AI has become one of the fastest-growing lines in the technology budget, and a growing share of it is bought directly from model providers rather than through a cloud marketplace. Those direct contracts are invisible to technology-blind ERPs. The result is a governance gap: material, accelerating spend with no line on the technology P&L and no clear owner at period close.

Ternary now connects natively to Anthropic and OpenAI. Direct AI spend and token usage flow into the same governed data model as the rest of your technology portfolio, so AI can be governed with the same rigor Finance applies to real estate, headcount, and capital.

• Anthropic, every channel - Connect Console API usage and Claude.ai Enterprise usage as separate feeds, each attributed to its own organization.
• OpenAI API - Bring in reported costs alongside model- and token-level usage for every connected OpenAI organization.
• One unified technology P&L for AI - Direct spend lands in the Unified AI Schema next to Claude and GPT procured through AWS Bedrock, GCP Vertex AI, and Azure AI Foundry, normalized by provider, model, version, and token type.
• Finance-grade attribution - Allocate, show back, and charge back direct AI spend through the Billing Rules Engine. Existing Custom Labels apply automatically.
• Budget vs. actuals for AI - Set budgets and alerts on cost or token consumption by model, provider, or cost center, and act before the invoice arrives.

For Anthropic Enterprise, and any AI purchased through AWS Bedrock, spend and usage are also automatically attributed to User Email, extending ROI accountability to the individual consumer. User-level attribution for more providers, including OpenAI, is on the way.

This is Technology Investment Intelligence applied to the most consequential spend category of the decade: judgment infrastructure that puts every AI dollar in front of the CFO, so leaders make better judgments with all the data.

Connect a provider from Admin > Integrations > New Integration, then select Anthropic or OpenAI. Once ingested, data appears in AI Observatory in the left-hand navigation and throughout Ternary.

Dashboard and report subscriptions already put technology spend in front of the people accountable for it. They now arrive ready for the review itself, with the context your team needs to act on the numbers in the email or Slack post itself.

Every dashboard subscription now opens with a financial briefing built on the same Finance-grade attribution behind your unified technology ledger. The charts you already receive, including budget vs. actuals, follow below it.

• Cadence-matched comparison - Each metric is shown against the prior period, aligned to the subscription’s daily, weekly, or monthly schedule.
• Plain-language summary - A written readout of what rose, what fell, and by how much, ready to carry into a financial review.
• Top drivers - The largest movers behind each widget, so variance arrives explained.
• Richer Slack delivery - Channels now receive the same figures, drivers, and charts as the email.
• Clearer filter labeling - Filtered widgets are now marked in both the metric tile and the chart.
• Named subjects - Subject lines now identify the dashboard or report, such as “Weekly Dashboard: Finance Overview.”

Report subscriptions move to the same refreshed layout, with the chart, active filters, and send date up front.

Existing subscriptions will receive the upgraded experience automatically. To add a new one, open any dashboard or report and select Subscribe from the three-dot menu.

Chargeback is only as credible as the labels beneath it. Spend without a department, product, or budget code sits outside the unified technology P&L and outside anyone’s accountability. AI raises the stakes further as boards want to know what an AI investment returns, and answering that requires cost and revenue in the same frame.

Ternary’s MCP server now answers both questions in plain language, from the AI assistant your team already uses. ERPs are technology-blind and FinOps tools are Engineering-owned. These two new tools extend Ternary’s judgment infrastructure, built for Finance, into every conversation about technology investment:

• Label coverage: See how much spend any Custom Label covers over a period, with each provider ranked by the size of its gap.
• Unallocated spend, located: Pinpoint the 20 largest account and service pairs carrying unlabeled spend, so remediation starts where the dollars are.
• Label hygiene: Surface likely misspellings, placeholder values such as “unknown,” values outside your approved list, and spend missing a required companion label.
• AI ROI: Weigh AI cost against revenue for the whole tenant, or by product, team, or any Custom Label.

Every figure is period-close compatible. Unallocated spend is defined exactly as a “not set” filter in the Reporting Engine, so the numbers reconcile to your reports. AI cost comes from Ternary’s own classification of each billing row, not from service names or tags. When revenue is missing for the period, Ternary reports AI cost alone and names the gap rather than estimating a ratio. That is finance-grade attribution, held to the same rigor as every other line of the P&L.

Connect the Ternary MCP server to your AI assistant and ask, for example: “How much of last quarter’s spend has no department label?” or “Show AI cost against revenue by product.”

