Using Generative AI to enhance business banking signups and credential checks

Highlights

The Challenge

“In a week, create a proof of concept that demonstrates how AI can be used to improve the Business Bank signup and KYC checking process.”

Understanding the problem

Getting insights from experts

• Having had previously worked on Dun & Bradstreet’s Risk Analytics KYS platform, I had some contextual understanding, but felt that it was better to have some more in-depth insights
• As we had no users to work with and limited time, my company provided time with two subject matter experts, who helped explain the KYC process in detail, which was invaluable, as we had little insight into the employee processes involved, as well as provide some suggestions as to where we could make improvements.

Auditing existing systems

• As a starting point, we conducted a UX Audit of the existing business banking signup process (which doesn’t involve an AI assistant at the moment)
  • Some key points we discovered:
    • The process was kept direct, and free from distractions, encouraging the user to focus on the task, and see it through to completion
    • Exploring the different paths in the product, we found that the visual identity switched when following some options, which gave a disjointed feel to the experience.
    • Of course, we could only go so far in the journey without actually starting the application process
    • But also, we noticed the reminders from the bank to try and encourage us to come back and finish our application were missing any ability to decline or unsubscribe.
• We also evaluated the information provided by Companies House and the APIs of other data brokers, so that we could understand how providing a company number could provide important data for the KYC process.

Exploring the problem

• Using everything we had learned above, I was able to map out user journeys for both the Customers and Employees, demonstrating just how the two overlapped, and how information and messaging could be transmitted from one side to the other.
• With this process, I could then interrogate and refine my ideas, making notes, identifying potential problems, and highlighting opportunities for improvement
• By then presenting my ideas to my colleagues I could get their perspectives and feedback for further refinement.
• Some key points emerged from this work:
  • We wanted to make the signup process as simple and easy for the Customer, where each step is clear and easy to follow
  • We should also understand that it may not be entirely linear – Customers might have questions they want to ask, or may need to leave the process and come back again, so we needed to build in affordances for these, as well as reminders to encourage them to return (with the option to abandon the process, if they wished)
  • Customers expect a quick response, and therefore it was important to ensure that the system would remind employees and help them to keep to these expectations.

Leveraging AI to improve the process

• Researching AI capabilities, we identified several ways in which AI could be used to enhance the process we were building:
• Primarily, Generative AI provides a “concierge” experience, being able to not just guide the Customer through the process, but to also pivot and answer questions during that process, and adapt the process to suit the needs of the user
• This experience could be extended to all communications, and with an Agent-style design, be made to feel like a dedicated representative, responding to the needs of the Customer individually (as opposed to the common customer experience of being passed between representatives, which can be a point of frustration for many Customers).
• This could also be applied to the Employee experience, making them aware of service-level deadlines, helping them to prioritise work, as well as being able to answer questions around the process and findings.
• The key value of the AI on the Employee side would be to scan documents and data to flag keywords and issues, presenting them to the Employee for human review. Used well, this could save considerable time.

The Customer Journey

Starting the journey

Creating our solution, we began with the Customer journey:

• We redesigned the Business banking homepage to provide an introduction to the AI Agent, as well as provide a quick summary of the signup process and what would be required for it. This ensures that the Customer is aware of the agent, knows what to expect, and everything that they will need before they begin.
• Once they start the process, the Agent introduces themselves, and are presented as an anthropomorphic character, to further encourage feelings of human relatability and equate the experience with that of a human counterpart. A progress bar is also provided, so that the Customer can see all the parts of the process, and understand at any given time where they are on it.
• The AI Agent starts by asking for simple details, including the Customer’s name, email address and telephone number. This not only provides a way of starting the conversation, but also allows the system to start the checking process by verifying the contact details, and ensures that they have a method of contacting the customer, should they abandon the process at any point.
• When the user receives communications from the system, they are made to resemble the AI Agent, reinforcing the idea of a dedicated representative. The messages not only contain verification codes, but also provide a link back to the conversation, providing an extra route for the Customer to return to the process at any point, and are encouraged to keep the message for that very purpose.
• The system retains the place in the process, so that the Customer can rejoin from where they left off. The chat history is also kept, so that the Customer can always scroll up and remind themselves of what was previously said.
• During the process, the Agent can check for errors, ensuring that details such as telephone numbers and email addresses are correct, providing real-time feedback and encouraging the user to check and make corrections.

