How I led Pathao's research initiative to build a personality-driven intelligence layer - designing VALS-based psychographic segmentation from scratch using employee pilots, gamified surveys, and clustering algorithms to unlock what behavioral data alone never could.
Pathao is Bangladesh's dominant super-app - rides, food delivery, and logistics serving millions of daily users across Dhaka and beyond. By 2025, the behavioral data was rich: spend patterns, ride frequency, food preferences, cross-service usage. The personalization engine was good at serving users based on what they had done before.
But a recurring gap kept surfacing in strategy discussions. Two users with identical spending patterns might order for completely different reasons. One buys from North End every day because they're time-poor professionals who need reliable food fast. Another orders from the same restaurant because it's the aspirational brand they associate with their lifestyle. Same behavior, radically different psychology - and therefore different ideal messages, promotions, and product experiences.
This is the psychographic gap: behavioral data tells you what people do, but not why they do it. And the "why" is what determines how to speak to someone, what product experience feels right for them, and which promotions will actually convert vs. get ignored.
The core insight driving this initiative: Behavioral clustering groups users by past actions. Psychographic clustering groups them by underlying personality, values, and lifestyle orientation - the drivers that explain why those behaviors exist and predict how they'll evolve. Layering both gives Pathao a significantly stronger intelligence foundation for personalization, cross-sell, and even credit decisions for BNPL users with thin financial histories.
This was not a standard feature build. There was no existing dataset to analyze, no validated instrument to adopt off the shelf, and no precedent within Pathao for this kind of research-to-product work. I owned it end to end: from framing the business case, to designing the research methodology, to building the application framework that connects psychographic segments to real product decisions.
I didn't start with a solution looking for a problem. I started by mapping where current personalization was visibly failing - where knowing what users had done was insufficient to know what to do next for them.
Most product research fails at the implementation stage because the gap between "interesting findings" and "product decisions" is never properly bridged. I designed this initiative so that every research output fed directly into a product decision - no insights sitting in decks without owners.
A key early decision: Start with Pathao employees, not users. Internal pilots let me test the survey instrument, validate clustering logic, and identify question weaknesses without the consequences of a broken experience reaching paying customers. It also created a natural "friend matching" use case that built internal buy-in and generated genuine engagement data before a single line of production code was written.
VALS (Values, Attitudes, and Lifestyles) is a psychographic segmentation framework with decades of academic validation and commercial application. It segments people across two primary dimensions: primary motivation (ideals, achievement, or self-expression) and resource levels (income, education, confidence, health). The intersection produces eight segments.
I chose VALS over other frameworks (Big Five personality traits, Maslow's hierarchy, OCEAN) for three reasons: it directly maps to consumption behavior rather than abstract personality; it has demonstrated commercial validity in marketing contexts; and its segment definitions translate meaningfully to Bangladeshi consumer culture with moderate adaptation.
A deliberate scoping decision: The standard VALS model has 8 segments. I scoped the Pathao implementation to 6, excluding "Makers" (DIY-oriented, minimal relevance in food delivery) and merging "Innovators" into the Experiencer profile for this context (the two segments overlap significantly in Pathao Food usage patterns). Fewer, cleaner segments produce better ML clustering performance and more actionable personalization rules.
Survey design for psychographic clustering is not the same as a standard UX research survey. Every question must do double duty: feel natural and engaging to the respondent, while precisely mapping each answer option to a specific psychographic dimension. I owned the question design, the answer mapping logic, and the scoring framework.
Design principles I set: Maximum 15 questions (completion rate drops sharply beyond this). All mandatory - partial responses corrupt the clustering vectors. Mixed question tones - some fun/cultural (superhero alter ego, rainy day food), some direct behavioral (payment method, restaurant exploration style). Answer options must cover all 6 target clusters without any cluster being un-selectable in any given question.
The full 15-question instrument covers five psychographic dimensions: service usage and lifestyle preferences (4 questions), social and communication style (3 questions), decision-making and values (4 questions), lifestyle and personal preferences (2 questions), and urban living and convenience orientation (2 questions). Each dimension contributes meaningfully to the cluster assignment without overloading any single behavioral dimension.
The core clustering challenge is that you can't wait for 400+ responses to assign a cluster to the first respondent - but you also can't lock in rigid rules without data validation. I designed a two-stage methodology that handles both constraints.
Every answer option maps to a cluster. When a respondent completes the 15 questions, each answer increments a counter for its mapped cluster. The cluster with the highest score is the primary assignment. The 6D vector is preserved for all respondents.
