Conversational AI for banking only works when every reply is one you would be happy to show a regulator. This article is for lenders, credit brokers, wealth firms and other regulated financial businesses of 10 to 250 people. It explains how to set up an AI agent that answers enquiries, collects facts and books appointments, while every piece of advice stays with a qualified person. You get faster responses and a full written record, without handing judgement to software.
What the AI should do and what it must not do
Conversational AI in banking is often sold as a replacement for advisers. That is the wrong framing for a firm of your size. The useful job is narrower: answer factual questions about your process, gather the details an adviser needs, and put a meeting in the diary.
Draw the line in writing before you build anything. The agent may explain documents required, opening hours, how an application progresses and how to book. It must never comment on suitability, rates for a specific person, eligibility or affordability. Those questions go to a person, every time.
Your compliance officer should approve the agent's instructions and knowledge base before it goes live. Treat them like any other customer-facing document you file.
Before you start
- Your account with contacts, calendars and the channels you use: web chat, SMS, WhatsApp or email.
- A list of approved answers, signed off by whoever owns your financial promotions.
- A written list of topics the agent must refuse, with the exact refusal wording.
- An adviser calendar with real availability and a round-robin rule if several advisers share enquiries.
- Custom fields for the facts you need before a first meeting, such as purpose of the enquiry and preferred contact time.
How to set it up
- Go to AI Agents > Conversation AI and click Agents List.
- Click the agent, then Bot Goals.
- In Information Collection, select fields such as Name, Email and Phone.
- Add custom questions for the details a banking enquiry needs.
- Use Update Contact Field to choose where each answer is saved.
- Turn on Skip if Already Filled for details already on record.
- Turn on Appointment Booking once the adviser calendar is configured.
- Select Trigger Workflow if the agent should start an automation.
- Set up Email Notification Settings so the right team is alerted.
- Turn on Enable Conversation Summary for the adviser who follows up.
- Click Save and send sample messages in the test panel.
Why this matters for a firm with flat revenue
Most regulated firms lose business in the gap between enquiry and first conversation. An enquiry arrives on Friday evening, nobody sees it until Monday, and by then the prospect has spoken to a competitor. The agent closes that gap. It replies in seconds, collects the facts and books the meeting.
It also replaces a lot of manual typing. Advisers stop re-asking basic questions because the answers are already on the record. Your administrators stop chasing missing details by phone.
Suppose the firm receives 220 enquiries a month, a third of them outside office hours, and those out-of-hours enquiries wait an average of 15 hours for a reply. With an agent answering and booking, each would get a reply in under a minute, and the adviser would start the first call with the facts already captured. That is roughly 70 enquiries a month no longer sitting overnight.
The written record is the real benefit
Every conversation sits on the contact record in Conversations, with timestamps. If a customer later says they were told something, you can read exactly what was said. That is better evidence than a phone note typed from memory.
Keep it that way. Do not let staff move conversations to personal mobiles or private messaging accounts. One inbox means one audit trail.
Running time each week
Expect about an hour a week. Read every conversation the agent handed over, check that refusals were triggered correctly and add any new factual question to the approved list. Once a quarter, have compliance re-approve the instructions.
Firms often load their whole website into the knowledge base. Marketing pages contain phrases like "best rates" that the agent may repeat out of context. Load only material written for factual answers.
When not to use it
Do not use an AI agent for complaints, vulnerable customer conversations or anything involving a decision on an application. Route those straight to a person with a workflow that tags the contact and notifies a named manager. If most of your enquiries are complex from the first sentence, a fast human callback may serve you better than an agent. For phone-heavy firms, read about AI phone answering before deciding on chat.
Troubleshooting
The agent answers an eligibility question. Your refusal list is too vague. Name the topics in plain words and give the agent the exact sentence to use instead.
Customers drop out before booking. The agent is asking too many questions first. Cut the intake to the three facts the adviser truly needs before a call.
Two advisers get the same lead. Check that only one workflow assigns the contact and that round robin is set on the calendar, not in two places.
Replies quote an outdated fee. An old document is still in the knowledge base. Remove it and keep one current fee page.
Frequently asked questions
Is conversational AI in banking allowed by regulators?
Regulators care about what is said to customers, not which tool says it. If the agent gives only approved factual information, records every conversation and hands advice to qualified staff, it fits the same controls as your website and emails. Check with your own compliance adviser for your permissions.
Can the agent take documents from customers?
It can send a secure form or upload link, and the files attach to the contact record. Do not ask customers to paste sensitive details into a chat message.
Should we start with chat or voice?
Start with chat. Written conversations are easier to review and approve. Once your answers are stable, the same rules can carry over to an AI receptionist on the phone line.
How is this different from a general chatbot?
A general chatbot answers questions. This set-up answers, collects facts, books and records, inside the CRM. The conversational AI overview covers the general case, and AI customer support covers service conversations with existing customers.