Read and interpret organization intelligence¶

Use organization intelligence to understand service use and investigate gaps. Read the definition and availability of each measure before using it in a management or executive decision.

Before following these steps: the current frontend is work in progress and HiveAI entry is disabled in the inspected version. The steps describe the intended Final Screens conversation when available. They do not confirm deployment. The complete dashboard, period controls and recovery screens still need verification, and current metric-delivery gaps can prevent the assistant from returning numbers.

Before you begin¶

Check that you are in the intended organization. You need current HiveAI access and the permissions appropriate to organization metrics. The dashboard also requires advanced access and organization-head or governance authority. Some conversational work metrics are limited to the organization head even where delegated dashboard access is allowed.

Decide what you need to understand: use of HiveAI, quality evidence, credits, budget controls or source health. These are different questions with different time bases. A job title or an apparently relevant question does not override access controls.

If HiveAI is unavailable, use the established authorized reporting process. Ask your administrator which experience is available; do not assume that a design example is a release announcement.

Ask about organizational use¶

1. In the intended HiveAI conversation, use Your question. You can start by asking “Can you show how our team uses HiveAI?”
2. Ask a bounded question such as “How has our organization used HiveAI this month?”
3. Read HiveAI use this month and the availability message. The inspected design shows Some measures unavailable.
4. Separate the reported observations from the information that could not be checked. In the example, adoption and source-processing observations are shown while quality measures are unavailable.
5. Before sharing the answer, confirm the organization, period, definitions and freshness. If any of those are unclear, request clarification or obtain the approved underlying report.

The intended screen says HiveAI can show adoption, source health and quality measures when available. Its sample figures are not your organization's data. They also do not prove that the current assistant can compose every dashboard measure.

The inspected screens do not establish a dashboard-opening button, period picker, downloadable report or evidence drill-down. Do not depend on an unverified control. Asking a question does not change a budget, process a source or resolve a governance issue.

Choose a period carefully¶

The dashboard service accepts a positive period of up to 366 days, ending no later than the present. The exact control for selecting that period remains unverified.

Recorded adoption and base quality indicators use the selected period. Source health and some policy/control information represent the latest available state instead. Credit balances currently come from account totals rather than a reconstructed historical balance.

A heading such as “this month” does not make every section a monthly measure. When comparing reports, keep the same scope and definitions, identify which observations are current snapshots, and retain each section's freshness. Do not treat a recent-looking overall timestamp as proof that every source was recently checked.

Understand adoption¶

The current adoption calculation uses recorded HiveAI audit activity:

• The field labelled eligible users counts distinct subjects present in those records. It does not count all licensed people or everyone who could have used HiveAI.
• Active users counts distinct subjects with a successful recorded outcome.
• Executions counts the selected audit records. It is not a count of unique conversations.
• Feature use counts distinct capability codes represented in the records. It does not tell you how many features each person used.

Consequently, an active-to-eligible comparison is about the observed audit population. Do not present it as the percentage of your whole workforce adopting HiveAI.

A person missing from the observed population may not have used the service in the period. That absence does not prove that the person lacks a license or permission. A small observed population may be suppressed.

Understand quality indicators¶

Citation coverage measures how often successful audit records include at least one citation. It does not prove that a citation supports every statement or that its source was correct.

The refusal rate counts denied or refused records against all selected records. A refusal may reflect an appropriate access or policy boundary. It is not automatically a service failure.

The field labelled tool accuracy currently uses a recorded no-safety-violation outcome among successful tool records. It is not an independent evaluation of whether the tool answered correctly. Grounded-answer evaluation is currently unavailable.

Treat unavailable quality measures as unknown. Do not replace them with zero, infer that every answer is wrong, or treat the other indicators as an accuracy score.

Understand credits and budgets¶

Credit information concerns tenant microcredits, not a currency invoice. Reserved credits, settled credits and remaining credits have different purposes; do not add them into a single spending total.

Historical spending is not verified by this dashboard. It calculates a difference between balance requests for the period start and end, but the underlying service reads current account totals. A zero difference can therefore occur even where the selected period had usage. Use the approved billing or usage reconciliation process for a financial decision.

