Aaleen
Khattak.

Context

A regional bank wanted to launch an AI-powered digital assistant, but before defining its personality, capabilities, or tone, they needed to understand how customers actually feel about AI in a banking context. I was asked to run the foundational discovery that would shape every downstream decision.

Approach

• 6 in-depth interviews per cohort across three specific cohorts: (a) personal banking users who had contacted customer care for app-related issues, (b) personal banking users who had never contacted care, and (c) account managers for larger businesses - 18 interviews total
• Explored sentiment toward AI in general and specifically for banking service requests
• Probed what an ideal AI assistant would look and speak like, and what tasks it should handle
• Investigated what trust and security mean in the context of banking AI, and how the assistant's personality could address those concerns

Key findings

• Customers now clearly distinguish AI from basic chatbots - but feel uneasy when AI behaves in an overly friendly or casual way
• Inconsistencies across AI platforms (e.g. Gemini vs. ChatGPT giving different answers) have eroded baseline trust
• Users prefer AI for information-seeking over task completion - and expect outputs to need editing. This was especially pronounced for anything involving money, where the sensitivity of the task made AI-driven completion feel less acceptable
• Participants wanted to be spoken to directly, efficiently, intelligently, and courteously - less "fluff"

Customers viewed AI as a tool and expected it to act as such - delivering exactly the information asked for, and all of it.

Outcome

The study shaped the brand identity and personality framework for the bank's digital assistant. I worked closely with UX Content Designers on the tone of voice for the assistant, translating research insights into language guidelines. It contributed to measurable increases in user satisfaction and a reduction in customer care call volume post-launch.

Learnings

This was the first time I saw research translate directly into a tangible product output - working with UX Content Designers to turn findings into tone of voice guidelines for the assistant. It taught me that research doesn't end at the insight; it extends into how those insights are applied, and that collaborating closely with content design is one of the most direct ways research can shape what users actually experience.

Context

The dining vertical team at a super-app approached me to run a usability study on a proposed AI assistant for restaurant discovery. The assistant was already in prototype - but it had been developed without any prior research to explore how users in the region feel about AI.

The proposal

Rather than jumping straight into usability, I proposed expanding the scope: run an exploratory phase first to understand user sentiments toward AI, then carry out the usability testing with those sentiments in mind. I facilitated a workshop with stakeholders to align on what we knew, what we didn't, and what we needed to learn.

Approach

• Step 1: 7 in-depth interviews (4 existing users, 3 non-users) - exploratory
• Step 2: 7 usability testing sessions on the prototype
• Post-research washup session with stakeholders to share top-level insights and start conversation
• Thematic analysis synthesised from quotes and observations

Key findings

• The service was not seen as a discovery tool - users used Google, social media, or word-of-mouth for restaurants and only used the platform for discounts
• The assistant did not provide clear value beyond using filters, and some participants saw typing a query as more effort than pre-selected filters
• The design presented open-text fields resembling established AI tools while messaging set low expectations - creating conflicting user expectations
• AI is relatively new in the region, and any break in the experience caused users to immediately abandon the feature

"Whenever I see a lot of investment needed for a thing, I would probably just drop it and go through the manual process."

Outcome

The research identified that the assistant needed to solve a real pain point - dish-specific recommendations, budget filtering, or review summaries - rather than replicate what filters already did. These insights shaped the repositioning of the feature and informed the approach for future AI releases across the platform.

Learnings

This study taught me how to push back on a brief constructively. I could have just run the usability test as asked, but recognising the missing foundation - and proposing the expanded scope - changed the outcome entirely. It also showed me that the way a feature is positioned matters as much as how it works; the AI assistant's core issue wasn't usability, it was value proposition.

Context

A regional bank was redesigning its business banking app end-to-end. I led the discovery and evaluative research across two phases, with insights feeding directly back into the design process between rounds.

Round 1 - onboarding, cards, payments

• 7 in-depth, task-based interviews (70 minutes each) with SME owners and operations leads
• Found that users expected OTPs to auto-fill and existing customer data to pre-populate after providing a phone number
• National digital ID auto-fill was a delighter - every participant used it - but post-onboarding locked-state task completion felt daunting without effort estimates
• Package and pricing language was confusing - participants misunderstood fee structures and felt blindsided by costs after investing effort in onboarding
• Trade license upload was preferred over manual TL number entry, which could streamline a full step

Round 2 - accounts and spaces

• 7 in-depth, task-based interviews (50 minutes each) on a revised prototype. I got access to Microsoft Bookings between rounds, which cut participant recruitment from 3 days down to 1 - allowing me to increase the sample size and move faster
• Account-level actions like "View details" and "More" felt disconnected - participants read them as general actions rather than tied to a specific account
• Savings were expected in the main accounts view - users mentally model all their money together, not separated by product type

When actions aren't clearly anchored to an object, the breadth of their relevance becomes questionable.

