FOCUS™ and the AI Spend Gap

There is a quiet standard reshaping how technology billing data works, and if you run finance at a company spending real money on AI, it is worth ten minutes of your attention.

It is called FOCUS — the FinOps Open Cost and Usage Specification — an open specification maintained as a Joint Development Foundation project under the FinOps Foundation. The problem it attacks is one every finance team knows by feel: every vendor invents its own billing format. AWS says one thing, Azure says another, every SaaS tool says something else, and before anyone can analyze technology spend, someone has to normalize all of it.

FOCUS replaces that tower of Babel with one schema: common column names, common definitions, and common rules. The FinOps Foundation currently lists AWS, Microsoft Azure, Google Cloud, Snowflake, Databricks, MongoDB, Vercel, and other providers with FOCUS-format support at different versions. That is genuinely useful, and it is important to say so before discussing what normalization does not solve.

The part almost nobody noticed

In May 2025, FOCUS 1.2 added support for virtual-currency economics. Its pricing-currency requirements let billing generators describe provider-defined units such as credits, tokens, and DBUs, their purchase and consumption, and their relationship to national currency.

Read that with a CFO hat on. The industry's shared cost language can now carry AI-consumption units. The people building technology-finance plumbing recognized the same shift behind Agent FinOps: AI spend is becoming a first-class cost category, and it often bills in units that traditional cost tooling was never designed to carry.

Then, on June 4, 2026, the FOCUS Steering Committee ratified version 1.4. That release added Invoice Detail and Billing Period datasets, expanded contract-commitment context, and standardized more of the correction, delivery, completeness, and invoice-reconciliation rules needed by finance teams. The standard is moving normalized cost closer to the finance office.

What is still missing when the data is perfect

Imagine the best case: every AI vendor you use emits clean FOCUS data tomorrow. Every credit, token, seat, and API charge arrives in a normalized dataset with consistent service, account, period, currency, and invoice context.

What do you have?

You have a perfectly organized set of costs that still does not answer all the questions a board or controller asks. Which agent spent this? Who owns it? Which workflow, department, and business outcome benefited? What happens to spend nobody claims? How does it land in the general ledger at month-end? What is the accrual for usage already incurred but not yet invoiced? Can the subledger tie out to the books?

Normalization is necessary and not sufficient. FOCUS tells you, with useful consistency, what was billed and how the provider classified it. It does not — and does not try to — decide internal accountability or turn source data into approved accounting.

That gap between normalized cost and closed books is where the month-end work lives:

• Attribution. A cost row with tags is not yet an owner. Someone must map spend to agents, workflows, departments, and cost objects. Costs that do not map belong in a visible suspense account, not an arbitrary even allocation.
• Chargeback. Finance does not close against a dashboard. It needs balanced journal entries, mapped to the chart of accounts and dimensions, reviewed, locked, and importable.
• Accrual. Usage-based AI services create metered-but-uninvoiced spend at every cutoff. The strongest estimate comes from complete usage through the period, followed by reversal and invoice reconciliation.
• Tie-out. A subledger earns the name by reconciling to source bills and the general ledger. If the numbers cannot tie out, the result is analysis, not a financial record.

FOCUS in. GL out.

This is why Meridian is designed as the accounting layer after normalization, and why a standard billing format makes it stronger rather than redundant. Meridian attributes costs to agents, owners, workflows, and departments; holds unresolved amounts in suspense; and produces chargeback entries, accruals, and reconciliation evidence that a controller can use.

We are building Meridian to ingest FOCUS-formatted Cost and Usage data, starting with version 1.2 and later. The standard handles portable billing vocabulary. The ledger handles accountability and the books.

FOCUS in. GL out. Clean source data enters; reviewed financial meaning comes out.

If your organization already receives FOCUS exports — or your board has started asking what AI work actually costs and returns — a two-week, read-only Agent Spend Assessment can establish the attribution and control gaps without giving Onaro credentials to your systems. You keep the findings either way. See the current source roadmap on the integrations page.

---

FOCUS is a trademark of the Joint Development Foundation. Onaro is not affiliated with or endorsed by the FinOps Foundation or the FOCUS project.

Adapted from the FOCUS Specification by FinOps Foundation, licensed under CC BY 4.0. Changes were made for summary, commentary, and Meridian accounting context.