# Progressive Automation
URL: /response-workflows/progressive-automation
Type: concept
Description: A structured approach to moving from AI-assisted responses to fully automated customer support.
Keywords: progressive automation, rollout, review, trust
Progressive automation is Stylo's approach to building trust in AI-generated responses. Instead of choosing between "fully manual" and "fully automated," you start with human-reviewed suggestions and gradually increase automation as you build confidence in the AI's quality.

The automation ladder [#the-automation-ladder]

| Stage                         | How it works                                                                                                                                      | Who sends the response?       |
| ----------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------- |
| **1. Assist**                 | Stylo generates a free-form suggestion based on the ticket and your knowledge base. The agent reviews, edits, and sends.                          | Agent                         |
| **2. Assist suggestions**     | Response workflows pre-generate structured responses based on your specific instructions. Agents review and send with one click.                  | Agent                         |
| **3. Conditional automation** | Workflows that consistently produce high-quality responses are switched to auto-send for straightforward cases. Complex cases still go to agents. | AI (simple) / Agent (complex) |
| **4. Full automation**        | Proven workflows handle end-to-end resolution. Agents focus on escalations and edge cases.                                                        | AI                            |

Each stage builds on the last. The data from stage 2 (acceptance rates, edit distances) is what gives you confidence to move to stage 3.

How to progress [#how-to-progress]

Stage 1 to 2: Add structure [#stage-1-to-2-add-structure]

When you notice agents repeatedly handling the same type of ticket:

1. Create a response workflow with clear strategy instructions
2. Set it to **cache** mode so it pre-generates suggestions
3. Write a specific "when to use" description

The workflow now handles the research and drafting — agents just review and send.

Stage 2 to 3: Build trust with data [#stage-2-to-3-build-trust-with-data]

Monitor your workflow's performance:

* **Acceptance rate** — how often agents use the suggestion vs. dismiss it
* **Edit distance** — how much agents modify the suggestion before sending
* **Confidence scores** — how well the workflow matches the right tickets

When you see consistently high acceptance rates (>80%) with minimal edits over a sustained period, the workflow is a candidate for automatic sending.

Stage 3 to 4: Expand coverage [#stage-3-to-4-expand-coverage]

Once a workflow is reliably auto-sending:

1. Review the escalation rules — make sure edge cases are properly routed to agents
2. Monitor the quality check rejection rate
3. Consider creating new workflows for related ticket types

What to measure [#what-to-measure]

| Metric                      | What it tells you                                  | Target for graduation          |
| --------------------------- | -------------------------------------------------- | ------------------------------ |
| **Acceptance rate**         | How often agents use the suggestion                | Above 80% over 2+ weeks        |
| **Edit distance**           | How much agents change the text                    | Under 15% average modification |
| **Confidence distribution** | Whether the workflow matches the right tickets     | Above 0.8 average confidence   |
| **Quality check pass rate** | Whether generated responses meet quality standards | Above 90%                      |
| **Escalation rate**         | Whether edge cases are properly caught             | Stable, not increasing         |

Internal notes as a stepping stone [#internal-notes-as-a-stepping-stone]

The **internal note** automation mode is a useful intermediate step. Instead of sending directly to the customer, the AI posts the response as an internal note. This lets you:

* See exactly what the AI would have sent in a real ticket context
* Monitor quality at volume without customer risk
* Build confidence before switching to public replies

This is especially useful for workflows that handle sensitive topics (refunds, cancellations, complaints) where you want extra verification before enabling customer-facing automation.

Tips [#tips]

* **Don't rush to automate.** The value of suggestions is that they make agents faster without any risk. There's no deadline to move to full automation.
* **Automate the boring stuff first.** Simple, high-volume, low-risk tickets (order status, shipping updates, thank-you responses) are the best candidates for early automation.
* **Keep escalation rules tight.** Every workflow that auto-sends should have escalation rules that catch edge cases. It's better to over-escalate than to send an inappropriate response.
* **Review regularly.** Even after automation, periodically review a sample of auto-sent responses to make sure quality hasn't drifted.

Related [#related]

* [Background Automation](/response-workflows/background-automation)
* [Escalation Rules](/response-workflows/escalation-rules)
* [Workflow Testing and Execution History](/response-workflows/testing-and-history)