Outcome

Insights shifted product strategy and directly informed information architecture, usability, package models and pricing, and user education strategy for the business banking app.

Learnings

This was a lesson in research operations as much as research method. Getting access to Microsoft Bookings between rounds completely changed what was possible - a 3-day recruitment process compressed to 1 day, freeing up more time to focus on analysis and discussion guide writing. It showed me that operational improvements can have as much impact on research quality as methodological ones.

Context

A super-app was preparing to launch paid add-ons for its subscription program - extra rides with cashback, no minimum order value for food delivery, dining discounts, and a bike family pass. The team needed more than gut-feel pricing; they needed data that commercial teams could defend.

Approach

• Survey with three sections: purchase acceptance and likelihood, Van Westendorp pricing (too cheap, cheap, expensive, too expensive), and add-on appeal ranking
• 1,104 responses across six non-mutually-exclusive user segments, spanning single-service users, multi-service users, and B2B partnership customers
• Tested five distinct add-on concepts with segment-specific appeal analysis

Key findings

• 76% of respondents were willing to pay for removing minimum order value on food delivery - the highest-acceptance add-on across all segments
• Add-on appeal was strongly correlated with usage - ride-heavy users preferred ride add-ons, dining users preferred dining add-ons
• Customers were willing to pay roughly half their subscription value for add-ons, and willingness did not vary based on platform usage intensity - suggesting price-elastic behaviour
• Non-purchase was driven primarily by irrelevance ("wouldn't use it") rather than unwillingness to pay on top of their subscription

If an add-on is relevant to a user's activities, they are usually willing to pay for it. The barrier is relevance, not price.

Outcome

The study produced defensible price ranges per add-on, segment-specific appeal rankings, and a clear recommendation to personalise add-on offerings based on user behaviour. It also flagged that the pricing experience itself would need qualitative testing before launch.

Learnings

This was my first large-scale quantitative study and the first time I used Van Westendorp in a real product context. I learned how to design a survey that serves both research rigour and commercial needs - the commercial team needed numbers they could take into negotiations, not just insights. It also made me more comfortable moving between qualitative and quantitative methods depending on what a question actually requires, and more confident in trying new things when it's needed rather than defaulting to what I already know.

Context

A regional bank was redesigning its personal banking app. Because how people manage their money is fundamentally shaped by how much they earn - affecting lifestyle, available products, and financial literacy - I structured the study around three customer tiers based on income and banking relationship.

Approach

• 15 in-depth interviews (70 minutes each), 5 per cohort across Mass, Select, and Elite tiers
• For the Elite cohort, traditional recruitment channels weren't working - these are high-net-worth clients who don't respond to standard outreach. I pivoted to recruiting through the bank's Relationship Managers, who had direct access. All 5 participants were recruited within hours
• Tested dashboard, payments, notifications, profile, apply, and support flows
• Explored mental models around core banking products alongside usability tasks

Key findings

• The "You have" and "You owe" asset/liability dashboard concept was widely understood and matched how participants think about their money
• 73% of participants misunderstood the eye icon's functionality - guessing "show more details" or a menu, not "hide balances"
• Requesting money from contacts was perceived as culturally inappropriate in the UAE - participants described it as rude or tacky, regardless of the context
• The spending insights feature was seen as a delighter, but participants expected monthly specifics for budgeting - the current format wasn't immediately actionable

The distinction between assets and liabilities was clear and matched how users think about their money. The grouping stayed.

Outcome

The study validated core design concepts while identifying specific interaction, labelling, and cultural issues across all three tiers. Findings and strategic recommendations were presented directly to C-suite.

Learnings

This was the first time UX research was being presented to the bank's C-suite, so I planned responses to the most likely questions before going in. It made a real difference - I could focus on the conversation instead of scrambling for answers. It taught me that planning responses is something I can do for every presentation, not just high-stakes firsts, and it's become a standard part of how I prepare.