Sector
AI driven FinTech
McKinsey & Co
Project Time
3 Months
My Role
requirement gathering,
LLM integration design,
proposals,
wireframing,
Hi Fi Ui design,
testing,
analytics and OKR metrics
Applying for a loan or mortgage requires you to provide the lender or landlord with your financial information. For many users, gathering and curating this financial information can be difficult.
The whole process can be somewhat antiquated, requiring applicants to physically scan and sometimes even fax important documents to established institutions.
Fincap approached McKinsey’s product development team to discover how they could bridge this gap and solve these pain points and create a better solution for their end users leveraging AI and NLM technology to garner confidence and trust in the experience.
Having delivered on similar fintech projects, I spearheaded the initiation and used existing domain knowledge to deliver a solution that produced optimised results for both Fincap as a client and their end users alike.
Competitive Analysis
Securing a loan or mortgage requires users to share their financial data, a task often hard to manage. We still grapple with a centuries-old problem: the need to furnish increasing amounts of information, not all of which are digitised or stored in "the cloud".
We sought to identify the true problem that this presented to users so that we could work towards a solution to this pain point.
Initially, we conducted a competitor analysis to determine the novelty of our concept (spoiler: it wasn't a first). After that, we benchmarked similar concepts with the objective of constructing a superior, more efficient application.
As a general rule: Don’t try to reinvent the wheel!
Survey and Interviews
In the first round of user research, we accessed certain users of Fincap and used surveys to understand the strongest use cases as well as pain points of the platform.
The categories were around: Car Loan, Home Loan, Personal Loan and surprisingly, Apartment Rental.
In synthesising results from 85 participants, there were 7 areas that would need to be addressed however using the Pareto Law, we saw that 80% of pain points came from 2 specific problems:
• Including a guarantor
• Miscommunication
Including a guarantorIncluding a guarantor or cosigner proved tedious. The process of applicants serving as intermediaries between guarantors/cosigners and landlords or brokers led to delays because it involved very slow document coordination and transfer.
MiscommunicationThe key issue observed was frequent miscommunication between borrowers or apartment applicants and lenders or landlords. For instance, one home loan applicant spent a month repeatedly sending documents due to lack of clarity on whether they were received or misplaced by the underwriting agent.
Analysis
We started by organising the gathered data points into an affinity diagram, categorising interview segments into seven groups.
Interestingly, our analysis showed users as either confident or anxious.
Confident users were generally 50-60, with stable careers, sound financial knowledge, aware of their loan eligibility and also with retirement on the horizon.
Anxious users, primarily younger, first home buyers, often with no prior property or loan experience and those looking at first time investment property.
Understanding these user insights informed how we would proceed with designing. Importantly we created 2 user personas to visualise and remind us of who we were designing for.
HypothesisAI automation can foster confidence in several ways:
1. Personalised Guidance:
Using NLM we can provide personalised guidance throughout the loan application process, helping users understand each step and ensure they're filling out information accurately. This can make the process less intimidating and more approachable.
2. Instant Feedback:
An LLM can provide real-time feedback and error checking, reducing the likelihood of mistakes and rejections, which often lead to anxiety and lost confidence.
3. Predictive Analysis:
AI can use historical data to predict the likelihood of loan approval, helping users understand their financial standing and what they can do to improve their chances before submitting their application.
Design
Sketches and Wireframes
We set up design sprints to conceptualise the user's primary navigation paths and LLM prompt and interaction points.
The completion of a detailed flow chart facilitated a comprehensive understanding of the steps in the journey that the integration could facilitate and help users with.
We transferred with AI and analytics team to sense check and because the model was simple enough, it was feasible to build in our given time frame.
ProposalWorking with the product owner to ensure we had a solid hypothesis, backed by user evidence and data points that told us what immediate pain points needed to be addressed, we pitched our concept to stakeholders both in Fincap’s leadership as well as some of the clients Fincap was working with at the time who we identified as power users and SMEs from the user research.
Testing
For our first round of usability testing, we tested with 4 power users that we were lucky to have access to.I gave them five different tasks:
1. Access the dashboard by registering an account
2. Go through the new application process, leveraging the AI assistant we had built
3. Ensure the AI assistant had captured their personal information correctly
4. Come to a decision on application
With our first user, we ran into problems almost immediately.
One of my users pointed out that the flow of the app should actually be less automated, at least when starting the flow for a new application. For example, when you apply for a new application, you can’t choose what kind of offer you get because it’s automatically generated.
We realised we needed to strike the right balance between AI automating everything and a user’s manual inputs.
Prototype 2.0
Utilising the feedback and information we got from our first batch of users, we made a few key changes.
I took the advice of my users, and made it so that when you click “Offers & Rates,” you are now able to see the different categories your offers and rates fall under rather than the system ‘knowing this’.
Now, I’d broken it into four different categories for users to manually select from:
1. Home
2. Auto
3. Credit Card
4. Personal Loan
Results and Next StepsTake Aways
Following on from this, I created handover documentation to our dev team and worked with them around compliance and implementation.
Ultimately we were able to deploy and conducted UATs with over 94% user satisfaction, 87% use confidence and an increase of 25% in user retention over a 2 month period.
The results were encouraging and as the team onboarded more users in the following weeks, those metrics continued to increase in the same trajectory.
I wanted to include this case study in my portfolio because it was for me, a pivotal example of AI enhancing user confidence and streamlining processes in these scenarios. AI offers personalised guidance, instant feedback, predictive analysis, and seamless document management, which together contribute to a less intimidating, more manageable user experience.
At the same time, it's really important to strike a careful balance between AI automation and user control. Users should maintain control over key decision-making elements to foster a sense of agency and trust.
Over-automation may create a disconnect and potentially erode trust, making the user feel sidelined in their own financial affairs.
In essence, a harmonious blend of AI capabilities and user control not only optimises confidence and trust in the processes of workflows that deal with sensitive information but also enhances the overall user experience of the application.
