Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.onsomble.ai/llms.txt

Use this file to discover all available pages before exploring further.

Action workflows are structured processes your business defines in Onsomble and that AI assistants walk customers through. Where knowledge content answers questions, action workflows complete tasks. If a customer asks ChatGPT “can I get a quote from [your business]?”, an action workflow is what turns that from a vague answer into an actual conversation — the assistant asks the right questions in the right order, captures the customer’s information, applies your business rules, and hands you a qualified lead or a completed booking at the end.

​
What an action workflow actually is

An action workflow is a structured series of steps. Each step typically does one of a few things:
  • Ask a question. “What kind of cover are you looking for?” “When would you like to book?”
  • Offer options. A fixed set of choices for the customer to pick from.
  • Branch based on answers. Different paths through the workflow depending on what the customer says.
  • Apply business rules. Eligibility checks, pricing logic, availability windows.
  • Produce an outcome. A quote, a booking confirmation, a qualified lead handed to your team.
The customer experiences all of this as a conversation with their AI assistant. Behind the scenes, the assistant is following the structure you’ve defined.

​
Good candidates for action workflows

Not every process is a good fit. The strongest candidates share a few qualities:
  • Well-defined inputs. The questions to ask are known and don’t depend heavily on unstructured judgement.
  • Rules-based outcomes. Eligibility, pricing, availability can be expressed as rules.
  • Real customer demand. The process is something customers actually want to do, not something you’d like them to do.
Typical examples:
  • Get a quote — a structured series of questions that produces a price or an enquiry handoff
  • Book an appointment — capture preferences, check availability, confirm
  • Check eligibility — a product, service, or scheme the customer may or may not qualify for
  • Request a callback — capture contact details and the reason with enough context for the team to respond well
  • Move through an application process — step-by-step intake for more involved flows
Processes that are highly unstructured, require empathetic judgement, or involve sensitive information you don’t want captured through a third party aren’t great candidates. Use workflows where structure helps, and keep the unstructured conversations human.

​
The visual builder

Action workflows are built through a visual builder in the Onsomble dashboard. You don’t write code. The builder lets you:
  • Add steps — questions, option lists, information screens
  • Define logic — branching based on previous answers, validation, required vs optional
  • Set outcomes — what happens at the end of the workflow (handoff, confirmation message, follow-up email)
  • Test as you go — preview how each change affects the flow without publishing
The feel is similar to building a form or a flowchart, not writing software.

​
Creating your first action workflow

1

Start with a real customer process

Pick something concrete. “Quote request for home insurance” is a good first workflow. “General enquiries” is too vague and will lead to a sprawling, unfocused flow.
2

Map the steps on paper first

Before opening the builder, sketch the flow: what questions would you ask a customer in person, in what order, what are the decisions that change the path? A five-minute whiteboarding session saves an hour of builder work.
3

Build the happy path first

In the builder, add the steps for the most common path straight through to outcome. Ignore edge cases initially. Get a working end-to-end version, then add the branches.
4

Add branching and validation

Once the happy path works, add the “if they answer this way, go there instead” logic, required fields, and input validation. Small and explicit is better than clever — customers will find edge cases you didn’t plan for.
5

Define the outcome

What happens at the last step? A lead handed to your team? A confirmation email? A booking written into your calendar system? Every workflow needs an outcome — otherwise the customer completes the flow and nothing happens.

​
Previewing and testing

Before publishing, use the preview to experience the workflow as a customer would. This is where most issues surface — confusing question wording, missing branches, outcomes that don’t trigger, flows that feel long. A few things worth testing explicitly:
  • Run the happy path end to end. Confirm the outcome actually happens.
  • Try the branches. Every branch should lead somewhere sensible.
  • Enter deliberately odd inputs. Empty responses, very long text, obviously-invalid answers. See how the workflow handles them.
  • Have someone else try it. Fresh eyes catch issues you’ve trained yourself to ignore.
If the preview feels long and tedious, the workflow is probably long and tedious for real customers too. Look for questions you could cut, combine, or defer.

​
Publishing

When the workflow is ready, publish it. Once live, it’s discoverable by AI assistants visiting the connected website — no additional per-assistant setup. A published workflow stays live until you explicitly unpublish it. Edits to a published workflow take effect on the next customer interaction.

​
Iterating on workflows

The first version of a workflow is rarely the best version. Once customers start using it, you’ll see patterns:
  • Drop-off at specific steps — a question that’s confusing, intrusive, or unnecessary
  • Unexpected answers — customers giving responses you didn’t plan for, suggesting a missing branch
  • Outcome quality — leads that aren’t qualified well, bookings that aren’t specific enough
Monitoring usage (see Managing workflows) surfaces this information. Iterate in small, frequent updates rather than large occasional rewrites.

​
What’s next

Managing workflows

Monitor usage, edit live workflows, and retire ones that are no longer needed.

Knowledge content

The other half of workflows — curated information AI assistants can query directly.