Yes, for the factual questions. An AI agent can tell a customer the rate on their account, when a fixed rate ends, how interest is calculated and what the bank's published rates are, if it reads from the current approved source. Advising which rate to choose, negotiating or predicting rates stays with a person at the bank.
What makes a rate question safe to answer?
A rate question is safe when the answer is a fact the bank already holds and the customer can check. "What is the rate on my savings account?" has one correct answer in the account system. "How much interest will I earn on 10,000 over a year at the current rate?" is arithmetic on published terms.
The risk lies in the source. Rates change, and an agent reading last month's rate sheet gives a wrong answer with full confidence. A customer who moves money on the strength of that answer has acted on a misstatement, and the bank may have to honour it or compensate them. So the conditions are about where the number comes from, not how the agent phrases it.
What are the conditions?
- The source is current and approved. The agent reads the rate sheet or product terms in force on the date of the question, and the customer's own rate from the account system.
- The answer carries its basis. Each answer says which account or product it refers to and the date the rate applies from.
- The question is about facts, not choices. "What is" and "how is it calculated" are in scope. "Should I" is not.
- No forward-looking statements. The agent does not predict rate changes or say what a variable rate will be next year.
What may the agent do, and what stays with a person?
| Question type | The agent | Stays with a person |
|---|---|---|
| The rate on the customer's own account | Answers from the account system | Disputes about whether the right rate was applied |
| When a fixed or bonus rate ends, and what follows | Answers from the product terms | Nothing routine |
| How interest is calculated | Explains the method from the terms, with a worked figure | Unusual cases the terms do not cover |
| The bank's published rates | Quotes the current rate sheet with its date | Rates not on the sheet |
| Which product or rate is best for the customer | Lists the options with their published terms | The recommendation |
| Whether rates will rise or fall | Declines to predict | Nothing; the bank does not forecast for customers |
| Requests for a better rate | Passes the request on | Any decision to vary a rate |
Explaining a charge already applied is a related job, covered in whether an AI agent can answer questions about a customer's bill and in how to answer billing questions with an AI agent. Applying interest is a different matter again, covered in whether an AI agent can apply fees or interest to an account.
A worked example
Take a regional bank's mortgage service line. A customer asks: "My fixed rate ends soon. What will I pay after that, and should I fix again?"
The agent reads the mortgage record. The fixed rate ends in eight weeks, after which the loan moves to the bank's standard variable rate. The agent quotes the current standard variable rate from the rate sheet, with its effective date, and calculates the monthly payment at that rate on the current balance. It notes that the variable rate can change.
On the second question, the agent does not give a view. It lists the fixed-rate products the customer is eligible for, with their published terms, and offers an appointment with a mortgage adviser to discuss which suits them. The record shows the sources used for each figure and that the advice question was passed on.
What controls make it safe?
- Versioned rate sources. Rate sheets are loaded with an effective date. The agent reads the version in force, and an expired sheet cannot be used.
- A figure, a source, a date. Every rate quoted is filed with the document and version it came from.
- A refusal pattern for advice. Questions that ask the agent to choose are answered with options and a route to an adviser. Test this on real customer questions before go-live.
- No commitments. The agent states that published rates may change and that a quote is not an offer, unless the bank's process makes it one.
- Staff use the same sources. If colleagues ask the agent rate questions, it should draw on the same approved sheets, as in answering staff policy questions with an AI agent.
The wider limits are set out in AI agents in banking: what they may and may not do.
Questions people ask
What if the customer's rate looks wrong?
The agent can show the rate on the account, the rate in the product terms and the dates for each. If they differ, it raises a case for a person, because correcting an applied rate means adjusting interest already charged or paid.
Can the agent compare the bank's rates with other banks'?
It should not. Other banks' rates are outside its approved sources and change without notice, and a comparison drifts towards advice.
Can the agent explain an annual rate versus a monthly rate?
Yes. Explaining how rates are expressed, with a worked example on the customer's own balance, is information and helps customers understand their account.