Use forecasts and executive briefings¶

Use this guide to read process forecasts, preview an existing briefing schedule and interpret a produced briefing. The conversation examples follow the inspected Final Screens. The frontend is still being built, so the steps describe the intended experience where your environment provides it.

The designed examples do not prove that a live answer can currently deliver all forecast numbers. If HiveAI is unavailable in your workspace, ask your administrator about delivery status. Do not look for an undocumented alternative entry.

Before you begin¶

You need the organization's permitted advanced HiveAI access and organization-metrics access. The direct interfaces additionally require organization-head or delegated governance authority. Policy, licensing and the relevant feature controls must allow the request. An ordinary membership, external invitation, executive job title or recipient entry is not sufficient by itself.

For a preview, an enabled briefing schedule must already exist and have a next due time. For a produced briefing, a scheduled occurrence must have been processed. A preview cannot substitute for either requirement.

Keep the business question aggregate: for example, whether recorded workflow activity is rising. Do not ask HiveAI to score individual employees, compare named people or infer a person's performance from activity counts.

Ask for a process forecast¶

The intended forecast conversation uses the following sequence:

1. In your available HiveAI conversation, ask, “Can you forecast workflow activity?”
2. Read the explanation of the required history and minimum group size. The illustrated journey explains that at least eight complete weeks are needed and that a range accompanies the estimate.
3. Ask a specific aggregate question such as, “What is the workflow activity at the 30-day point?”
4. Read the point estimate together with its interval, observation history and freshness. Confirm which work domain the answer describes.
5. Use the result to decide what to investigate. Check current work and operational context before committing resources or changing a process.

The inspected successful screen uses an example of 28 events per week with a range of 26–30. Those are design examples, not your organization's measured results. The current conversation delivery gap means a missing estimate must stay missing; the assistant should not reconstruct it from a label or invent a plausible value.

The supported source domains are workflow, cases, projects, correspondence and records. More specific department, employee or arbitrary process filters are not established by the current forecasting source. Ask for supported aggregate scope rather than repeatedly narrowing a group to reveal withheld information.

Interpret the estimate and range¶

A forecast is based on recorded contribution events, with duplicate claims for the same event counted once. A completed task, an open request and a recorded activity event are different things. The forecast does not promise a task count, processing duration, service-level result or staff requirement.

The source forms complete seven-day buckets from the start of the chosen observation period. It uses between eight and 52 complete buckets within an observation window of at most 366 days. A trailing partial week does not contribute a weekly count.

The horizon can be one to 90 days, but the calculation rounds the horizon up to whole weeks. Read “at the 30-day point” as a weekly trend estimate in the corresponding future bucket. Do not multiply it by 30 or add it to the observed count as though it were a period total.

The displayed interval reflects uncertainty under a straight-line model. A “95%” label does not establish independently measured forecasting accuracy or a guaranteed probability for your process. Changes in work, missing source evidence and non-linear patterns can make a trend misleading. A very narrow range can result from an unusually regular history; it is not proof of certainty.

Check the observation dates and source freshness separately. The time an answer was prepared does not prove that all contributing systems supplied recent, complete records.

When more history is needed¶

The inspected “More history needed” screen shows seven complete weeks where eight are required. It gives no estimate or range and suggests checking after another week.

Wait for sufficient real evidence and repeat the aggregate question. If the observation period is too short, ask an authorized administrator to verify the selected period. Do not substitute zeros or made-up activity to obtain a forecast.

Other causes of unavailability need different recovery. A small group remains protected even when the history is long enough. A stale source needs investigation of evidence delivery. An unsupported filter needs a supported aggregate question. A timeout may justify a later retry; it does not establish that no work occurred.

When HiveOrbit is not ready for a forecast¶

A forecast can be unavailable even when sufficient history already exists. HiveOrbit must currently permit operation for this workspace. Ask an authorized administrator to check its setup, operating period, pause or recovery state and any temporary access-verification problem. The forecast's general unavailable message may not identify the exact cause.

Do not treat this as a request to add invented history, change your question repeatedly or purchase more HiveAI credits without checking the cause. Waiting for another week only addresses a genuine shortage of historical evidence. Once the workspace is ready, the ordinary permission, history, cohort and freshness checks still apply.

Check an existing briefing schedule¶

The intended conversation starts with “Is my weekly workflow briefing scheduled?” The illustrated answer identifies a weekly schedule, its time and either the next occurrence or the latest produced occurrence.

Read the response as a schedule check. It is not evidence that a new schedule was created. Verify the recurrence, next due date and the meaning of the displayed time zone with your administrator. Current scheduling uses a fixed UTC hour; a bare “8 AM” needs that context.

A weekly schedule advances by seven days. A monthly schedule retains its chosen day where possible and uses the month's last day otherwise. An administrator changing settings should verify the next occurrence again, especially around month ends or local daylight-saving changes.

Preview the next briefing¶

1. Check the existing schedule in the conversation.
2. Ask, “Preview the next workflow briefing.”
3. Confirm that the response is labeled as a preview and identifies the next occurrence.
4. Read observed facts, the forecast, any recommendation and the assumptions separately.
5. Note unavailable measures or missing sections before using the preview in a discussion.

The preview uses a window ending at the next due time, which can be in the future. Information for that whole window cannot yet be complete. A preview is a view of what the available evidence supports now, not an advance copy guaranteed to match the future result.