Released-credit information is unavailable. Budget rows represent configured scopes currently applicable to the reader. They are not guaranteed to cover every user, team or department budget. Overlapping scope counters must not be summed into tenant spending.

A warning means a control needs attention. Reading it does not raise a limit or authorize more use. Contact the responsible administrator through your organization's budget process.

Understand knowledge-source health¶

The latest processing snapshot groups sources by their current observed processing stage:

These are source counts, not document-page, chunk or question counts. Source-type coverage includes non-removed sources that may still be processing; it does not mean every counted source is searchable.

The snapshot can lag behind source changes. Its reported freshness is based on snapshot publication, and the dashboard does not apply an automatic age threshold that makes an old snapshot unavailable. An apparently complete section can still need a freshness check.

A backlog or retry count does not provide a completion estimate. Use your support process to investigate persistent problems. Repeating the same question does not start or accelerate source processing.

Understand governance¶

The base incident figure counts selected audit records with incident references; multiple records can refer to the same incident. It is not necessarily a count of distinct incidents.

Policy identity prefers the governance snapshot effective at the period end, while the suspended-control count uses current controls. If that snapshot is missing, policy identity falls back to current policy information. This mixed view does not establish that every control has been evaluated or that your organization has passed a compliance audit.

Where extra governed-operation measures are available, read their own definition and availability. Do not merge their counts with base dashboard totals or infer that a proposal, approval or reported attempt proves completed work.

When some measures are unavailable¶

The intended partial screen explicitly says quality measures are unavailable and that answer accuracy cannot yet be judged. Preserve that explanation when summarizing the result.

The overall response can appear available even when individual sections are partial or unavailable. Check every section. Absence of an overall warning is not evidence that all measures are complete.

Recover from access or answer failures¶

If access is denied, verify your current organization and ask an administrator to check the appropriate role, policy and commercial entitlement. Delegated dashboard authority does not automatically grant organization-head access to conversational work metrics.

If the question or answer is withheld, keep the request at an organization/process level. Do not ask for worker rankings, individual productivity or employment recommendations. Do not try to infer a person through a small team or a difference between totals.

The current wording checks can also reject an innocently phrased question. Ask support to investigate a false refusal using the safe context of the request; changing spelling to evade the check is not a recovery method.

For an unavailable source, retry later if advised and use an approved report for time-sensitive decisions. For inconsistent results, preserve the selected period, visible explanation, section freshness and safe support reference if shown. Do not send confidential document content or personal records in a public support channel.

For managers and executives¶

Use these observations to direct investigation and improve service support. They do not establish employee performance, answer accuracy, return on investment or an audited financial result.

For a management summary, include the scope, reporting period, observed population, unavailable sections and any current-state snapshots. Keep the current historical-spending limitation visible. Avoid a single green/red judgment that hides missing evidence.

Before acting on a work-process count, confirm whether it counts contribution records or current business items. Use the relevant WorkOS or organization analytics guidance to inspect the actual process.

For administrators, platform staff and support¶

Check the intended organization, current authority, enabled capability and underlying source availability. A dashboard read cannot fix a permission, budget, provider or ingestion problem.

Preserve privacy suppression and safe unavailable messages during investigation. An organization head's access does not authorize disclosure outside the organization. Use approved internal operational procedures for remediation rather than inventing controls in the user journey.

Example: review monthly use¶

A leader receives an answer with adoption figures, a knowledge-processing backlog and unavailable quality measures. The leader records the period and explains that adoption covers observed audit subjects. They request investigation of the backlog and keep answer accuracy unknown.

They do not use the dashboard's calculated credit difference as verified monthly spending, rank staff from aggregate use, or conclude that an unavailable measure is zero. The resulting summary remains useful because its limits travel with the observations.

Continue with feature scope, conversation recovery and knowledge and citations.

Use the work-facts guide when interpreting case, project, meeting or leave results. Their item lists, source dates and current balances must not be treated as complete organizational metrics.