Skip to content
FLOWTYPE

AI WORKFLOWS

Put AI steps to work inside workflows you control.

Classification, extraction, summarisation and decision support are steps in a workflow. You define inputs, outputs and what happens when confidence is low.

START BUILDINGEXPLORE PLATFORM
  1. Customer message
  2. AI classification
  3. Priority
  4. Routing
  5. Action

Problem

AI output is only useful when it lands in a process.

A classification or a summary does little on its own. It has to feed the next step, and people need to know when to review it.

  • Outputs arrive as free text that other systems cannot use
  • There is no clear path when the result is uncertain
  • It is hard to see what an AI step received and returned

FLOWTYPE solution

AI as a bounded, inspectable step.

In ORCHESTRATE an AI step has a defined input, a structured output and a route for low-confidence results. It is a node on the canvas like any other.

  • Structured output your next step can use
  • Confidence thresholds route to human review
  • Input and output are visible in the run log

Product visualization

Try the message routing example

This runs in your browser with simple local rules to show the workflow shape. No AI service is called.

AI step types

Assign a label and a confidence score to text, then route on the result.

Input

"My invoice shows a duplicate charge."

These are product capabilities shown as examples. Outputs are static samples, not results from an operating AI service.

Example workflow: support message routing

Simulated in your browser

Edit the text or pick a sample. Rules run locally; no AI service is called.

  1. Customer message
  2. AI classification
  3. Priority
  4. Routing
  5. Action

Select RUN CLASSIFICATION to send the message through the workflow.

Capabilities

What you get

  • Classification

    Assign labels and priorities to messages and records.

  • Extraction

    Turn documents and text into structured fields.

  • Summarisation

    Condense long threads for the next step or person.

  • Decision support

    Suggest a next step with reasons, for a person or rule to confirm.

  • Confidence routing

    Send uncertain results to an approval step.

  • Full visibility

    Inspect what each AI step received and returned.

Workflow example

Example: Support Escalation

A ticket arrives. An AI step classifies and summarises it, a rule checks priority, and urgent issues are escalated.

  1. 01

    Trigger

    A helpdesk webhook delivers the new ticket.

  2. 02

    AI

    Classification and summarisation produce structured output.

  3. 03

    Logic

    A rule compares priority to the escalation threshold.

  4. 04

    Actions

    The on-call team is paged and the support lead notified.

  5. 05

    Result

    The run records the AI output alongside every action.

  1. Ticket created
  2. Classify & summarise
  3. Urgent?
  4. Escalate to on-call
  5. Notify support lead
  6. Completed

Technical explanation

What an AI step looks like

An AI step declares its type, input mapping, output schema and review rule. This is an illustrative configuration.

  • Output is validated against a schema
  • Low confidence can route to an approval step
  • Provider selection is abstracted from the workflow
ai-step.json (illustrative)Demo example
{
  "id": "classify",
  "type": "ai.classify",
  "input": "{{ trigger.message }}",
  "labels": ["billing", "incident", "question"],
  "on_low_confidence": { "below": 0.7, "route_to": "review" }
}

Continue exploring

BUILD A WORKFLOW WITH AI STEPS.

Start with a template and adjust the AI step to your data.

START BUILDINGEXPLORE PLATFORM