Transcript: How We Built an AI-Powered Order Support System with Voiceflow, Make & OpenAI
We have a user messaging the Blue Coffee website.
We added a chat widget to the site using JavaScript. When a visitor opens the website, the chat widget loads automatically. They can click it and start a conversation.
This chatbot is powered by Voiceflow — a platform that lets us design conversation flows. We can add buttons, capture inputs, and build logic for what happens in the chat.
Once a message is sent, it triggers a webhook. A webhook is like a smart link — it receives data instantly. The message goes to Make, where the automation starts.
Inside Make, the AI agent powered by OpenAI decides what to do next. For example, if the customer asks about an order, the bot first verifies their identity. It asks for an email, sends a verification code, and then checks the code.
We built separate functions in Make for:
- Generating and emailing a code
- Verifying that code
- Checking order status
- Checking shipping status
- Creating an order
These functions ensure the AI has just enough context to respond — and only to the relevant case. It doesn’t have full access to the CRM. It only receives what’s needed for that one conversation.
Also important: OpenAI doesn’t train on this data. The assistant uses the API, and privacy is maintained — no data is shared for training purposes.
Now, back in Voiceflow, we built a flow that listens for requests like “Help me with my order.” It captures the user’s message and sends it to the webhook. The AI then processes the request and replies instantly. This creates a conversation loop between the AI and the customer — all happening on the website.
The same logic could be reused in WhatsApp or Telegram. If you connected those channels, the same AI agent would work there too.
From there, the scenario continues. The message goes to OpenAI, a response is generated, structured, and sent back. If the AI needs to trigger a function (like checking shipping), it does that too.
Here’s an example:
- A user wants to track shipping.
- The AI sends the request to our shipping provider via API (in our case, SameDay).
- We fetch the latest status and return that info to the user.
Or, for example, we can:
- Generate a code
- Store it
- Email it
- Create Shopify orders via API
All of this happens in Make.
To build something like this, start with OpenAI’s API, not ChatGPT. Go to the Assistants tab and create a new assistant.
Define clear instructions — what the assistant should do, what it can access, what rules it should follow.
Then, create functions like:
- Get Order Info
- Create Auth Code
- Validate Code
- Get Shipping History
Each function has a name, description, and parameters. You can use OpenAI's examples or write your own. If you want to build a function to send an email, for example, just specify the parameters like email address, subject, and message.
You can even use a custom GPT to help you generate these functions faster.
Back in Voiceflow, the assistant setup includes:
- Trigger conditions
- Options the user can select
- The webhook to Make
- Capturing the thread ID (so the conversation remains continuous)
In Make, once you create the webhook, every time a message is sent, the data goes to that endpoint. From there:
- You send the message to OpenAI
- It runs through the assistant and functions
- The response is sent back to Voiceflow or another interface
Let’s look at a quick demo:
When I type “Hi, can you give me the status of my order?” — the system triggers immediately. The bot replies: “To help with your order, I need your reference number or an email to authenticate you.”
Once the user gives their email, it triggers the email verification function. The AI sends the code and waits for validation.
To wrap up:
- Voiceflow handles the chat interface
- Make handles the automation and AI connections
- OpenAI powers the assistant’s decisions
Everything is modular, flexible, and secure — and can be used across different channels and tools.