Finance plans, forecasts, and reports live in spreadsheets. The Ternary add-on for Google Sheets brings Technology Investment Intelligence directly into that workflow. Budget models, variance analyses, and board materials now run on finance-grade attribution, not pasted values.

• Import any report. Pull any Ternary report into the open sheet as a block.
• Refresh in place. Update one block or all of them, and your surrounding formulas, notes, and formatting stay exactly as you left them.
• Ternary data with your formulas. Reference live report data in individual cells, so figures stay current through every forecast cycle and period close.
• Scoped access. The add-on reads only the spreadsheet it is open in, and your API key goes only to Ternary's API.

Add-on or MCP: same data, different destination.
The add-on is deterministic: the complete report lands in cells, identical for every user and refreshable on demand. You can also use Ternary's MCP to help you build and save a new report and then import it with the Sheets add-on for complete, audit-ready numbers.

Open any Google Sheet → Extensions → Add-ons → Get add-ons → search "Ternary", then connect with your Ternary API key. Or install from the Google Workspace Marketplace.

Since launch, the Ternary MCP server has let Finance and FinOps ask any AI assistant questions against live technology P&L data. No more exports or stale CSVs. That was read-only by design: the assistant could see everything you could see, and nothing more.

Now it can do something with what it finds. When an assistant surfaces an answer worth keeping, like a budget at risk or a cost driver behind a dashboard swing, it can save that query as a report directly in Ternary, so it's there for the next period-close review without you re-create it by hand.

What's new:

• Save reports from a conversation. Ask a question, get an answer, and tell the assistant to keep it. It's saved privately to your account and appears in your saved reports, ready to reference, share, or build a dashboard on. 
• Savings recommendations, surfaced on demand. The assistant can now pull optimization recommendations from Ternary's Optimization Hub. Get idle resources, rightsizing, and other savings opportunities ranked by estimated monthly savings, filterable by provider. The same view you see on the Recommendations page, now answerable in a sentence.

Same governance model as before: OAuth-authenticated, permission-scoped to exactly what the requesting user can already see. No new access, no new risk surface, just a shorter distance between a question and an ROI decision.

Access: Already connected? These tools are available now. Otherwise, learn how to connect your AI assistant.

AI spend is growing fast, and most of it arrives unattributed. Your business needs to go deeper than the top-line total, understanding which team, project, or product is driving that cost. AI provider invoices don't carry that detail on their own, but your LiteLLM gateway already does: every request is priced and tagged by team, key, and model.

Ternary now documents a direct path to bring that gateway detail into your unified technology P&L. Point LiteLLM's built-in FOCUS export at a bucket you control, ingest it through Bring Your Own Data, and pair it with your AI provider invoice in the Billing Rules Engine: the invoice supplies the dollars, the gateway supplies the proportions. The output is a zero-sum reallocation of real invoiced spend across teams and cost centers that is reconciled to the bill, not estimated.

• Team, model, and provider breakdowns - report on AI spend by team, model, and provider, alongside your other cloud costs.
• User- and key-level granularity - map LiteLLM's Tags field to break down by user and API key with no extra tagging work.
• Invoice-true reconciliation - Reallocate Costs billing rules split invoiced AI dollars using gateway proportions.

Requires a self-hosted LiteLLM proxy (v1.98.0+) writing its FOCUS export to a GCS or S3 bucket.

Learn how to get started with our LiteLLM AI Gateway guide.

AI assistants are now the first place people go to ask questions about their business, including questions about AI and technology spend. The problem comes when someone asks "what did we spend on AI last quarter," the assistant either guesses, or someone pastes in a CSV that's already stale by the time it's answered. That's a governance gap: judgment made on secondhand data, outside any controls, outside your permission model.

The Ternary MCP server closes it. Connect Claude, ChatGPT, VS Code, or any MCP-compatible assistant directly to your live technology P&L. No more exports, copy-paste, or shadow spreadsheets standing in for the system of record. Data spans every Custom Label, Custom Measure and provider integrated into Ternary. Ask questions of your data from Anthropic, AWS, Azure, GCP, Oracle Cloud, Alibaba Cloud, Databricks, Snowflake, MongoDB Atlas, OpenAI, and any custom source brought in via BYOD.

• "What did we spend on BigQuery last quarter, broken out by project?"
• "Is the Q3 infrastructure budget on track, at risk, or over?"
• "What moved on the Engineering dashboard this week, and which services drove it?"
• "Pull the saved AI Observatory report and show me token spend by model family."

