Antonym: The ROI Trap Edition

Eyes on the AI prize.

Dear Reader,

We love numbers. We love charts. And we love to say we’re getting more done, faster.

But focusing too hard on short term gains may blind us to real opportunity, especially when it comes to accelerating our organisations with AI.

ROI: Really Only Imagined?

The Jagged Frontier paper from Harvard Business School came out 18 months ago, and it still holds the best scientifically proven numbers for productivity gains—roughly, a 20% increase in productivity and 40% improvement in quality. By now, logic suggests we should have even greater returns, but there’s a problem: no one is measuring it properly.

Not enough studies have emerged since with repeatable, peer-reviewed studies. This is partly because it is very hard and time-consuming to measure methodically, to have control groups and run experiments in busy corporate settings.

So why aren’t there more success stories from businesses? We heard about Goldman Sachs delegating 95% of the work on S1 filings to AI, but this is a glimpse of a process speeding up, albeit dramatically, not hard numbers that might be regarded as ROI (return on investment).

Leadership teams like “ROI” because it offers a sense of control and an objective measure that can settle disagreements in the top team and act as evidence for investment.

But as Warren Buffet said:

Any business craving of the leader, however foolish, will be quickly supported by detailed rates of return and strategic studies.

So when you are asked for ROI for an AI project, is it scientific rigour (an investment in itself) or a more performative theoretical business case that’s being requested?

ROI calculations often feel like certainty, but as former Bank of England governor Mervyn King and Paul Kay argue in Radical Uncertainty, businesses often retrofit justifications to support their existing decisions rather than truly evaluating impact.

King and Kay offer Edinburgh’s tram project as evidence of this. Consultants predicted profitability, yet it ended up twice over budget and requiring subsidies to stay afloat. ROI models say one thing, reality delivered another. AI investments can fall into the same trap—focusing on efficiency ROI instead of how AI actually helps businesses adapt to complexity.

Two questions a proposer of an AI project might ask in response to “how will we measure the efficiency ROI on x?”:

1. How do we measure productivity now? (It is rarely measured.)

2. How much do we want to spend on the measurement? (What’s the ROI on the ROI!)

The Productivity Sugar Rush

The business case for investing in AI literacy is easy to prove with efficiency gains, but we shouldn’t stop there. It’s a possibility-limiting way of looking at this technology.

AI’s initial productivity boost is like a sugar rush: an instant boost to performance and confidence; but it is only a prelude to the longer term value that will come from AI literacy, governance, and long-term innovation.

Keen readers will remember the AI literacy framework we proposed in the Prepared Minds paper. A simpler version I’ve been using with Brilliant Noise AI-B-C clients is this one, with three levels of AI literacy and corresponding levels of transformation:

The table describes integration of as a three-phase journey: doing what you already do, but better; doing things in new ways; and finding entirely new things to do. In the first phase, AI speeds up everyday tasks, enhancing productivity and quality in areas like writing and reporting. The second phase involves re-evaluating processes to leverage AI capabilities, such as using meeting conversations to automatically generate reports. The final phase is about system-level redesign, imagining new business models, operating models and products that were previously impossible..

This sequence mirrors the development of AI literacy within an organisation. As teams become more AI literate, they move from foundational knowledge and task automation to using AI for innovative workflows and, ultimately, to creating entirely new AI-driven strategies. This journey requires continuous learning and experimentation, developing knowledge and mastery to realise the full potential of AI and gain competitive advantage.

The biggest value from AI two years or more from now will be in business innovation. The new things that people have dreamt up because they have been working with AI and spotted the opportunities. We use a heat map of the value stack to illustrate this idea. Right now, the money has rushed into infrastructure – GPUs, the Nvidia chips powering the construction of the AIs, and the cloud technology they are running on. In the early days of the web revolution the money went to Cisco, whose equipment was building the web, while in the early days of the mobile revolution companies like Nokia and Ericisson were building the foundations. At the start of a revolution we haven’t imagined the places where the biggest value will be created – in the 90s at the start of the web and mobile waves, no one had thought of Google, Facebook, Uber, Spotify etc.

So, where does that leave us? Focusing on efficiency gains is responsible and efficient—for a while–but it is focusing on a base level of progress when we need our heads up and looking where the trajectory is headed. In terms of our diagram, we’re practsing at level one, but should be planning for level 2 and dreaming of level 3.

That’s all for this week

Thank you so much for reading and for leaving likes 🤍 → ❤️. Last week Antonym’s subscriber numbers grew a lot – welcome, new readers! – and the highest number of views for an edition yet. I really welcome feedback and especially the time you take to read this.

All the best

Antony