We’ve moved through two AI eras fast.
2010–2022 was the Data Era. Whoever had the most data won. 2022–2025 was the Compute Era. GPUs became the battleground. Today we’re in a third era and most executives haven’t caught up to it yet.
The Token Era. And it changes the economics of everything.
The Math Most Leaders Are Ignoring
When enterprises deploy AI agents at scale, millions of interactions become billions of tokens. Billions of tokens become millions in monthly operating costs. The question has shifted from “Can we build this?” to “Can we afford to run this?”
Traditional Tech ROI: Revenue − Cost = Return
AI Value Efficiency = Business Outcome ÷ Total AI Operating Cost
Here’s what that it looks like in practice:
The companies winning with AI aren’t deploying the most models they’re extracting the most value per token.
We’ve moved through two AI eras fast.
2010–2022 was the Data Era. Whoever had the most data won. 2022–2025 was the Compute Era. GPUs became the battleground. Today we’re in a third era and most executives haven’t caught up to it yet.
The Token Era. And it changes the economics of everything.
The Math Most Leaders Are Ignoring
When enterprises deploy AI agents at scale, millions of interactions become billions of tokens. Billions of tokens become millions in monthly operating costs. The question has shifted from “Can we build this?” to “Can we afford to run this?”
Traditional tech ROI: Revenue − Cost = Return
AI Value Efficiency = Business Outcome ÷ Total AI Operating Cost
Here’s what that it looks like in practice:
Company B wins not because of better models. Because of better economics.
Why Most AI Programmes Stall
85% of enterprise AI pilots never reach production. The usual suspects model quality, employee resistance, immature tooling gets the blame. But the real culprit is a measurement failure.
Executives celebrate copilots deployed, agents created, use cases launched.
They should be tracking cost per decision, cost per customer interaction, cost per revenue dollar generated.
Activity is not transformation. Token consumption is not value creation.
The New Priority Stack
The best AI leaders have reordered their agenda:
1. Outcome first. Not “improve CX” but “reduce Tier 1 support response time below 20 minutes at under $12 per interaction.” Specificity is what makes measurement possible.
2. Data quality before model sophistication. Poor data quality costs enterprises an average of $12.9M annually (IBM, 2024) a figure that compounds when AI operates at speed on bad inputs. No frontier model fixes a broken data foundation.
3. Govern token spend like cloud spend. AI FinOps is now a real discipline. Leading teams set per-workflow token budgets, track cost-per-outcome dashboards, and apply prompt engineering to eliminate token waste.
4. Compute last. More GPUs don’t fix a broken business case. Infrastructure optimisation is the final layer, not the first.
The Question Boardrooms Will Ask Next Year
2023: “What’s our AI strategy?” 2024: “What’s our AI return?” 2025: “What’s our AI margin?” 2028:?
That last question will separate the winners.
The organisations that win won’t be the ones who deployed the most, they’ll be the ones who extracted the most value per token spent.
The race isn’t about intelligence. It’s about efficiency.
Start tracking what matters:
- Cost per outcome
- Revenue per token
- AI margin
How is your organisation measuring AI value beyond activity metrics? Drop your approach in the comments.