Company details

• After the initial contact details, the Customer is then asked for details about their company. They only need to confirm three details; type of company, size of turnover, and company number. With these details, the system can get everything else it needs from the Companies House API.
• As the process takes the form of a chat conversation, the Agent can not only guide the Customer through the application process, but can also answer questions at any stage along the way, including clarifications on terminology. The Agent should be trained to answer questions around a specific list of subjects, but should always be provided with limitations, to prevent it from saying anything that is not authorised. If a Customer asks a question that is not on the specific list, the Agent is trained to respond that it can’t answer the question, but can put them in touch with a human colleague to answer it for them, which would happen inside the same chat window, for a seamless experience.
• Once the Customer has provided the company number, the Agent provides them with a synopsis of the basic information from the Companies House API, so that the Customer can verify their company. This also includes details of the company officers, confirmation of which will lead them on to the Documents stage of the process.

Providing documents

• After providing typed information, the Documents stage is a little more complex, as it involves not just the Customer to provide information, but also the company officers who will also have access to the account. The process starts by asking if the Customer can provide everything themselves, or if they need the Agent to contact the officers on their behalf. If so, the Customer can then provide contact details for the officers, again for communication and verification purposes.
• As before, the communications from the system maintain the feel that everything is coming from the Agent, reinforcing the impression of a dedicated representative. The messaging provides an introduction from the Agent, that mentions the Customer to provide reassurance of validity. The user can then follow the link on the message to provide their document details.
• The process can work on both mobile and desktop devices, making use of the camera or webcam to capture the document for verification. The process starts with a page the introduces the verification process, ensuring that they know what’s required before they begin.
• The screen then switches to a camera view for the user to show the document. The screen checks the positioning of the document, indicating the correct positioning in each corner and the legibility of the text through OCR. When the Officer pushes the button, both the image and the text is captured, so that both can be verified by a human. 
• The final stage of the document process involves the Officer verifying the captured information.

The Employee journey

KYC checks overview

• The KYC dashboard gives the Employee an overview of all of the checks that they need to perform, as well as a reference of previous checks
• Each KYC check has an overview of issues that the AI has located, as well as the number of issues that have been reviewed, and any flags that have been raised for Compliance.
• The system reminds the Employee of checks which are getting close to service agreement limits, or going over them
• The AI Agent helps the Employee to prioritise work, or if it looks like they are not going to finish it on time, request a transfer of the work over to a colleague.

Company overview

• The Company overview shows the three sections where checks have been carried out – Company Details, People and Financial Records
  • By breaking this into three sections, Employees can check one part before information has been submitted for the others.
• The same issues found, issues reviewed and flags raised are shown for each component of the overview, as well as a summary Red/Amber/Green status
• The AI Agent provides a summary of findings, and can guide the Employee through each section, highlighting areas which require review.

Officer details

• The Person details provide an overview of the officers in the company, as well as people who have been found to be associated with them, allowing the Employee to check any potential Compliance issues such as Politically Exposed Persons, or those on Watchlists, identified through data house APIs.

Aftermath

The designs above was made into a working proof of concept, which led to a successful bid with the client bank.

Retrospective

Project takeaways:

• Overall, I feel like we made a good effort within the limited time we had for the bid, understanding the context and identifying some key points which helped develop an approach which would have made for a better product
• Of course, I would have liked more time to research and test my ideas in more depth, and feel that if I had been able to do so, I would be able to create an even better outcome
• I’m also happy to have had the chance to understand and apply AI to a project, recognising its benefits and limitations to enhance the KYC process.