The story
Everyone is spending on AI. Almost no one can say what it bought.
Token budgets are being written into job offers. Consumption leaderboards are celebrated. And the honest answer to "what are we getting for it?" is still a usage chart. We have run this play before — cloud adoption measured adoption instead of efficiency, and by industry estimates roughly a third of cloud spend went to waste. TIER is the meter for the yield, not the burn.
The problem
The bill is precise. The return is a guess.
A finance team can tell you the AI line item to the dollar. Ask what that dollar produced and the room goes quiet — because nothing in the stack was built to answer it. The spend is measured. The yield is invisible.
The spend is real, the yield is invisible
Engineering leaders keep saying the same thing: "We gave everyone access. We have no idea what it's producing." The invoice is exact. The return is a shrug.
Every dashboard measures burn, not yield
Vendor consoles, usage exports, spend reports — they all answer "how much did we consume." None of them answers "what did the consumption buy."
Budgets are set on a number nobody can defend
Renewals, per-seat vs. per-token decisions, next year's allocation — all negotiated against a consumption chart, because there is no efficiency figure to negotiate against.
Without a yield number, you cannot even see waste
Spend that produced nothing looks identical to spend that shipped a feature. A team burning 4x the budget could be doing 4x the work, or leaving the lights on. The chart cannot tell you which.
The precedent
We have run this play before.
Cloud adoption in the 2010s measured the wrong thing. Companies tracked adoption — "what percent of workloads are in the cloud" — instead of efficiency — "what do we get per cloud dollar." The bills climbed with no change in revenue, and the reckoning came later, one over-provisioned account at a time.
= roughly a third of the spend wasted before anyone noticed
~30%Token spend is the same curve as cloud, earlier and steeper. The difference: this time there is a way to measure the yield from day one, instead of clawing back the wasted spend years later.
The cloud-overspend wedge
The reframe
Stop measuring burn. Measure yield.
The fix is not another usage dashboard. It is changing what sits on top of the ratio — replacing "how much did we spend" with "what did the spend produce," in dollars, per team, over time.
A usage dashboard cannot tell a team that shipped an SSO integration on $300 of inference from a team that burned $3,000 iterating on code that got reverted. TIER is exactly that distinction, computed.
The name is the thesis
TIER — Token Impact & Efficiency Ratio
A double meaning, on purpose. It is the ratio — impact over efficiency, shipped outcomes over dollars, computed and re-derivable. And it is the tier — one comparable grade for how well spend becomes real work, so an organization can finally compare on efficiency instead of usage.
Where this sits in history
The fourth attempt to measure engineering — built to survive the pattern.
Lines of code measured typing. Story points measured estimates. Token dashboards measure burn. Every prior metric put the wrong thing in the numerator, and every one pointed at individuals was gamed and abandoned. TIER puts shipped, quality-weighted outcomes over dollars — and grades the work, not the worker.
Read the 60-year lineage — LOC, function points, velocity, DORA — and why TIER is different →