a cluster of independent processes; it is a living, recursive organism. A change in one subsystem inevitably reverberates through others - often where no metric exists to register the shock. AI’s consequences are systemic, not local.

To test them in isolation is to measure a wave by observing a single molecule of water.

Real life example. You've swapped the counter-productive strategy of burning a large client base with repellent cold-mailing. You replaced it with sophisticated campaigns targeting narrow audience segments, complete with a high degree of interactivity and standardized mechanisms. The result? A nearly threefold increase in orders. Splendid.

But, hold on a moment. Were you aware that production only had 12% spare capacity?

Do you realise the costs you generated for our logistics department? The lack of two-person crews on long-haul routes completely eroded our sales profitability.

And are you aware that the sad young woman in the company car park, the one with a cardboard box of personal belongings at her feet- was a pillar of our Key Customer loyalty process? She was a quiet, yet highly effective generator of significant profit margins, stemming from her ability to maintain warm relationships with the very clients who guaranteed our revenue stability. And were you perhaps a little surprised, that your 'artificially intelligent' chatbot possesses, a slightly weaker intuition when it comes to cultivating those relationships?

The Paradox of Local Rationality

Here lies the enduring paradox. Local implementations often generate data that appear robust yet guide the firm towards error. Projects with low local ROI may yield powerful systemic benefits; those with high ROI may produce silent damage elsewhere. As Kumar et al. (2024) observe, the value of AI arises only in cross-functional synergy - when marketing, operations, finance, HR and analytics interact as one cognitive whole. Isolated deployments, by contrast, misrepresent both risk and reward. To any systems engineer this is axiomatic; to much of modern management, it remains a revelation.

Beyond ROI – Towards Measures of Organizational Intelligence

According to the Canonical Model of Business, post-industrial value no longer stems from singular investments but from the integration of processes, competencies and knowledge. True performance is systemic, not financial. AI should therefore be introduced broadly yet lightly - through a constellation of small, inexpensive initiatives across every domain: technological, market, financial, managerial, human, regulatory and socio-political. Its purpose is not to generate a percentage return but to elevate the organisation’s cognitive agility — its ability to learn, adapt and self-regulate in real time. Such agility is measurable not by ROI but by integrated indicators of effectiveness, efficiency, quality and operational risk awareness. These are the true currencies of post-industrial competence.

AI as a Cultural, Not Merely Technological, Phenomenon

Artificial intelligence is not an IT project; it is a cultural phenomenon that redefines the very grammar of organizational thought. Where industrial technologies multiplied physical productivity, AI multiplies cognitive capacity. It reframes the boundaries between data and meaning, human and tool, decision and reflection. ROI, grounded in the bookkeeping of cost and profit, cannot apprehend this transformation. As Herbert A. Simon (1969) reminded us, organizations are 'artificial systems of cognition' whose effectiveness depends on the quality of information they process. AI does not merely augment such systems - it changes what they are. Its worth lies not in a financial dividend, but in the expansion of awareness and learning throughout the institution.

Conclusion – A New Epoch of Metrics

ROI was the perfect compass for an industrial landscape: clear boundaries, stable conditions, measurable outputs. In today’s terrain - complex, data-saturated, and constantly shifting - it points nowhere. To ask 'Does it pay off?' is to ask the wrong question. The mature inquiry is: Does AI make us wiser? More integrated? More resilient? Only then does it truly 'return.' There is no ROI on a cognitive revolution — only progress, or the failure to evolve.