Organization / Governance data
How governance numbers work
What spend, waste, output, and audit views mean — and what they deliberately do not claim.
Spend
Three kinds of cost, and why they are not added up the same way
Spend on this platform is not one number with one meaning. Three kinds sit behind it, and collapsing them is how a spend report starts lying.
Usage is metered consumption — tokens and calls, attributable to a model, a tool, a team, and a person.
Research is purchased and then recognised over time. This is the one that surprises people: the Research tab lists purchases incurred in the window you selected, which is not the same as the research cost this period carries. A purchase made months ago and still being spread contributes to this period without appearing in that list at all. For what a model has cost this period — usage plus recognised research — read the Models tab, not the Research tab.
Committed subscription floors are contracted at the connector or account level. Telara shows them separately and does not divide them across models, because the contract does not. Any per-model split of a floor would be an invention, and it would be the number someone later quotes in a budget conversation.
So: a model’s cost is its usage plus its research. Floors sit beside that total, never inside it.
Spend granularity
Daily feeds ≠ individual runs
Platforms that only expose daily or bucketed usage never show as individual runs. Platforms marked Per-request / Per-run report real run-level cost and outcomes. On platform lists, success and latency show as n/a when runs are not visible.
Spend vs output
Correlation, not attribution
Output metrics (PRs merged, suggestions accepted, and similar) are productivity proxies joined to spend by person and week. The view shows correlation only — it never attributes a specific output to a specific AI run. Scores normalize within the current view, so they are only comparable inside that chart.
Waste inbox
Estimates, not guaranteed savings
Rows come from per-person waste scores (acceptance vs cohort, abandoned sessions, BYO-key share). “Recoverable” is spend × waste score — an estimate, not a guaranteed saving. Dismissing clears a row for this session only until durable workqueue actions ship.
Audit
What the trail covers today
Live today: full-text search, filters, and CSV export over Telara’s platform audit trail — MCP sessions, tool calls, and action gates (CSV is the current page). SIEM streaming, saved searches, and unified search across externally ingested platform logs are not available yet.
Reports
On-demand CSV from live data
Reports download as CSV from the current period’s live data. Custom builders and scheduled email are not available yet. The compliance evidence pack covers up to the most recent 500 audit entries. The pilot-week value report prints counts Telara already records (tokens, tool calls, gates, HITL, estate findings raised) and lists what it will not claim: assumed hours, finding resolve, session transfer, or a “value Telara created” figure.
Your access log
Always on by design
The personal access log cannot be turned off. “Targeted” means someone viewed data scoped to you; “Directory” means a people-wide view that includes you. That self-visibility is intentional transparency.