how to migrate from a rule-based chatbot to an AI agent.
heysteff is an AI platform for customer support and sales across WhatsApp, Instagram, Messenger, Gmail and Shopify; Steff is the AI agent that runs it. This guide is for anyone who already runs a decision-tree chatbot (say, in ManyChat or another flow tool) and wants to move to a conversational agent — without losing overnight the control they already had over what the bot can and can't say.
◆ prerequisites
Access to your current chatbot's configuration (so you can review its flows), an active heysteff workspace, and a human team available for the copilot period — the migration works best with active supervision in the first weeks.
Step 1: Inventory your current flows and classify them
Before migrating anything, list the flows running in your chatbot today and sort them into three types: the informational ones (FAQs, policies, catalog — everything that answers with fixed data), the transactional ones (booking, checking stock, starting a checkout — everything that executes an action), and the marketing ones (broadcasts, campaigns, nurture sequences). Each type migrates differently, so this classification is the step that saves the most time later.
Step 2: Migrate the content
The informational flows become material for the agent's trainable knowledge base: upload the replies, policies and catalog data you already had written into your decision tree. The transactional flows, on the other hand, are reframed as tools — if your current chatbot calls an API or webhook to execute an action, that same endpoint can be reconnected as a custom tool for the AI agent.
Step 3: Run in copilot mode for 1–2 weeks
Don't switch on autopilot from the start. In copilot mode, the agent suggests the reply and a human on the team reviews it, adjusts if needed, and sends it. This is the period where you catch content and tone gaps fastest, without the end customer noticing — the one they see in the conversation is always a person.
Step 4: Turn on autopilot in stages
Once the copilot's suggestions are consistent, enable the agent's automatic replies gradually: start with the simplest, most repetitive cases (the ones you probably already covered well with fixed rules) and add coverage as you gain confidence — rather than turning off the rule-based chatbot and switching on the full agent the same day.
Step 5: Keep your escalation rules
If your rule-based chatbot already had cases that always went to a human, replicate that same logic in the AI agent with escalation rules and keywords. The migration shouldn't mean losing that control — on the contrary, with a conversational agent you can be more precise about when to escalate, because it understands context, not just exact text matches.
What NOT to migrate as-is
Be honest about the limits: mass broadcast flows without context (an identical message to your whole list, no personalization) don't benefit from becoming an AI conversation — traditional outbound campaigns are still better for that. The migration makes most sense for everything that's repetitive one-to-one conversation today: FAQs, order tracking, lead qualification and scheduling. Keep the decision-tree tool running in parallel for the flows you haven't moved yet — there's no prize for cutting over everything at once, and running both side by side for a few weeks is the safest way to confirm the agent covers what the old bot did before you retire it.
FAQ
Can I migrate only part of the chatbot and leave the rest as is?
Yes, and it's the recommended approach: migrate the highest-volume informational and transactional flows first, and leave mass marketing campaigns as they are if they already work well.
How long does the copilot period last?
It depends on your conversation volume and how complete your brand material is from day one; one to two weeks is usually enough to catch the main gaps before moving to autopilot.
Do I lose my escalation rules when I migrate?
No, they're recreated in the new agent — and you can actually make them more precise, because the agent understands context, not just exact keyword matches.
◆ next step
Migrate your rule-based chatbot to Steff with supervision, at your own pace.