Patrick McGrath writes these. Mostly about the distance between an AI demo that works and a system your team will actually run.
Most players decide to leave in their first sitting, so a churn score that fires after a week of silence reports a decision already made. Three games this year showed us how to find that moment, and how to prove a fix worked.
Every live-ops platform now sells AI, and the prices are public. Buy the part that ships config. Build the part that decides what to ship, because it runs on your game's own history.
A vendor offering to be paid only if the number moves sounds like the fix for paying for hours. Whether it is depends on a definition most proposals leave out.
You can get the readout out days earlier and still lose the meeting to an argument about where the number came from. Speed is the cheap half of an analysis layer, and the expensive half has no screenshot.
The AI that paid off in live ops sat under the calendar, in churn and LTV forecasting, where it puts a decision in front of a person days sooner. The talking NPCs and self-filling calendars vendors pitch first are the wrong place to start.
A Head of AI runs about $500,000 a year once equity is counted. The cases where you should make the hire, and why a firm that sells the alternative would tell you.
Half the game industry thinks generative AI is harming it, and only 7% think it is helping, down from 13%. What that means if you run a studio and still have work to get done.
The most-quoted statistic in enterprise AI comes from 153 survey responses, describes something other than failure, and is now used to end conversations it should start.
Enterprise AI stalls in two places the pitch never mentions: the harness you didn't build, and the team that never picked it up.
Pieces go to a small list first. Nothing else is sent to it, and no cadence is promised.
If you would rather talk than read, that is the better use of your time.