The preview does not save a produced briefing or advance its schedule. It also does not run the scheduled paid narrative generator. Its recommendation can be a simple trend-based suggestion. The final scheduled recommendation, if generated, can differ. Your surrounding conversation can still incur its normal usage.

If the schedule is disabled or has no next due time, ask an authorized administrator to review it. Repeating a preview request will not enable it.

Read a produced briefing¶

The inspected journey asks, “Show the latest weekly workflow briefing.” The answer labels the produced briefing with its date and separates observed use, the process forecast and a recommendation.

Confirm the due date as well as the produced result. “Latest” does not necessarily mean fresh enough for today's decision, and an organization's latest occurrence may belong to a different schedule. If more than one schedule is relevant, ask for the intended one and verify the returned details.

Read each part:

• Observed facts: apply the organization-intelligence definitions. Active HiveAI users and recorded requests do not equal workforce size or business output. Unavailable answer-quality measures remain unknown.
• Forecast: check that an estimate is available, what it counts, its horizon and its range.
• Recommendation: treat it as something to evaluate. “Review backlog and handoffs” does not prove that a particular handoff caused a delay.
• Assumptions and uncertainty: check the amount and freshness of the evidence and any withholding or incompleteness reason.

A completed or “Briefing ready” result can still contain unavailable dashboard measures. A partial briefing can retain facts while omitting a generated recommendation. An unavailable section must not be interpreted as a zero value.

Reading a saved briefing can also recheck access to a Secret section. If that access no longer applies, the section can be replaced by an unavailable placeholder. Previously seeing a section is not permanent permission to retrieve it.

Create, change or disable a schedule¶

These operations exist in the backend, but a complete screen journey has not yet been verified. Work with an authorized administrator through the supported interface in your environment; this guide does not invent menu names or controls.

Prepare these details for review:

An owner does not automatically have to be included in the recipient list. The list also does not restrict all briefing readers: the current authorized read scope is organization-wide. Do not use it as a confidential distribution list.

Changing the schedule replaces its settings and re-enables it. Review that effect if the schedule was disabled. Disabling removes the next due time and preserves previously produced occurrences. The inspected interface does not offer a separate delete or regenerate operation.

If a schedule change is proposed through HiveAI, follow the approval-request guidance. Review the exact owner, recipients, recurrence, time and scope before approving. A preview or proposal is not a completed change. If the schedule changes while you review it, refresh and prepare a current proposal rather than approving stale settings.

After an uncertain response to a change, read the schedule before submitting a new change. The first request may have completed even if its response was lost. Ask support to reconcile the outcome when the visible state is unclear.

Understand credits and missing recommendations¶

Forecast calculation and scheduled narrative generation are separate activities. The scheduled narrative generator estimates usage and attempts to reserve tenant credits before calling the model. Applicable user, team and department budget scopes are resolved from the schedule owner.

A denied reservation prevents that model call. A provider failure or an empty answer can trigger release of the reservation and leave a fact-only briefing. An unsuccessful settlement can also remove the recommendation, with some cases requiring reconciliation.

A missing recommendation does not prove that no credits were reserved or consumed. A displayed zero or an incomplete result is not a billing receipt. Ask the administrator or support to check the credit outcome before requesting additional paid work.

Previewing does not invoke this scheduled generation path. Normal conversation usage still applies when you ask HiveAI to explain the preview or produced briefing.

Recover from missing or incomplete results¶

Scheduled processing is driven by an hourly job. A configured due time is not a promise of delivery at that exact minute. The current code does not establish an automatic repair path for every interrupted briefing occurrence. Support should investigate before anyone recreates schedules or repeats paid generation.

For managers and executives¶

Use a forecast as one input to a planning conversation. Ask what the events represent, which historical period was used, how recent the source evidence is, and whether the trend remains plausible. Compare with current operational knowledge.

Keep observed facts, predictions and recommended action distinct in presentations. Preserve the date, scope and uncertainty if you quote a result. Do not convert activity counts into individual performance scores, financial return or staffing requirements without separately justified evidence.

For administrators, platform staff and support¶

Confirm current tenant and feature authority, exact schedule settings, active executive membership and the relevant credit controls. An owner's workforce record and sign-in identity currently need careful reconciliation; do not ask users to guess identifiers when a valid person is rejected.

Check the produced occurrence and its sections, not only the top-level ready state. Investigate delayed processing, missing evidence and uncertain paid outcomes before promising a retry or refund. Automatic recipient notification, local-time scheduling and exact-once paid production have not been established by this documentation audit.

Handle access or deletion requests through the governed platform process. Disabling a schedule preserves produced history, and participant deletion handling can replace identifiers without erasing shared content. Avoid including confidential briefing content in a support message unless the support channel is authorized for it.

Example: prepare a weekly planning meeting¶

An authorized leader asks about workflow activity at a future weekly point. They check the eight-week history, range and source date, then compare the estimate with known upcoming work.

They preview an existing weekly briefing to see which facts are currently available. At the meeting, they open the produced occurrence for the relevant due date. If answer quality is unknown or the recommendation is missing, they say so. They use a suggested backlog review to start a discussion and separately follow the normal approval process for any change.

The result is a more informed review of the process. It is not an automated decision or a guarantee of future performance.