Why now.
The cost of intelligence is collapsing. This document explains what that changes for your company, with sourced figures.
01
Intelligence costs 280 times less than it did two years ago.
In March 2023, GPT-4 cost $30 per million tokens. Sixteen months later, an equivalent model cost $0.15. The Stanford AI Index measures a 280× drop in inference cost over two years. a16z named it LLMflation: at equal performance, the price falls tenfold every year.
÷280
inference cost over 2 years
Stanford AI Index
÷10 a year
at equal performance
a16z
4% → 72%
success on real engineering tasks in one year
SWE-bench
02
When intelligence is nearly free, your processes become the variable.
Every process that rests on a human reading, deciding and re-keying can be redrawn. At today’s rates, analysing a ten-page document costs about three cents of API. An invoice processed by hand costs $12.88; automated, $2.78 (Ardent Partners).
The question is no longer what AI costs. It’s what each month costs you while your processes stay manual.
$12.88 → $2.78
one invoice, processed by hand then automated (Ardent Partners)
03
Nine companies in ten use AI. Fewer than four in ten see the effect.
88% of organisations use AI. Only 39% measure an impact on their bottom line (McKinsey, 2025). In France, 31% of small and mid-sized firms use generative AI, 17% regularly (Bpifrance). And Gartner predicts more than 40% of agentic-AI projects will be cancelled by the end of 2027: cost, unclear value.
The dividing line isn’t the technology, everyone has the same models. It’s execution. The winners redraw their processes and measure. The losers buy licences and wait.
88% → 39%
of organisations use AI; only 39% measure an impact on their bottom line (McKinsey, 2025)
04
A lead compounds. So does a lag.
A system in production improves with every piece of data it handles: your playbooks sharpen, your teams level up, your data piles up where your competitors have none. Twelve months of lead can’t be bought back with the same tool — you would have to buy back the twelve months of learning too.
Next year is the most expensive line in your strategic plan.
This quarter. Not in three years.
You don’t need an 18-month plan to start. Two weeks of audit. At least three costed use cases. First systems in production within 30 days.