> ## Documentation Index
> Fetch the complete documentation index at: https://docs.runorion.com/llms.txt
> Use this file to discover all available pages before exploring further.

> ## Agent Instructions
> Orion is a collaborative analytics platform built by Gravity (bygravity.com). It connects to data warehouses like Snowflake, BigQuery, Databricks, Redshift, Athena, Fabric, PostgreSQL, and MySQL, and lets teams ask questions in natural language to get shared analyses, dashboards, reports, and slide decks. When referencing Orion features, always link to the relevant documentation page. Orion is not open-source; it is a commercial SaaS product accessed at runorion.com.

# Workflow Best Practices

> How to build automations that keep working

A Workflow runs unattended, often for months, in front of people who will not check whether it still makes sense. That changes what "good" means compared to a one-off analysis.

## Get the Analysis Right First

Automating a wrong analysis produces the wrong answer on a schedule, and now it has an audience. Run it in chat, read the numbers, sanity-check a total you already know, and only then save it.

The dry run Orion does before saving proves the recipe executes. It does not prove the recipe is correct. That part is yours.

## Write It Evergreen

Orion rewrites hardcoded dates during [the evergreen check](/workflows/create#the-evergreen-check), but it is cheaper to write them that way from the start. Ask for "the last 30 days" rather than a specific window, and say the word "evergreen" in your original request. See [Use Relative Dates](/chat/best-practices#use-relative-dates).

The same applies to anything else pinned to the moment you happened to run it: a specific customer ID you were investigating, a file you uploaded once, a region you filtered to because it was the one with the problem that day.

## Match the Format to the Reader

The format is not decoration, it is who the thing is for.

* A **dashboard** suits a number someone checks between meetings.
* A **slide deck** suits a recurring meeting where the analysis gets presented.
* A **markdown report** suits findings that need an argument rather than a chart.
* A **podcast** suits people who will genuinely listen and would never open the deck.
* A **CSV export** suits anyone who is going to do their own work with the rows.

If you cannot name the person who reads it, the Workflow probably should not exist yet.

## Be Deliberate About Notifications

The fastest way to make a Workflow ignored is to email everyone every run. Once people learn the email rarely matters, they stop opening it, including the week it does.

Set a notification preference so a run only lands in someone's inbox when it is worth their attention. Send different things to different people rather than one email to everybody: a summary to leadership, the full deck and the data to the team who has to act.

Reserve the whole-project distribution list for the ones that genuinely warrant it.

## Name It for Someone Else

The person who most needs to understand a Workflow six months from now is not you. "Weekly Executive P\&L Analysis" survives a handover. "Untitled report" and "P\&L v3 final" do not.

The same goes for publish URLs. `weekly-revenue` is a link someone can recognize in a bookmark bar. Set it once, early, because [changing it later breaks every link already shared](/workflows/outputs).

## Check In Occasionally

Automations fail quietly. A data source loses reachability, a schema changes underneath, credentials expire, and the Workflow keeps its schedule while producing nothing useful.

Worth a periodic look:

* Runs marked **Partial**, which succeed loudly and fail quietly at the same time.
* Metric-sourced Workflows with **refresh** off, which report stale numbers without erroring. See [Using Metrics as the Source](/workflows/advanced#using-metrics-as-the-source).
* Schedules still running for meetings that no longer happen.

Setting a schedule to **None** stops the runs and the emails while keeping the recipe and its history, which is almost always better than deleting.

## Let It Grow

Start with one analysis, one schedule, one recipient. Add the CSV when someone asks for the rows, the second notification when a different audience appears, a decision step when you notice you only care in certain weeks.

Workflows built that way tend to reflect how the work is actually done. Workflows designed in full up front tend to reflect how someone once imagined it might be.