What it does:

• Live cost queries - The assistant discovers your real dimensions and measures, then runs queries against FOCUS-normalized, finance-grade attribution - the same data model behind every report in Ternary.
• Saved reports and dashboards, on demand - Pull existing reports or ask what changed on a dashboard week-over-week, with drivers surfaced automatically.
• Budget status, instantly - Spend to date, projected month-end, and health status.
• Governed by design - OAuth-authenticated, read-only, and permission-scoped to exactly what the requesting user can already see in Ternary. No new access is created.

Learn more on how to connect your preferred AI assistant here.

Every budget has its own narrative. By recognizing the details early on, we can transform a variance into an informed decision rather than an unexpected surprise.

The budget detail page has been rebuilt from a simple status check into a full analytical workspace. 

Where you once saw three charts, you now get a complete view built for real financial review:

• Spend vs. budget analysis - See performance against budget with the depth needed to explain variance, not just flag it.
• Projections - Forward-looking spend trajectory using the latest AI/ML models, so you know where a budget is heading before period close, not after.
• Variance drivers - Surface the specific changes in spend responsible for variance, down to the cause.
• Linked reports and materials - Attach reports from the Ternary Reporting Engine and outside materials directly to a budget, so context lives where the review happens.
• Configuration, visible in place - Review how a budget is built without leaving the page, giving all users the context of the budget.
• Flexible time horizons - Filter the entire view across any time period your review cycle requires.

This extends across the hierarchy: search finds any sub-budget instantly, and both budgets and sub-budgets can be renamed to match how Finance actually talks about the business.

Alerts now go deeper, too. Users can now subscribe to sub-budgets directly, and see every subscriber and threshold set across a budget and its sub-budgets in one place.

This is judgment infrastructure for technology investment. A budget page designed to meet the level of scrutiny your business needs to make informed decisions.

To access the new view, select Budgets from the left-hand navigation, then select any budget to explore the new detail view.

Every report contains signal and noise. The ability to separate them quickly is what turns a data table into a decision.

Limit Displayed Results brings a new level of precision to the Ternary Reporting Engine. Now available directly in the filter bar, you can limit displayed results to exactly the rows that matter. You can now filter on any measure in your query, not just the top one.

• Show Top N - surface the highest or lowest contributors across any measure
• Greater than, less than, or equal to - zero in on rows with totals that meet a specific threshold, filtering out everything that doesn't

Whether you're isolating the ten most expensive AI models, finding every product line where unit cost exceeds a target, or stripping out negligible spend to focus a board presentation, the Limit Displayed Results filter puts that control directly in the report, without post-processing in a spreadsheet.

Limit Displayed Results is available now in the filter bar of any report.

AI spend is not a single line item. It's Anthropic models running via AWS Bedrock. It's Azure OpenAI alongside direct OpenAI API calls. It's Gemini, Grok, Qwen, and a dozen other models procured through different channels, billed in different formats, and invisible to any single reporting surface.

AI Observatory is Ternary's global dashboard for AI spend and token usage. A single, governed view across every AI provider, vendor, model family, and version in your technology portfolio. This dashboard will be consistently updated to reflect the latest models.

Unified AI Schema Powering Observatory is a provider-agnostic framework that normalizes AI-specific data from every source into a standardized schema. This enables consistent querying, reporting, and analysis regardless of where the spend originates. Every record is attributed by provider, publisher, modality, input/output tokens, model, and version. Claude, ChatGPT, OpenAI's reasoning models and Whisper, Moonshot Kimi, xAI Grok, Alibaba Qwen, Zhihu GLM, MiniMax, Amazon Nova and Titan are all in one place and in the same format.

What you can do with it

• Cloud Provider AI Analysis - Break down AI spend procured via GCP, AWS, and Azure, with vendor and model-level attribution across Anthropic, OpenAI, Google, Meta, and more.
• AI Guardrails - Create budgets and alerts around token usage and cost by model, provider, or any dimension. Control what happens at the point of consumption, not after the invoice arrives.
• Allocate, Showback, and Chargeback - Use Ternary's Billing Rules Engine to attribute AI spend directly to the teams, products, and cost centers that incurred it.
• Custom Labels - Existing label rules apply to AI spend automatically. Create new labels to organize AI investment in ways that map to your business.

A Starting Point, Not a Ceiling. Ternary's AI Observatory is designed as a launchpad. Every report underlying the dashboard is available in the Ternary Reporting Engine, ready to be filtered, grouped, and customized to reflect the specific dimensions, models, and cost structures that matter to your business.

AI Observatory is available now. Navigate to AI Observatory in the left-hand navigation to get started.