Suggestions and proactive prompts
How Orion surfaces starter questions, autocomplete, follow-ups, next-action suggestions, and Orion Pulse without being asked.
When to use it — When you want a starting point, faster question entry, or an early heads-up on what changed before you go looking.
Where to find it — The Orion question box (starter chips + autocomplete + follow-ups) and the home page (Orion Pulse cards). Critical Pulse findings also post to Slack.
Who can use it — Tracer, Weaver, and Luminary.
Most of Orion is question-and-answer: you ask, Orion answers. The features on this page are the parts that work the other way around — Orion suggests what you might want before or after you ask, and it keeps an eye on your active metrics in the background so you do not have to.
Every one of these surfaces is built to degrade quietly. If a model call times out, a table is missing, or Orion is not configured, the suggestion simply does not appear — it never blocks the question box or breaks the page.
Overview
There are five distinct suggestion surfaces, each grounded in your workspace:
| Surface | When it appears | What it gives you |
|---|---|---|
| Starter suggestions | Before you type, on the Orion box | 5–8 short starter questions about your workspace |
| Autocomplete | As you type (3+ characters) | 3 completed, answerable questions |
| Follow-ups | With every answer | Suggested next questions Orion returns alongside the response |
| Next-action suggestions | After an answer, for a metric move | Rule-based next steps and a draft-message helper |
| Orion Pulse | In the background, on the home page | The top metric moves this week, ranked; criticals posted to Slack |
The first four live around the Orion question box. Orion Pulse runs as a scheduled background scan and surfaces its results as cards on the home page.
When to use it
- You do not know what to ask — Starter suggestions give you concrete, workspace-aware questions to click instead of staring at an empty box.
- You know roughly what you want — Start typing and let autocomplete finish the question with metrics and breakdowns that actually exist in your model.
- You just got an answer and want to dig deeper — Use the follow-up questions, or the next-action suggestions for a metric that moved.
- You want to be told what changed — Let Orion Pulse surface the biggest moves of the week so you can investigate the ones that matter.
Concepts
Starter suggestions — A short list of 5–8 starter questions generated from your workspace's domains, reveals, and key metrics. They are cached per workspace for one hour, so the same chips appear for everyone in the workspace during that window. Autocomplete — As-you-type completions for the question box. Orion needs at least 3 characters before it suggests anything, and it returns 3 distinct, answerable questions that vary in time range or breakdown. Follow-ups (follow_ups) — A list of suggested next questions returned alongside an Orion answer. They are produced by the answer engine and shown under the response so you can keep the thread going with one click.
Next-action suggestions — Rule-based next steps tied to a metric and its movement (for example, a metric that dropped). A companion draft-message helper turns an insight into a short, professional message you can send to your team.
Orion Pulse — A background scan over your active reveals that finds this week's most significant metric moves, ranks them, stores the top five, and posts the critical ones to Slack. The home page shows these as Orion Pulse cards.
Significance — A 0-to-1 score Pulse assigns to each finding. It drives the ranking and the Slack threshold: findings at 0.8 or higher are treated as critical.
Getting started
You do not turn these features on individually — they are part of Orion. To see them:
- Open Orion and look at the question box for starter chips (before typing) and autocomplete (as you type).
- Ask any question to see follow-ups and, for metric-related answers, next-action suggestions under the response.
- Open the home page to see Orion Pulse cards for the current week's top moves.
Prerequisites: Orion must be configured for your workspace (a workspace Anthropic key) for AI-generated suggestions, autocomplete, and follow-ups. Orion Pulse needs at least one active reveal with selected metrics. Slack alerts for critical Pulse findings require a connected Slack workspace — see Slack.
Step-by-step
Start from a suggested question
Open Orion. Before you type, the box shows starter chips drawn from your domains, reveals, and key metrics. Click one to ask it immediately, or start typing your own.
