The AI stack is splitting into four layers: infrastructure, observability, metering/billing, and monetization intelligence. The first two are crowded. The fourth — why margin dropped and what to do next — is where FluxMeter now ships product, not just narrative.
Billing tells you what happened
OpenMeter, Lago, Metronome, and Stripe excel at usage → invoice. Dashboards excel at traces and latency. Neither answers: why did AI cost rise 40%? Which customer loses money? Would switching models save six figures?
Public signals align: FinOps Foundation reports 98% of teams managing AI spend in 2026; executives cite budget blowouts in Q1. The gap is heuristic margin exploration on metered data — not another meter alone.
What FluxMeter Intelligence ships (3.0–3.1)
GET /intelligence/root-cause decomposes spend deltas with rule-based drivers. GET /intelligence/unit-economics flags negative-margin accounts when revenue is overlaid (product revenue uses a cost-share heuristic). POST /intelligence/simulate runs model-switch and promo what-ifs. v3.1 adds rule-based pricing recommendations, profitability views, forecast, alerts, and report export.
Metering and guardrails remain Pillar A — check, reserve, Gateway proxy (3.2). Intelligence reads the same rollups; it does not replace enforcement.
See the full map
The four-layer market map — players, maturity, and FluxMeter positioning — lives at /market-map. Compare pages explain how we complement OpenMeter and Lago rather than replace them.