Skip to main content
AI workflows let you package a set of instructions into a button that anyone on your team can click. When triggered, the AI follows those instructions — querying data, summarizing results, interacting with the user, or walking through a multi-step process — all without the user needing to write a prompt. Workflows are defined in the semantic model by modelers and appear as buttons in the AI chat interface.
This feature is currently in beta. To have it enabled for your organization, contact Omni support.

​
Use cases

Workflows are flexible. A workflow can be as simple as a one-line prompt or as detailed as a multi-step agent with specific output formats and decision logic. Some examples:
  • Standardized reports — Generate a weekly sales summary with specific metrics, groupings, and comparisons, consistently formatted every time.
  • Guided lookup — Ask the user for an account name, then pull relevant records and summarize findings in a structured format.
  • Data quality checks — Run a series of queries to identify anomalies, missing values, or threshold breaches, and summarize the results.
  • Onboarding helpers — Walk new users through a dataset by querying key metrics and explaining what each one means.
The key value is consistency: instead of relying on each user to write the right prompt, you define the instructions once and everyone gets the same experience.

​
How workflows work

When a user clicks a workflow button:
  1. The workflow’s input (if defined) and description are sent to the AI as a prompt
  2. The AI follows the instructions, which may involve generating queries, asking the user for input, summarizing results, or performing multi-step analysis — all routed through the relevant topic
  3. Results are displayed in the chat interface, and the user can continue the conversation to refine or follow up

​
Where workflows appear

​
Defining workflows

Workflows are defined in your model’s YAML files and can be set at two levels: Define workflows at the root of your model file to make them available across all topics: 
workflows:
  weekly_sales_summary:
    label: Weekly Sales Summary
    description: >
      Generate a summary of sales for the current week compared to
      the previous week. Include total revenue, order count, and
      average order value. Break down by product category.

​
Workflow properties

​
label
string
required
The display name shown on the workflow button. Keep this short — aim for 2-4 words that clearly describe the task.
​
description
string
required
The instructions sent to the AI when the workflow is triggered. This can range from a single sentence to a detailed, multi-step prompt with specific output formats, decision logic, and references to topics.Write this as if you’re instructing an analyst: be specific about which data to query, how to structure the output, and what steps to follow.
​
input
string
An opening message that sets the initial interaction. This is sent to the AI before the description and is useful for:
  • Greeting the user and explaining what the workflow does
  • Asking the user for input before proceeding (e.g., an account name or date range)
When input is set, the AI will typically respond to the input first, then use the description as its instructions for the rest of the conversation.

​
Examples

​
Simple: one-step query

A workflow that generates a single, well-defined query: 
workflows:
  top_customers:
    label: Top Customers
    description: >
      Find the top 20 customers by total lifetime revenue.
      Include customer name, total revenue, order count,
      and date of most recent order. Sort by revenue descending.

​
Interactive: ask the user for input first

Use input to collect information from the user before running the analysis: 
workflows:
  account_lookup:
    label: Account Lookup
    input: >
      Ask the user for the name of the account they want to
      look up before proceeding.
    description: >
      Look up information for the account the user specified.

      1. Find all records matching the account name.
         Pull back key fields: account value, status, owner,
         and created date.
      2. Find their recent product usage trends
      3. Find their recent support tickets
      4. Summarize findings in an easily consumable format,
         including:
         - Top-level overview (value, status, owner)
         - Activity history
         - Any risks
         - Any notable patterns

​
Advanced: multi-step process with structured output

For more complex workflows, the description can include detailed steps, output formatting, and decision logic: 
workflows:
  data_quality_check:
    label: Data Quality Check
    description: >
      Run a data quality audit on the orders dataset.

      Step 1: Query for orders with missing customer IDs
      in the last 30 days. Output a count and percentage
      of total orders.

      Step 2: Query for duplicate order IDs. Output any
      duplicates found with their timestamps.

      Step 3: Check for orders where the total is negative
      or zero.

      Summarize findings in this format:
      - Overall health score (Good / Needs Attention / Critical)
      - Table of issues found with counts
      - Recommended next steps for each issue

​
Using file uploads with workflows

Users can attach files during a workflow conversation to provide additional context for the AI to work with. For example, a workflow might ask the user to upload a CSV of target accounts, then cross-reference that list against data in Omni. Supported file types include CSV, PDF, JSON, YAML, TXT, and images (PNG, JPEG). Files are processed client-side and are not stored on Omni’s servers. 
workflows:
  compare_targets:
    label: Compare Target List
    input: >
      Upload a CSV of target accounts and I'll compare them
      against our existing customer data.
    description: >
      The user will upload a CSV file containing a list of
      target accounts. For each account in the file:

      1. Check if the account exists in our customer database
      2. If it exists, pull the current status, lifetime revenue,
         and most recent activity date
      3. If it doesn't exist, flag it as a new prospect

      Summarize findings as:
      - Count of existing customers vs new prospects
      - Table of all accounts with match status and key metrics
      - Any accounts with high revenue potential to prioritize
The Query Assistant > File uploads AI setting must be enabled for users to attach files in the chat.

​
Workflows vs sample queries

Both workflows and sample queries appear as buttons in the AI interface, but they serve different purposes:
WorkflowsSample queries
PurposeProvide the AI with instructions to followDemonstrate example questions users can ask
How they workSend a prompt (simple or multi-step) to the AIRun a predefined query with specific field selections
Best forRepeatable processes, guided analysis, structured outputsShowcasing available data and common queries
You can use both together. Sample queries help users discover what data is available, while workflows package complex or repetitive tasks into a single click.

​
Tips for writing effective workflows

  • Be specific in the description. The more detail you provide, the more consistent the AI’s output will be. Include the fields, time ranges, groupings, metrics, and output format you want.
  • Use input to collect context. If your workflow needs user-specific input (like an account name or date range), use the input field to ask for it before the AI begins its analysis.
  • Structure complex workflows with numbered steps. For multi-step processes, numbered steps help the AI follow your intended sequence and produce organized results.
  • Specify output format. Tell the AI how to present results — tables, summaries, sections with headers, health scores — to get consistent, readable output.
  • Reference topics when helpful. If your workflow should use a specific topic, mention it in the description (e.g., “use the Orders topic”).
  • Test and iterate. Try your workflow in the AI Assistant to see how the AI interprets the instructions, then refine the description as needed.
  • Keep labels short. Labels appear as buttons, so aim for 2-4 words that clearly describe the task.