How to Build a Chatbot for Financial Services Without Tripping Over Compliance

Rule-Based Bot or AI Agent: Pick the Right Option

Before you touch a builder, make this call between a rule-based chatbot and an AI agent. If you get it wrong, your chatbot would just not solve any problems, but rather create more. 

Match the technology to the risk

Not every financial interaction carries the same level of complexity. Low-risk, predictable tasks are ideal for rule-based chatbots because every outcome is known in advance. 

AI agents cost more to build and govern, but they handle the open-ended questions customers actually ask. 

Building a Financial Services Chatbot: Step-by-Step

This is the part you came for. We'll use WotNot's no-code agent builder as the platform, but the logic applies broadly to any no-code builder with a rule-based and AI mode.

Step 1: Map the Conversation Before You Open the Builder

Before you touch a single block, sketch the flow on paper. For a financial bot, this matters more than it does for a retail site, because one wrong branch can lead a customer to the wrong product, the wrong process, or a compliance dead-end.

Decide up front: 

• What's the entry point? 

• What are the two or three things a customer will most likely want? 

• Where does the bot need to stop and hand off to a human?

For an account opening bot, that sketch looks like: 

Greet → Choose account type → Collect applicant details → Upload identity document → Confirm and close. 

That's the happy path. Now sketch the exits: 

• What happens if they choose a business account instead of personal? 

• What if document upload fails? 

Those branches are where finserv bots earn their keep or fall apart.

Step 2: Sign Up and Open the Bot Builder

If you're new to WotNot, sign up for a free 14-day trial, no credit card needed. After sign-up, you land on the welcome dashboard.

Click Go to Bot Builder, then Build a Bot, and select Build an Inbound Bot. This is the bot type that responds to users who initiate the chat. 

Next, select your channel. Web is the right starting point for most banks, insurers, and fintechs. If your customers primarily use WhatsApp (common in emerging markets and for collections), select that instead. You can always connect additional channels later.

Step 3: Choose Your Build Path

After selecting the channel, WotNot shows you two options: use a template, or build from scratch. Which one you pick determines how the next hour goes.

Path A: Using a Template (Faster, Rule-Based)

WotNot's template library includes the Account Opening Chatbot template, which is a good place to see how a finserv flow is actually structured.

The template opens with a Trigger block — the starting point of every WotNot flow. It defines when the bot activates: which page, which channel, which conditions.

From there, the account opening flow moves through these blocks:

• Greetings (Send Message block): Welcome message plus your AI disclosure if relevant: "Hi, I'm [Bank Name]'s virtual assistant. I can help you open an account, answer product questions, and connect you with a team member."

  

• Choosing Account Type (Buttons block): Savings, Money Market, or Checking. Each button routes to its own branch. The Buttons block keeps the conversation structured and stops customers from typing something the bot can't parse.

  

• Choosing Entity (Buttons block): Trust, Individual, or Company. Each entity type carries different document and compliance requirements, so this branch directly affects what you collect downstream.

• Collecting Name, Phone, and Email (Collect Input blocks): Three sequential steps. The Collect Input block stores each response as a variable you can use later for CRM pushes, notifications, or passing context to a human agent.

  

• Uploading Address Proof (Collect File block): Accepts document uploads directly in chat. For KYC flows — identity verification, address proof, income documents — this block handles it without routing customers to a separate form.

• Application Created (Send Message block): Confirmation with a reference number. In production, a Webhook or HTTP Request block connects this step to your core banking system to actually create the record.

• Conversation Ends: A clean close: "One of our executives will contact you shortly."

Every block is editable from the side panel. For most teams, this entire flow deploys in under an hour.

Path B: Building from Scratch (For an AI-Powered Financial Chatbot)

If you select Build from Scratch, you start with a blank canvas and a single Trigger block. This is the right path when you need a bot that handles open-ended customer questions, explains products, or answers queries a rule-based flow can't anticipate.

Click the + icon and open the AI tab to add an Agent block. Four things to configure:

• Prompt: Sets the persona and the guardrails. For finserv: "You are a virtual banking assistant for [Bank Name]. Answer questions only from the approved knowledge base. Do not provide personalized financial advice. If you cannot answer confidently, route to a human agent."

• Knowledge Base: Go to AI Studio, create a knowledge base, and add your sources — product documentation, rate sheets, fee schedules, loan eligibility criteria, FAQ pages. WotNot accepts URLs, PDFs, and CSVs. Set a refresh frequency so the knowledge base stays current as rates and policies change. This is the RAG layer: the bot answers only from what you've verified, not from what the underlying model thinks it knows.

• Functions: Give the bot the ability to act, not just respond. Functions can check appointment availability, look up account status, create a CRM record, or route to the right department. WotNot's functions guide walks through the setup.

• LLM Settings: Choose your model and set parameters. WotNot's AI Studio supports multiple LLMs, so you're not locked to one provider.

Once the Agent block is configured, add elements around it such as a Send Message welcome block, Buttons for quick-action shortcuts like "Check loan status" or "Speak to an advisor," and a Talk to Human block for anything the bot should step aside from — fraud, investment questions, anything flagged as uncertain.

The result is a hybrid: rule-based flows for predictable tasks, AI for open-ended ones.

Step 4: Add the Guardrails Finance Actually Needs

Before you test or deploy, configure:

• Escalation triggers: Keywords like "fraud," "dispute," "complaint," or "I want to close my account" immediately route to a human via the Talk to Human block, with full conversation history passed through. 

• PII handling: If your flow doesn't need to collect a Social Security number, don't add a field for it. If a customer volunteers sensitive data unprompted, have a policy for what the bot does with it. 

• Fallback flows: If a customer's input doesn't match any branch after two attempts, a direct Talk to Human block should catch it cleanly. 

These need to be thoroughly mentioned in the prompt to create the agent. 

Step 5: Test the Chatbot thoroughly

Click Test Bot in the top-right corner to run a live preview.

Walk the happy path once, then take up the edges. Ask the same question five different ways. Ask something the bot has no business answering and see if it holds its line. Try uploading a wrong file type. Type something urgent and see if the escalation trigger fires.

Two tests every finserv bot needs to pass: "I think there's fraud on my account" should never get a canned FAQ response. "Can I get a loan approved right now?" should never get a yes.

This is the step most teams rush, and it's not recommended at all.

Step 6: Deploy, Then Keep Watching

Click Deploy. For website bots, WotNot generates a JavaScript snippet. 

Just paste it into your site or push it through Google Tag Manager. For WhatsApp and other channels, connect your business accounts through integration settings.

Track conversation volume, resolution rates, and escalation frequency in the first few weeks. Update the knowledge base when products or rates change. A financial chatbot that was solid at launch can quietly degrade if nobody's maintaining the knowledge it answers from.