Tokenomics

Tokenomics is Finitizer's module for AI spend. It answers one question in order: who is accountable for AI, what it returns, and what it costs. The module is in Beta.

Why a bill is not enough#

AI spend arrives through several doors at once: provider APIs, cloud token services such as Bedrock and Vertex AI, tools with AI built in, rented GPUs and hardware you own. Each bills differently, and none of them names the workload the money served.

Seeing that spend clearly does not, on its own, tell anyone what it returned. In the Tokenomics Foundation's State of Tokenomics survey (September 2026, 472 respondents), three in four enterprises are not, or only slightly, confident they can connect AI spend to an outcome their CFO would accept. Proving value was the challenge named most, by 43%. Of those already moderately confident in their spend visibility, 60% still cannot show an outcome.

The module is built around closing that gap: spend grouped into owned workloads, each workload's outcomes and value beside its cost.

Survey figures: Tokenomics Foundation, State of Tokenomics, September 2026, licensed CC BY 4.0.

Where the data comes from#

Everything is computed from one set of rows, read from sources you already have:

• Cloud bills Finitizer already reads. Gemini and Vertex AI from the Google Cloud billing export, and Amazon Bedrock from AWS. No new credential.
• Nightly pulls. Claude Enterprise, OpenAI and the Claude Console, with a key an administrator stores. The key is sealed on the server and never shown again.
• Exports you import. Twenty-two formats, including OpenAI and Anthropic usage and cost reports, Azure cost exports, AWS cost and usage reports, FOCUS files, OpenRouter activity, LiteLLM spend logs, Claude Code, GitHub Copilot and Cursor reports, and a Databricks system-table query.

Every dollar carries its basis: from an invoice, a share of one, or a list-price estimate where a feed has tokens but no dollars. Files are parsed in your browser and only the rows are stored. Prompts are never stored.

The seven screens#

Overview#

Who answers for the spend, what AI returns, and what it costs this month: the person accountable for AI economics, how much spend sits in a named workload with an owner, value booked per AI dollar, and spend against the monthly budget with a projection and its range.

AI Value#

What each workload returns. Outcome metrics (tickets resolved, pull requests merged, units sold) give a cost per outcome. Value claims follow the Tokenomics Foundation's chain from an observed change to a dollar result, and only finance, admin and super admin can book one, once, on evidence they name. Total cost sets each workload's token spend beside the other costs it carries (platform and people, licenses, vector databases, evaluation, observability, GPUs), so a return can be read against the whole cost. Customer-facing workloads get their margin after AI.

Efficiency & Routing#

How well the tokens are used: cache hit rate and whether caching pays, batch and tier use, the model mix on the survey's Frontier scale, spend behind a router or gateway, why the bill moved (volume, mix or rate), what a workload would cost on an open-weight model, and, for models you serve yourself, cost per token beside the API price with utilization, energy and carbon.

Forecast & Planning#

The next three, six or twelve months with a range from backtests, against the budget and by workload; the drivers of each workload's cost; scenarios priced against the budget; and a record of how each month's forecast did.

Data sources#

Where the rows come from, which of the six channels of AI spend your data covers, what each feed can show, and who owns each spend source.

Cost Allocation#

Showback by team, workload, person, provider, channel, model, project, key, source, tag, region or environment. Workloads are named here with an accountable owner, a stage and a sunset date, and allocation rules assign new spend to them.

Budgets & Governance#

Budgets whose status comes from the projected month end, and six policies that find what a budget cap cannot: models outside an approved list, experiments past their sunset, unowned spend, new keys spending, runaway loops and Frontier share over a cap. Developer AI reads Claude Code, Copilot and Cursor against their seats and by team. Alerts go to email, a Slack channel per budget, or Jira.

Who sees what#

Viewers read every screen. Engineers and above can change things. Finance, admin and super admin book value and see spend and activity by person; everyone else sees the same totals with people counted, not named. Administrators manage provider keys.

What it cannot see#

The module says so on the page rather than reading a gap as zero: a channel with no data, a model with no price, a cost month not yet read. Edge and desktop models send nothing to read. A forecast cannot see a launch or a contract nobody entered. Finitizer never changes a prompt, a model or a provider setting, and a budget here does not stop spend: a hard stop is set at the provider.

Where to start#

1. Name the person accountable for AI economics on the Overview.
2. Pull the cloud bills and bring in the provider exports on Data sources.
3. Name the workloads on Cost Allocation, each with an accountable owner.
4. Add one outcome metric per workload on AI Value, then record the first claim for finance to book.