[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"article-body-the-day-100-ai-bill":3},"\nEvery enterprise AI programme has a day 100. For the first hundred days\nthe bills are absorbed by enthusiasm: the pilots are strategic, the\ntokens are cheap individually, the platform fee is an investment. Then a\nfinance partner does what finance partners do, and asks two questions\nthat end the honeymoon. What does this cost us per decision? And what\ndoes a decision return?\n\nFinance teams are converging on a name for this discipline, AI unit\neconomics, and it is overdue, because very few organisations can answer\neither question. The\ncosts arrive as infrastructure line items: tokens, seats, GPU hours,\nvendor subscriptions. The returns arrive as anecdotes and screenshots.\nBetween the two sits no ledger of any kind.\n\n## The unit is the decision\n\nThe instinctive unit of AI accounting is the token or the seat, because\nthat is how the invoices are denominated. Both are the wrong unit, in\nthe way that \"cost per kilowatt\" is the wrong unit for a factory. The\nthing the spend exists to produce is a changed decision: a forecast\nsomeone acted on, an exception someone resolved, an intervention chosen\nover a default.\n\nDenominate in decisions and the questions become answerable, and\nuncomfortable:\n\n- **Cost per decision.** Total programme cost divided by decisions\n  actually influenced. A copilot with heavy usage and no decision\n  attribution has an infinite cost per decision, which is the honest\n  number for a great deal of current spend.\n- **Return per decision.** What did acting on the system's call earn\n  against not acting? This requires the outcome to be recorded against\n  the decision, which is bookkeeping most programmes never built.\n- **The substitution line.** When is an agent cheaper than an analyst\n  hour, and when is it an expensive way to generate work the analyst\n  must then check? Without a cost-per-decision figure, the debate runs\n  on vibes in both directions.\n\n> Gartner's finding that [45% of CFOs' AI investments lean toward\n> productivity while 20% lean toward decision quality](https://www.gartner.com/en/newsroom/press-releases/2026-07-20-gartner-survey-shows-45-percent-of-cfos-say-their-ai-investments-lean-towwards-productivity-while-20-percent-say-these-investments-lean-towards-decision-quality)\n> is a unit-economics verdict: cost per task compressed, cost per\n> decision unexamined.\n\n## Why the books were never kept\n\nThe missing ledger is not an oversight. It is a structural consequence\nof how programmes are set up. Pilots are scoped to demonstrate\ncapability, not to attribute outcomes, so the measurement is bolted on\nafterwards, which means never. MIT NANDA's State of AI in Business\nresearch found 95% of pilots showing no measurable profit and loss\nimpact, and\nthe operative word is measurable: much of the missing value is not\nabsent, it is unattributed, which for a CFO is the same thing.\n\nThe repair is unglamorous. Route the AI's outputs to named decisions\nwith owners. Record what was recommended, what was decided, and what\nhappened. Meter the costs against the same decision records. From that\npoint the day-100 conversation changes character entirely: the\nprogramme has a cost curve, a return curve, and a defensible answer to\n\"why should this budget survive?\"\n\nThe organisations that build this ledger first will also, not\ncoincidentally, be the ones whose AI improves fastest, because a system\nwhose decisions are scored is a system that can learn which of its\ncalls deserve trust.\n\n*Enthusiasm funds the first hundred days. Only unit economics funds the\nnext thousand.*\n\nProphesee keeps those books by construction. Every prediction is wired\nto an owned decision and every decision to an outcome, so cost and\nreturn per decision are queries, not archaeology. To start the ledger\non your own data, [start here](/contact).\n",1786786820104]