RBC → Hyper-personalized ai search





RBC’s hyper-personalized ai search

After the successful implementation of the Help Centre and Async Chat, my team and I were tasked with the redesign and development of the search results page.

Dates
Jan 2024 - June 2024

Role
Design Lead

Team
Franz T, Lara S,  Kellie M, Jenica V, Jonathan G

Product Vision
Give each and every client the ability to self-serve by creating a hyper personalized ai powered conversational search experience when searching for information, looking for support, trying to conduct transactions, resolve their inquiries or problems and increase efficiency.

Product Goals for Search Results
- Empower Clients to self-serve
- Hyper-personalize Client Experience for Support, increase accuracy for Search when Clients are trying to resolve their inquiries/problems
- To: increase efficiency for our Clients to get to their end-result (either an answer, or an action that gets them to end of job)
- To: increase confidence in RBC and our experiences, increase satisfaction
- As a result: increase self-serve and decrease costs / need for Advice Centre escalation






Scope
The RBC search results page was utilizing intelli response to generate search results which was driving clients away from online banking search to google and then calling the advice centre resulting in high call volumes and client frustration.

 

HMW support our clients and enable them to self-serve to complete/answer their banking needs, all within search?

Competitor Research


The first step was to understand how search results were displayed across different sites, how conversational search engines were functiontioning and how clients interacted with them.

I took this opportunity to look at You ai, Perplexity, Bing, Chatgpt and Google.





Requirements

As this was a fairly new ask coming from product and business, it was very important to set up meetings with stakeholders to understand what’s in scope, refine the user flow, what use cases we could utilize and highlight limitations if any. I also referenced some of our previous research pertaining to search.

After meeting with all the stakeholders and gathering all the user research I drafted a set of high level requirements.




Unmoderated User Testing


User feedback

From our previous research we knew our clients trusted and used google the most. Therefore we wanted to get user feedback on how they were using google search and conversations search.

Methodology

10 Participants were recruited through the UserTesting Panel for unmoderated interviews.

Criteria - age between 30-50 (based on live data from the Advice Centre), primary online banking channel is web, active on the website.


Findings

Insights

● Google’s generative ai: Customers wanted the information source to be clear. They felt hesitant using generative ai for financial information. Despite the lack of sourcing, they still liked generative summary component as it presented information right away without looking elsewhere.

●  Google’s Search results: The many different sections of the page felt cluttered and customers did not like the long scrolling. They also indicated some results felt redundant. But despite the excess of information customers appreciated the simplification of information and how it was presented on the page.

●  On follow up questions: Customers felt it helped them reword and narrow search results. It also felt like a conversation. The people also ask section provided a similar value.

●  Thumbs up and down: Customers did not notice the thumbs up/down icons but were aware of it. They did not find much use for it. They were also confused whether the thumbs up/down was pertaining to all the search results or one particular section.

Flow Diagram

After we conducted our research we had an understanding of how this new search would function, I mapped the user’s interaction with RBC’s search.



To Be Scenario

We selected the credit card limit increase flow for the wireframing activity.

I drafted a To be Scenario for how this experience would be for our clients.




Client Segmentation

It was also essetial to map out  the tone of the ai along with user’s behavior pertaining to using the ai powered search.


i - User mindset and roles


ii - AI tone



iii - Searching, Conversation & search results





Wireframes Mid-Fi

1: Client Searches for “increase limit”.

2: Client is asked to disambiguate - which limit increase. Client selects credit card limit increase

3: Conversation search prompt showing client has an offer. They can accept/decline or request a higher limit.

4: Toggle between conversation and search results


MVP



After many design workshops and brainstorming sessions we finalized the design of our mvp.

Help Centre was to be the entry point for hyper-personalized conversational search.

Scenario:

Step 1: The client looking to increase their credit card limit would search for “increase limit”






Step 2: As the search is personalized, the client would then be asked to select which product they are looking to increase the limit for.








Step 3: The client is presented a credit card limit offer, which they can accept, decline or inquire more about it using the ask follow up link.

The page also provides a snippet about the limit increase, what other people are asking, faq’s and articles.






Step 4: Client accepts credit limit offer and the ai asks for further confirmation.





Step 5: Client provides confirmation and their limit has been increased.





Usability Testing

Overview

This research sought to determine and validate the feelings and experience of RBC clients when interacting with hyper personalized search. The new RBC search feature will provide the capability to have conversations with clients allowing them to perform tasks and solve their banking needs. Currently clients can converse with RBC’s search on any inquiry pertaining to limits and branch information.

Research Approach

10 moderated in-depth interviews:

●  25-50 y/o

●  Banking with RBC min. 5yrs

●  Min. 3 RBC Products

●  Has other products/accounts with other banks

●  OLB prime users (5) and MOB prime users (5)



Key Findings: Client Expectations






Key Findings: Client Understanding






Key Findings: UI and Visuals





Key Findings: Search output usefulness




Looking Ahead


What we learned

Entry point to the feature is hidden. Clients are enthusiastic about its potential, but don’t know that it’s there.

Inclusion of both conversational search results and traditional search results provide actionable and informational pathways forward.

While some indicated the FAQs and People Also Ask were not relevant, others indicated that without them it may require additional steps to complete their task.

Participants appreciate the confirmation step after accepting or declining the offer, but would like it to feel more “official”

Top toggle, bottom follow-up field and async chat can co-exist on the page, but some refinements could provide additional clarity of what each offers

Participants responded positively to the flow, but suggest it could be more visually engaging



Final Designs