You want to introduce AI into your payment support, but where do you actually start?
Do you let it answer customers straight away? Connect it to your transaction systems? Give it permission to take action? And what about VIP customers who still expect to speak to a real person?
The good news is that you don't have to figure all of this out on day one.
Echomind works as an additional AI layer on top of your existing support platform, such as Intercom or Zendesk. You can start with simple questions, test the results and gradually introduce more advanced workflows.
Here's how to approach the integration in four steps.
Step 1. Connect your support platform and start with simple questions
The first step is straightforward: connect Echomind to your existing CX platform, add your knowledge base and decide which questions the AI should handle first.
There's no need to connect your payment infrastructure yet.
At this stage, Echomind can answer common questions such as:
How long does a bank transfer usually take?
Which payment methods do you support?
How can I request a refund?
What documents do I need for verification?
Your team defines how the AI should respond, what information it can use and when a conversation needs to be passed to a person.
Before going live, you can test everything in Echomind's test environment. It lets you see which knowledge the AI used, how it reached an answer and where something needs to be improved.
Also configure Judge LLM to evaluate conversations against your quality criteria, including accuracy, helpfulness and adherence to support guidelines. Once the AI is live, these evaluations help your team identify answers that need improvement.
What you get: An AI agent that handles simple questions, with ongoing quality evaluation.
Before connecting real data: security and compliance
Before giving Echomind access to transaction or account data, involve your security and compliance teams.
Echomind uses self-hosted AI models, so customer data isn't sent to external cloud LLM providers. This gives you more control over data processing, but the integration still needs to be reviewed against your applicable PCI DSS, GDPR, AML/KYC and internal security requirements.
Agree on four things before moving forward:
Data access: What information does the AI actually need?
Permissions: Which functions can it call, and what actions require approval?
Data protection: How will access, processing and retention be controlled?
Audit trail: How will function calls, approvals and escalations be recorded?
For example, a transaction-status function can return the status and permitted error details without exposing full card numbers or unrelated customer information.
Once your team approves these boundaries, you can start with read-only functions and expand the integration gradually.
Step 2. Let AI check real-time payment information
A knowledge base can explain how long a transfer usually takes. But it can't tell a customer where their specific transfer is right now.
This is where Function Calling comes in.
Function Calling allows Echomind to trigger predefined functions that retrieve real-time information from your systems. Your engineers make the necessary APIs or workflows available, and you decide exactly what information the AI can access.
At this stage, we recommend starting with read-only functions. The AI can retrieve information but cannot change account details or initiate financial operations.
Imagine a customer asking:
"I sent the money yesterday. Where is it?"
Echomind triggers your existing transaction lookup function, receives the current status and explains what is happening based on the information returned.
The same approach works for several common payment requests:
Declined payment reason. A customer's card payment failed and they don't know why. Echomind checks the transaction record and returns the decline reason your processor made available: insufficient funds, expired card, bank decline — instead of a generic "please try again."
Transaction limit or hold. A customer's payment was blocked and they assume it's an error. Echomind checks the account's current limits and usage, and explains whether the block is a daily/monthly cap, a temporary hold, or something that needs manual review.
Account verification. A customer asks why their account is still under review. Echomind checks the verification status and identifies whether additional information is required, if that information is available and permitted to be shared.
Fee or FX charge breakdown. A customer disputes why a cross-border payment cost more than expected. Echomind retrieves the itemized breakdown for that specific transaction — the exchange rate applied, service fee, any intermediary charges so support can explain the number instead of guessing at it.
The AI responds using the information your systems actually return. If a decline reason isn't available, it won't invent one.
What you get: An AI agent that answers account-specific questions without requiring an agent to perform every lookup manually.
For more on the technology, read our Function Calling guide
What happens to your transaction data?
Echomind only calls the functions your engineers make available. You control which data each function returns, without giving the AI direct access to your entire payment database.
Our self-hosted models keep customer conversations and transaction data out of external cloud LLMs. We don't store your data. We don't use this data to train or fine-tune shared models.
Step 3. Let AI take action and handle escalations
Checking a transaction is useful. But sometimes the customer needs more than a status update.
Let's say a transfer has failed and requires technical investigation.
Echomind checks the transaction and the returned error shows the issue needs technical investigation. Instead of immediately handing the conversation to a human agent, it triggers your predefined escalation workflow.
That workflow can create a Jira ticket with the transaction ID, available error details and information collected during the conversation.
Echomind then tells the customer that the issue has been passed to the relevant team.
Your support agents don't need to repeat the lookup or create the ticket manually.
The same approach can support other workflows:
Refund requests. Check eligibility, collect the required information and trigger an approved refund workflow after the necessary confirmation. Requests outside your rules go to a specialist.
Account restrictions. Retrieve information that can be shared with the customer and guide them through permitted steps. Fraud, compliance and sensitive account decisions remain with the appropriate team.
Issue follow-ups. Where configured, a scheduled workflow can check ticket status and initiate an update when new information becomes available.
When does a human take over?
Echomind handles requests within the rules you've defined and escalates when:
The issue needs manual investigation or a sensitive decision.
An action needs human approval.
Identity verification cannot be completed.
The system returns an error or insufficient information.
The customer requests a human agent.
These conditions are configured during integration. Before escalating, Echomind can collect the relevant context so the next person doesn't have to start from scratch.
What you get: An AI agent that can retrieve information, trigger approved actions and escalate cases with the relevant context.
Step 4. Keep VIP support human with an AI Copilot
Not every customer conversation should be automated.
For VIP customers and high-value accounts, you may prefer to keep communication with your support agents or account managers.
Echomind's AI Copilot helps with the work behind those conversations.
Imagine a VIP customer asking about a delayed transfer. Instead of manually opening the payment dashboard and searching for the transaction, the account manager can use Copilot to retrieve the relevant information within 15 seconds and prepare a suggested response.
The account manager reviews it, adds context and decides what to send.
The customer still speaks to a person, while AI handles much of the information retrieval and response preparation.
Copilot can be introduced alongside any of the previous stages. You don't need to complete the entire automation rollout first.
What you get: Human-led VIP support with less manual work.
How much work is involved
Each stage requires a different level of involvement from your team.
Timelines depend on your knowledge base, API readiness, security approvals and available resources. We recommend starting with one use case and agreeing on a realistic rollout plan after reviewing your setup.
How do you know when to move to the next step?
Test each stage before expanding the AI's responsibilities.
After Step 1: Confirm that the AI answers your selected questions correctly and meets your quality criteria.
After Step 2: Verify that functions return accurate data and the AI handles missing information and API errors correctly.
Before enabling actions: Test permissions, confirmations, approvals, escalation rules and failure scenarios.
Your team defines what acceptable performance looks like. Echomind's test environment and Judge LLM help you evaluate the results.
VIP Copilot can be introduced independently whenever your team is ready.
Start with one request, then build from there
You don't need to automate your entire support operation to start seeing where AI can help.
Pick one repetitive request, connect the information Echomind needs and test it with your team. Once it works the way you expect, move on to the next workflow.
Some conversations can be fully automated. Others, especially VIP or sensitive payment issues, are better kept with people. Echomind supports both approaches without asking you to replace your existing CX platform.
Not sure where to start? Tell us how your payment support works today, and we'll help you identify a first use case for a pilot.
Run a pilot on your chat
We pick one conversation segment together and fix success criteria up front. No rip-and-replace.