Most AI support agents can answer a question. But answering a question and solving a problem are two different things.
A customer asking about a refund doesn’t just want to know how refunds work. They want the refund processed. This is where Function Calling comes in. Function Calling in AI customer support allows an AI agent to trigger predefined functions during a conversation to retrieve real-time information or perform specific actions. With Function Calling, Echomind can go beyond answers and take action in your systems.
What Function Calling is
By calling functions, our AI agent can interact with your systems during a conversation and perform specific actions. Instead of explaining how to cancel a subscription, the agent cancels it on its own. Instead of describing the refund process, it initiates the refund procedure.
You decide which endpoints the agent can access: checking an order, updating an email address, or canceling a pricing plan. When a customer submits a request via chat, the agent determines what the customer needs, sends the appropriate request to your system, and provides a real, up-to-date response.
The scope of access granted is entirely up to you. Some actions may be read-only, others may require customer confirmation before changes are made, and still others may require prior human approval. Most teams start with limited access and expand it as they see how the agent performs.
For example, a customer asks to change the delivery address for an order. The agent first calls a function to check the order status. The system returns that the order hasn't shipped yet, so the address can still be changed.
The agent asks the customer to confirm the new address and then calls another function to update it. Once the system confirms the change, the agent lets the customer know it's done.
At no point does the agent need direct access to the company's database or internal systems. It can only call the specific functions the company has made available to it.
What this looks like in practice
SaaS. Сustomer asks to downgrade their subscription. The agent checks the current plan, confirms when the change will take effect, asks the customer to confirm, and updates the subscription- all within the same conversation.
Financial services. Customer spots an unfamiliar transaction on their card. The agent verifies the transaction and explains the situation in detail, then immediately blocks the card if the customer confirms they did not make the payment.
E-commerce. Customer asks where their order is. The agent checks the delivery status, can see that it’s delayed, and, if the address can still be changed, updates it right in the chat without redirecting the customer to another page.
Telecommunications. Customer says their mobile data has stopped working while abroad. The agent checks the mobile data balance, roaming status, and plan restrictions, and then, after receiving the customer’s consent, activates the necessary services.
All four examples show the same trend: the agent doesn’t just tell the customer what to do next. It can do it for them.
Where your data goes
Simply put: nothing is stored permanently anywhere. The agent sends a request directly to your system via the API you’ve connected, receives a real-time response, and uses it just that one time- only within the context of that specific conversation. Absolutely nothing is recorded in a separate database on our end.
It differs from a typical integration, which involves a separate tool keeping a current copy of your data. Here, each request is a snapshot, not a persistent copy.
A single read-only access is a good place to start: order status, account balances, and subscription details. As soon as this is running smoothly, it will be much easier to decide which features to expand that will truly save your team time.
Start with one action
You don’t need to give an AI agent access to dozens of functions from day one. Pick one repetitive request your support team handles manually: checking an order, changing account details, canceling a subscription, or something else that requires an action in your system.
Connect that function, test it, and see how it works in real conversations. Then add the next one.
Want to see how Function Calling could work for your specific use case? Leave us a request, and we’ll show you in a demo.
Run a pilot on your chat
We pick one conversation segment together and fix success criteria up front. No rip-and-replace.