This guide helps you migrate a legacy AI chatbot setup to Hugo while keeping the useful parts of your existing automations.

Hugo is the recommended way to automate support with AI in Crisp. Instead of maintaining large workflow trees for every possible question, you can let Hugo answer from your knowledge, follow clear instructions, start workflows when a structured process is needed, and escalate safely when a human should take over.

Why migrate to Hugo

Legacy AI chatbot setups were powerful, but they often required a lot of branching, fallback handling, manual source selection, and maintenance. Hugo simplifies this by acting as an AI Agent connected to your Crisp workspace.

Migrating to Hugo helps you:

• Reduce workflow complexity → fewer branches are needed for regular support questions
• Improve answer quality → Hugo can answer from your enabled training resources and current conversation context
• Use safer escalation → Hugo can escalate when it lacks enough information, detects frustration, or reaches a rule you configured
• Keep procedural automation → workflows remain available when a fixed flow is the right tool
• Connect tools and integrations → Hugo can use approved integrations and MCP tools for real actions
• Measure impact → Hugo analytics and topics help you understand what is being handled automatically

A good migration does not mean deleting every workflow. It means deciding which parts should become Hugo knowledge, which parts should become Hugo instructions, which parts should become Hugo routing rules, and which parts should stay as workflows.

Migration video guide

If you prefer to see the migration flow visually, this video walks through the main ideas behind moving older AI chatbot setups to Hugo.

Start the migration

Open Crisp, then click AI Agent from the left menu.

If you have not activated Hugo yet, Crisp may ask a workspace Owner to confirm the migration. This does not mean Hugo instantly takes over all live conversations. Your existing workflows remain available in AI Agent → Automate → Workflow Builder, and you can configure Hugo progressively before enabling it for customers.

After the migration, review two areas first:

• AI Agent → Train → where your knowledge sources are managed
• AI Agent → Automate → Workflow Builder → where your existing procedural flows are stored

From there, choose one of the two approaches below.

Choose your migration approach

AI-first approach

This is the recommended path for most teams. Hugo handles the conversation first, answers from your resources, and triggers workflows only when a procedural flow is needed.

Use this approach when you want to:

• Answer common support questions automatically
• Reduce the number of workflow branches you maintain
• Let Hugo decide when to escalate or route
• Keep workflows for structured tasks such as demo booking, lead qualification, or data collection

To configure it, go to AI Agent → Agent → Activation, enable Hugo for new conversations, and set your default target to Answer with AI Agent. Then use AI Agent → Guidance → Routing to describe situations where Hugo should start a workflow, move the conversation, or escalate.

Workflow-first approach

This approach keeps a workflow at the beginning of the conversation and hands off to Hugo later. It is useful when you must collect mandatory information, enforce a strict qualification path, or run channel-specific logic before Hugo answers.

Use this approach when you need to:

• Collect required information before any answer is sent
• Separate users by channel, country, plan, or segment
• Show buttons or inputs before the AI takes over
• Keep an existing routing structure while introducing Hugo gradually

In the Workflow Builder, use the Answer with Hugo block when the flow should hand the conversation to Hugo. You can also pass context from the workflow so Hugo understands what already happened.

What to migrate where

Use this mapping when reviewing older chatbot workflows:

• Factual answers → move them to AI Agent → Train, such as Q&A snippets, knowledge base articles, website content, or files
• Tone and behavior rules → move them to AI Agent → Guidance → Instructions
• Topic-based decisions → move them to AI Agent → Guidance → Routing when Hugo should act on the situation
• Button and input flows → keep them as Workflows when the user must choose or submit information
• Fallback handling → usually let Hugo escalation settings handle it instead of building manual fallback branches
• Dynamic account or order data → use an integration or MCP tool instead of static training content

This keeps each type of information in the place where Hugo can use it best.

Migration steps

1. Audit your existing workflows

List the workflows that currently answer questions, collect information, route users, or trigger internal actions. Mark which ones are still useful and which ones only exist to compensate for older AI limitations.

Pay special attention to branches that contain long answer text. Those answers are usually better stored as training resources so Hugo can reuse them naturally across conversations.

2. Move knowledge into training resources

If a workflow contains factual content, turn it into a knowledge base article or Q&A snippet. Hugo should learn facts from training resources, not from hidden workflow branches.

Good training resources are specific, well-scoped, and easy to update. They should answer the question directly, include the relevant conditions or exceptions, and avoid behavioral instructions.

3. Recreate behavior rules as instructions

If a workflow contains tone, formatting, or answer-style rules, move them to Guidance → Instructions. Instructions should tell Hugo how to behave, not what product facts to memorize.

For example, Always ask for the order number before discussing delivery issues belongs in instructions. A list of delivery zones and prices belongs in training resources.

4. Recreate decisions as routing rules

If a workflow detects a situation and then routes the conversation, consider using Guidance → Routing. Routing rules are easier to maintain when the decision depends on intent, topic, urgency, or emotion.

For example, you can ask Hugo to start a refund workflow when the user clearly wants a refund, or move the conversation to a sales inbox when the user asks about a partnership.

5. Keep structured flows as workflows

Do not remove workflows that still serve a clear purpose. Demo booking, support qualification, user-detail collection, department routing, out-of-office flows, and product-specific forms can remain valuable.

The difference is that Hugo can now start these workflows only when they are useful, instead of forcing every user through the same path.

6. Test before activating live handling

Use AI Agent → Evaluate → Playground to test realistic customer questions. Review the guardrails and sources Hugo used, then adjust training, instructions, or routing as needed.

When the Playground results are consistent, enable Hugo gradually from AI Agent → Agent → Activation. You can limit channels, adjust escalation settings, or keep a workflow-first setup while the migration is in progress.

Frequently Asked Questions

Still have questions which were not covered in this article? Here is a collection of the most frequently asked questions on this topic.

Will the migration create downtime?

No, as long as you test before enabling Hugo for live conversations. You can configure Hugo, train it, test it in the Playground, and only then activate it from AI Agent → Agent → Activation.

Should I delete all my old workflows?

No. Delete or simplify workflows that only exist to generate AI answers, but keep workflows that perform structured tasks. Hugo and workflows are complementary: Hugo handles open-ended conversation, while workflows handle predictable processes.

Can I test Hugo before enabling it for customers?

Yes. Use AI Agent → Evaluate → Playground to test Hugo safely. This is the best place to validate training resources, instructions, escalation behavior, and routing before Hugo handles real users.

What if I am not ready to use Hugo everywhere?

Start with a limited scope. Enable Hugo on selected channels, use routing rules for specific topics, or keep a workflow-first approach while you migrate your knowledge and instructions progressively.

What should stay in workflows after the migration?

Keep workflows for fixed processes: collecting required fields, showing buttons, asking for consent, routing departments, triggering webhooks, booking meetings, or running a step-by-step process. Move broad support answers and AI-style fallback logic to Hugo.

Updated on: 04/05/2026