Let autocomplete finish your question
Type at least 3 characters. Orion offers 3 completed questions that reference real metrics and dimensions and vary by time range or breakdown. Pick one or keep typing.
Use the follow-ups after an answer
When Orion answers, it returns a set of follow-up questions under the response. Click one to continue the analysis without retyping context.
Act on a metric move
For an answer about a metric that changed, Orion can suggest rule-based next actions. Use the draft-message helper to turn the insight into a short message you can share with your team.
Review Orion Pulse on the home page
Open the home page. Orion Pulse cards show the week's top moves — direction, percent change, current vs. last week, and the reveal. Click Investigate with Orion → to open a pre-filled question about that move.
Examples
Orion Pulse ranks the week's moves and keeps only the top five. A critical finding (significance 0.8 or higher) is posted to Slack in this shape:
⚠️ Orion Pulse Alert
gross margin is ↓ 22.4% this week (Weekly Orders)On the home page, the same finding renders as a Pulse card showing the change, the current-vs-prior values, and the reveal name, plus an Investigate with Orion → button. Clicking it opens Orion with a ready-made question:
Why did gross_margin change this week?Pulse chooses its detection method automatically. When the ML service is available, it uses Prophet-based anomaly detection over the last 90 days of daily values (per metric, up to 3 metrics per reveal) and only keeps anomalies from the last 7 days. When ML is unavailable or there is not enough history, it falls back to a rule-based this-week-vs-last-week comparison and fires on moves of 15% or more.
Best practices
- Click a starter chip when you are unsure where to begin — they are tailored to your workspace, not generic.
- Type a few words before expecting autocomplete — it needs at least 3 characters and works best when you name a metric or dimension.
- Treat Pulse as a triage queue — start with the highest-ranked card, since findings are ordered by significance, then by size of change.
- Use Investigate with Orion rather than retyping — the pre-filled question carries the metric and the week so Orion has the right starting context.
- Curate key metrics in workspace memory — starter suggestions and autocomplete pull from your domains, reveals, and key metrics, so a well-maintained model produces sharper prompts.
Tips
Starter suggestions are cached for one hour per workspace. If you just renamed a reveal or added a metric and want fresh chips, expect up to an hour before the starter list reflects it — autocomplete and follow-ups update immediately.
Common mistakes
Reading a mid-week drop as a collapse. The current week is in progress. Pulse deliberately suppresses an in-progress-week "collapse" that is really just incomplete data — see Troubleshooting — so a metric that looks near-zero mid-week is usually missing days, not a real crash.
Assuming no chips means something is broken. Suggestions fail silently. An empty starter list or no autocomplete usually means Orion is not configured or a model call timed out, not that the page is broken.
Troubleshooting
| Symptom | Cause | Fix |
|---|---|---|
| No starter chips appear | Orion not configured, or the model call returned nothing (empty result is still cached for 1h) | Confirm a workspace Anthropic key is set; wait out the 1h cache or revisit after configuring Orion |
| Starter chips look stale after a model change | Starter suggestions are cached per workspace for one hour | Wait for the cache to expire; the list regenerates on the next request after that |
| Autocomplete never shows | Fewer than 3 characters typed, or the suggestion call failed | Type at least 3 characters; autocomplete returns empty on any error rather than blocking the box |
| No follow-ups under an answer | The answer engine produced none for that question | Normal for some answers; use a starter chip or autocomplete to continue |
| Orion Pulse shows nothing | No active reveals with selected metrics, or no move met the threshold this week | Activate a reveal and select metrics; Pulse fires on ML anomalies in the last 7 days or rule-based moves of 15%+ |
| A real-looking crash is missing from Pulse | The in-progress week looked like incomplete data — a drop to under 10% of last week is suppressed as a likely data gap, not a critical alert | Re-check later in the week once data is complete; genuine spikes and moderate moves still surface |
| Critical Pulse finding did not reach Slack | Slack not connected, or no finding reached significance 0.8 | Connect Slack; only findings scored 0.8 or higher are posted |