AI economics, from inside an operating business
Twenty years taking cost and time out of transaction-heavy industries — US mortgage and property management, industrial safety, B2B software, Indian real estate.
Most of what is here is my own operating data, including the launches that missed.
Standing balances
Verified weekly
Nearly 9 in 10
of organisations regularly use AI in at least one business function
McKinsey, The State of AI 2026
37%
attribute at least some EBIT impact to AI, about the same share as last year
McKinsey, The State of AI 2026
6%
are high performers: 5% or more of EBIT from AI, and significant value
McKinsey, The State of AI 2026
63%
of digital firms report AI adoption, against 16% of non-digital firms
Smeets and Olarte Rodriguez (IFC), 'AI Adoption Among Frontier Firms', WDR 2026 background paper
Above 16 hours
frontier 50% time horizon at METR's last update in May; METR says its suite is unreliable above that
METR, Time Horizon 1.1
Most recent entry
All entries →Everything up to the commitment ran on AI. I'd have bet against that.
Going into last year's launch I was sure the AI could take the information work, and unsure it could carry the decision. It carried both, all the way to the booking.
Earlier
The accounts
Being believed costs money. Most companies file it under something else.
Digitisation is the binding constraint across most of the world, long before intelligence is.
Cost per workflow, payback period, and the invoices behind both.
Where the line falls inside a job when part of it moves to a machine.
Evidence standards for AI claims, including the ones I make here.
Procedures you can run on a Monday morning. Limitations included.
The dispatch
What changed since the last issue, and what has stopped being safe to quote. Every figure carries its source and the date it was fielded.