This simulation illustrates how the predefined rules work before any ML clustering is applied. Each number represents the count of answers mapping to that cluster across 15 questions.
| Respondent | Striver | Achiever | Thinker | Experiencer | Believer | Survivor | Assigned | Ratio |
|---|---|---|---|---|---|---|---|---|
| A | 4 | 4 | 6 | 1 | 0 | 0 | Thinker | 6:4:4 |
| B | 0 | 1 | 0 | 9 | 0 | 5 | Experiencer | 9:5:1 |
| C | 1 | 6 | 1 | 2 | 3 | 1 | Achiever | 6:3:2 |
| D | 0 | 0 | 0 | 0 | 4 | 11 | Survivor | 11:4:0 |
| E | 6 | 2 | 3 | 2 | 0 | 0 | Striver | 6:3:2 |
Once 400+ responses are collected from the employee pilot, the predefined rules are stress-tested against data-driven clustering. The 6D response vectors are run through K-Modes clustering (the correct algorithm for categorical data - unlike K-Means, which assumes numerical distance) and hierarchical clustering to see whether natural clusters emerge that match the predefined segments.
When the algorithmic clusters match the predefined segments, this confirms that the question-to-cluster mapping is correct and that the predefined groups are meaningful real-world categories rather than arbitrary constructs.
Divergence signals either that a question's answer options don't cleanly map to the intended cluster, or that two of our predefined segments are actually the same cluster in the data. These divergences inform the survey instrument refinement before production rollout.
The KNN fallback: For respondents who fall at the boundary between two clusters (e.g., Respondent E above: Striver 6 vs Thinker 3 is a meaningful gap, but Respondent C's Achiever 6 vs Believer 3 might warrant review), a KNN (K-Nearest Neighbors) pass on the full response vector provides a distance-based assignment that's more nuanced than the simple maximum-score rule. The 6D vector preserved for every respondent makes this possible without re-running the survey.
Research that doesn't change product behavior is not product research - it's academic. I built the application framework to ensure every segment definition has a corresponding action across the three primary Pathao Food surfaces where personalization matters.
The food tile sort order on the Pathao homepage is the highest-leverage personalization surface - it's the first thing every user sees. Psychographic segment determines which restaurant types appear first, not just behavioral recency or rating score.
| Segment | Homepage Feed Strategy | Rationale |
|---|---|---|
| Achiever | Fast, reliable chains first. KFC, North End, established brands with strong delivery SLA. "Your regular order" shortcut prominent. | Achievers optimize for time and reliability. Novelty is noise. Surfacing their established patterns first removes friction from the repeat-order flow. |
| Experiencer | Trending and viral menus at the top. "New this week" and "Everyone's ordering" labels. Flash deal countdown timers visible immediately. | FOMO is the primary driver. Social proof and time pressure convert Experiencers more than discount depth. The feed should feel like a live trend feed, not a menu. |
| Thinker | Healthy restaurant section elevated. Nutrition tags visible on food tiles. Curated "Editor's selection" with rationale (not just rating). | Thinkers research before deciding. They need information architecture that supports rational evaluation - not emotional triggers. Nutrition data is a feature, not a filter. |
| Believer | Local neighborhood restaurants first. Bengali cuisine prioritized. Prayer-time meal alerts (iftar, sehri). "Your neighborhood's favorites" section. | Believers value familiarity, community, and tradition. The feed should feel like a trusted local guide, not a discovery engine pushing unfamiliar brands. |
| Striver | Premium brands at discount. "Normally X, today Y" price framing. BNPL eligibility surfaced on aspirational restaurant tiles. | Strivers want access to aspirational brands without the full price. Making the gap feel closeable (discount + BNPL) is the conversion trigger, not the food itself. |
| Survivor | Lowest-cost nearby vendors first. Free delivery options prominently labeled. No-minimum-order restaurants surfaced. Micro-promo (Tk 10–30) visible immediately. | Price is the only decision variable. The feed needs to surface the maximum value options without any browsing friction - every scroll is a potential dropout. |
Beyond what to show, psychographic segments determine how to say it. The same offer lands differently depending on the copy tone, urgency framing, and social context used.
Psychographic clustering also feeds into Pathao's BNPL underwriting model for thin-file users. Specific question answers correlate with financial behavior dimensions that traditional credit scoring uses financial history to capture - allowing psychographic data to serve as a proxy signal when transaction history is insufficient.
This answer maps to Achiever or Thinker - both associated with higher financial conscientiousness and lower default risk. In thin-file users, this single response meaningfully shifts the prior for creditworthiness.
This maps to Experiencer or Striver - clusters with higher risk appetite and impulse-driven spending. For BNPL underwriting, this is a negative signal that increases the default risk prior, enabling tighter credit limits for this cohort.
