A ledger of what AI costs and returns
Twenty years taking cost and time out of transaction-heavy industries — US mortgage and property management, industrial safety, B2B software, Indian real estate.
What gets published here is mostly first-party, and it includes the entries that did not work.
Standing balances
Verified weekly
89%
of organisations use AI in at least one function
McKinsey, Aug 2026
37%
can attribute any EBIT impact to it
McKinsey, Aug 2026
6%
report it mattering to the bottom line
McKinsey, Aug 2026
63%
AI adoption among firms with digital foundations, against 16% without
World Bank, Jul 2026
~70m
frontier agent horizon at 80% reliability, not the 12h headline
METR, May 2026
Most recent entry
All entries →On why the AI question in emerging markets isn't about AI
A World Development Report background paper, published at the end of July, went looking at 1,563 companies across 83 countries, all of them drawn from IFC's private-equity and venture portfolio — the kind of firm…
Earlier
The accounts
What it costs to make a stranger believable, and where that cost sits in the accounts.
Why the AI question in low-digitisation economies is rarely about AI.
What deployments actually cost and return, published with the numbers.
Which half of a job moves to the machine, and which half does not.
What would count as evidence that any of this worked.
Frameworks and procedures, with their limitations stated.
The dispatch
What changed, what became unsafe to quote, and the lines worth using on a stage. Every figure carries its source and the date it was fielded.