Docs/Orion/Decision Intelligence

Decision Intelligence

When a metric moves or an anomaly fires, Celiq generates ranked action recommendations with expected impact and confidence scores.

TracerWeaverLuminaryUpdated Jun 2026 · 5 min read

✦

In this section

Part of Celiq's semantic layer platform. Connect your warehouse, model your data once, query it everywhere.

📝

Note

Some UI surfaces for this feature are still in development. Backend capabilities are available via the API.

💡

Tip

What it is: Ranked action recommendations generated when a metric drops or a goal falls behind — with expected impact and confidence scores.
Why use it: Don't just know that CAC spiked. Know the 3 most likely fixes, ranked by expected impact.
Where to find it: Action Center (/actions) and inline in Orion chat
Who can use it: All roles.

Decision Intelligence

When a metric moves or an anomaly fires, Celiq generates ranked action recommendations with expected impact and confidence scores.

Where recommendations appear

Action Center (/actions) — when an anomaly or goal risk is detected, the corresponding action card includes ranked recommendations. Orion chat — when you ask about a problem metric, Orion appends recommendations to its answer.

Recommendation format

Each recommendation shows:

A. Increase Meta spend
   Expected: +$40K revenue · Confidence: 74% · 1-2 weeks

B. Reduce discounting
   Expected: +$18K · Confidence: 61% · 3-4 weeks

C. Reactivate lapsed customers (email)
   Expected: +$12K · Confidence: 55% · 2-3 weeks

The recommendations are ordered by expected impact × confidence — not just raw expected impact. A high-impact, low-confidence recommendation ranks below a moderate-impact, high-confidence one.

Recording outcomes

Click Take action to log which recommendation you chose. This creates an entry in your Decision Trails log.

Come back and record the outcome after the time window elapses. Over time this improves future recommendation accuracy — Orion learns which recommendations worked for your business specifically.

Confidence scores

Confidence is based on:

Historical similar situations — how many times this pattern has appeared in your data
Causal chain strength — how strong the correlation is between the action and the outcome metric
Experiments data — if you have logged experiments with matching variables, they inform the score

Orion never claims more than 90% confidence. If a recommendation would require more certainty than the data supports, it will say so.

What if no recommendations appear?

Recommendations require:

At least 8 weeks of historical data
At least one anomaly or goal risk event as a trigger
Sufficient correlation between available actions and the affected metric

If the data is insufficient, Orion will explain what's needed to enable recommendations.