Measuring a psychographic clustering initiative is more complex than measuring a feature launch. The primary output - a richer user intelligence layer - only creates value when downstream systems consume and act on it. I defined metrics across three horizons: data quality, engagement, and business impact.
| Horizon | Metric | Why it's on the dashboard |
|---|---|---|
| Data Quality | Psychographic coverage - % of active users with a cluster assignment | Personalization only works where data exists. Coverage is the precondition for every downstream application. Low coverage means the system is personalized for a minority of users. |
| Data Quality | Survey completion rate - % of quiz starters who complete all 15 questions | Partial responses corrupt the clustering vectors. A completion rate below ~70% signals that the survey is too long, too confusing, or not engaging enough to see through - all solvable product problems. |
| Engagement | Quiz viral coefficient - shares per quiz completion | The gamified result-sharing mechanic is both the growth engine and a signal that users found the output meaningful enough to broadcast. A high viral coefficient means low CAC for psychographic data collection. |
| Engagement | CTR lift on psychographically-targeted promos vs. behavioral-only baseline | The most direct test of whether psychographic clusters are producing better-matched messaging. If targeted promos don't outperform the behavioral baseline on CTR, the segmentation is either wrong or not being applied correctly. |
| Business | BNPL approval rate uplift for thin-file users with psychographic scores | Psychographic data's most commercially significant application. If approval rates increase without a corresponding increase in default rates, the clustering is genuinely adding predictive signal to the credit model. |
| Business | Retention - return order rate within 7 days for psychographically-personalized users vs. control | The feed personalization application has a compounding retention effect. Users who consistently see food options aligned with their personality are more likely to order again, more quickly. This metric captures the long-term business case. |
The Business Concept Document, questionnaire framework, and cluster mapping logic - as produced during the research initiative.
| Section | Content |
|---|---|
| BCD.1 | Problem: Behavioral data (spend, frequency) reveals what users do but not why - limiting personalization beyond transactional history. Psychographic gap prevents delivery of true lifestyle-aligned experiences. |
| BCD.2 | Primary objective: Develop psychographic segments and map users into clusters at scale. Metrics: CTR on personalized promos, BNPL approval rate for thin-file users, psychographic data coverage %, quiz engagement rate, behavioral prediction accuracy lift when psychographic traits are layered. |
| BCD.3 | Evidence for prioritization: Behavioral limits (spend/usage don't explain "why"), BNPL credit context (thin-file users need alternative signals), engagement potential (gamified quizzes are sticky and drive peer comparison virality). |
| BCD.4 | Solution: Lightweight psychographic segmentation powered by surveys. Questions map to Big Five/VALS models. Hierarchical clustering or factor analysis defines segments. Clusters cross-referenced with behavioral data and tagged to user profiles for use in recommendation systems and credit engines. |
| BCD.5 | Business impact: Higher marketing ROI from accurate personalization, greater conversion on cross-vertical campaigns from better understanding of user values, BNPL product expansion without compromising default risk, enhanced brand stickiness from gamified interactions. |
| # | Question | Psychographic Intent | Cluster Mapping |
|---|---|---|---|
| Q1 | Primary motivation for ordering food online | Core value driver identification | A→Achiever · B→Thinker · C→Experiencer · D→Survivor |
| Q2 | Willingness to try unfamiliar restaurants | Risk tolerance and novelty appetite | A→Experiencer · B→Achiever · C→Believer · D→Survivor |
| Q3 | Most important meal consideration | Benefit prioritization | A→Thinker · B→Striver · C→Experiencer · D→Survivor |
| Q4 | Group ordering frequency | Social orientation | A→Experiencer · B→Achiever · C→Believer · D→Survivor |
| Q5 | Key app loyalty factor | Brand relationship depth | A→Achiever · B→Striver · C→Experiencer · D→Believer |
| Q6 | Cuisine selection approach | Cultural openness | A→Striver · B→Believer · C→Believer · D→Survivor |
| Q7 | Health consciousness level | Ideals strength | A→Thinker · B→Achiever · C→Experiencer · D→Survivor |
| Q8 | Decision-making process | Information processing style | A→Thinker · B→Achiever · C→Believer · D→Survivor |
| Q9 | Food trend adoption | Innovation readiness | A→Achiever · B→Striver · C→Believer · D→Survivor |
| Q10 | Preferred payment method | Digital adoption and resources | A→Achiever · B→Striver · C→Believer · D→Survivor |
| Q11 | Festival/holiday ordering behavior | Tradition vs. convenience | A→Achiever · B→Believer · C→Striver · D→Survivor |
| Q12 | Environmental consideration in orders | Values consistency | A→Thinker · B→Achiever · C→Striver · D→Survivor |
| Q13 | Premium restaurant frequency | Resource level indicator | A→Experiencer · B→Achiever · C→Striver · D→Survivor |
| Q14 | Meal planning style | Structure vs. spontaneity | A→Thinker · B→Experiencer · C→Believer · D→Survivor |
| Q15 | Attitude toward sharing food experiences | Social expression | A→Experiencer · B→Achiever · C→Believer · D→Survivor |
How a completed quiz response becomes a psychographic segment tag on a user profile and drives personalization decisions across Pathao's product surfaces.