How to Implement a Great AI Assistant in Finance — Home Loans
How to Implement a Great AI Assistant in Finance — Home Loans
When building an AI assistant for a company in any industry, knowing their customers’ unique needs is key and getting to know them better overtime is what AI does best.
Social27 has been expanding its AI solutions to a number of different industries. Recently, I’ve been working on designing an assistant for a large banks home loan division. The pilot is out after months of scripting, so I thought I’d share the experience, some challenges that arose and how we addressed those challenges.
My hope is that this use case will give you some ideas on how to best implement an AI assistant in your business and prepare you to know what to expect throughout the design process. It’s also a great way to showcase how AI assistants can be customized to a specific industry, company and customer.
My Approach to Designing a Home Loan AI Assistant
When designing an AI assistant for a new client, the most important place to start is understanding top customer needs and most frequently raised questions and concerns. What this essentially comes down to is identifying the right scenarios, or use cases, for the assistant.
How do we start this identification process? We start by talking with the client about what they view as the top customer scenarios. This helps us understand the top priorities from the company and employees’ perspectives and what their goals are. The next step is to get an understanding of the customers’ perspective. To do this, we looked at the data.
What the Data Told Us
To uncover the most potent scenarios to script in the AI assistant’s initial design, we looked at historical records that showed us what kind of inquiries the company gets most frequently. For example, a lot of inquiries were around certain types of home loans. This tells us that we should include information on different types of home loans in the assistant’s prompts.
Furthermore, we were able to uncover that customers inquiring about particular home loan types most often opted to talk to an employee for further information. So, we made sure to add an option for customers to talk to an expert so the assistant can get them in touch with an employee right away or scheduling a future call time that works both for the customer and the right employee.
We also analyzed their existing entities for guidance with the assistant. This AI assistant has an audience international to us (we’re based in the US) but needed to resonate with the local user. To make sure the assistant aligns with the company’s branding and users on a regional level, we analyzed the company’s website, email, blogs and other assets to find what commonalities were resonating with users worldwide and leverage those successful messages for our AI assistant design.
Top Home Loan Scenarios for our Client’s Business
After analyzing the data, we were able to identify a number of top customer scenarios that align with our client’s AI assistant goals. The three most popular scenarios are applying for a home loan, meeting with a loan expert and learning more about home loans.
1. Apply for a home loan
This scenario is probably a given, but there are many variables around home loans that make this scenario a bit more complex. When a user selects this prompt, they are first asked to identify the type of home loan they’re looking for or whether they are looking for particular information about home loans.
If home loan is selected again, four additional prompts come up asking if they are looking for a home extension, top up loan, etc. Once they select the type of loan, they’re interested in applying for, they are able to either get more information about that home loan type or talk to a loan expert.
2. Meet with a loan expert
In a few simple steps, any user who’s ready to talk with an expert can get connected with the right expert in a way that works with their schedule. First, the assistant will ask for an email address to send where it can send the meeting invite to.
Next, it lists all the dates experts are available, followed by the times they are available to have a meeting.
Finally, the assistant has the user check all their information and change anything they would like to change before booking the meeting.
3. Learn more about home loans
Upon selecting “Learn more about home loans” a Playlist is launched. Social27 Playlists are essentially a streaming service that organizations can compile marketing assets and informational material that utilizes a recommendation engine to populate the most relevant content for each individual user. This allows our client’s customers to get as much information they want about our client and what they do all in one place at their convenience and only talk to an employee, expert or sales representatives when they’re ready.
Here’s our Playlist currently on the Social27.com page:
Back to the Data
To test the efficacy of the bot, as well as start the learning process for the algorithms, we do pilots with small sample customer bases and involve customer success departments for guidance on designing the best experience possible. Through set intents, the platform measures the most used paths as well as sentiment from conversations with loan experts. This data can be used to alter conversation copy, remove paths that are not of interest and design new paths.
To give you a detailed view of the data available to our customers, I’m going to outline how we at Social27 utilize our dashboards to better understand the conversations happening inside the AI Assistants and the content being consumed in the Social27 Playlists.
The “Overview” section lets you easily see the number of new conversations, contacts, meetings booked and customer satisfaction levels. This high-level view lets you easily see spikes or dips in conversations and meetings, as well as the overall satisfaction of the users interacting with the assistant.
This can help you find out who sent out the link to the campaigns, what new content was added or other actions that were taken during key moments that effect these spikes and peaks.
The “Team Progress” section lets you see how your agents are doing and rank their performance via a leaderboard. This section also graphs the number of scheduled meetings in a given time period.
The “Conversations Sentiment” section shows you the top positive, negative and neutral keywords that have been used during conversations with the assistant. We analyze the conversations between agents and the end user, as well as the bot and end user. This analysis spits out a list of words categorized in negative, positive and neutral.
Analyzing this data has helped us identify certain words to change in our conversations’ copy, as well as guidance to give to agents on what words resonate best with the end users and what words don’t.
The last section of the conversation’s dashboard is the “Customer Intent” section. When setting up a campaign, we define intents on our platform. These are keywords that most resonate with the campaign objectives. Once the AI Assistant and Playlist is launched, our platform tracks these intents and displays their usage (as shown below) so you know what you customers are most interested in and what is not resonating with buyers.
The next dashboard I’ll go over is the Playlist Dashboard. With this dashboard, we can track how users are interacting with our content inside our Playlist.
The Playlist dashboard is split into 3 sections:
Playlist Content and
The overview gives you key stats to gauge the health of your playlist including: unique visitors, total view time, total sessions, incremental view time — which is the view time after the first piece of content in the playlist, form submissions and average view time. This will help you compare the efficacy of various Playlists so you can take best practices from your best performing Playlists and incorporate those practices into existing and new ones.
Moving along the rest of the overview tab, we have graphs and tables that highlight top visitors, their distribution/activities and top content views. This is valuable information for your sales team so when you pass on a qualified lead to sales, they can ensure they’re having the most relevant conversation with those buyers.
The second tab in the Playlist Dashboard is the “Playlist Content” tab. Here you can see how many total views, average view time and likes that each piece of content in the playlist is getting. This will help you gauge which content types perform best and which aren’t as effective.
You can click on a piece of content to drill further into its individual stats. Here we can see key stats such as the cost per minute, ROI and visitor engagement so you now which efforts you should invest more in and which you can leave behind.
Finally, there’s the “Visitors” tab. This tab shows a table of all the interactions taken by each unique visitor including visit time, location, device, engagement score, total views and more. Clicking on a row will show engagement for all the content a user has consumed to ensure a highly relevant, personalized, ongoing conversation.
The final dashboard I’ll dive into is the Hive Dashboard. The Hive dashboard is the centralized control room where you can monitor, secure and manage your entire swarm (AI Assistants + Playlists) at scale. You will get valuable insights and ROI data on campaigns. Insights from customer conversations, content engagements across your entire channel delivered in a secure and compliant environment.
Social27 is an AI powered Automation, Augmentation and Analytics platform. Social27 Playlists empower B2B marketers to deliver a personalized and ‘binge-worthy’ content experience in real-time. Our AI Assistant identifies buyer intent, qualifies leads, and schedules meetings; accelerating sales and revenue growth. Social27 Deal Room automates all the repetitive and time-consuming processes in the deal workflow, keeping all calls, contracts, and compliance in one place. And finally, the Partner Ecosystem Accelerator enables velocity across your ecosystem by enabling a #NoFriction customer journey.