BigBasket Upselling Optimization

First, could you clarify what BigBasket is primarily trying to optimize for with this checkout carousel? Is it maximizing total profit per order, increasing overall upsell conversion rate, improving customer satisfaction, growing the average basket size, or strategically promoting private labels?

1. Current State Analysis & Understanding:

• Private Label Performance:
  • What is the current upsell conversion rate of the private label carousel (i.e., % of users who reach checkout and add at least one item from it)?
  • What's the average revenue and, more importantly, the average profit generated per order from these private label upsells? (Profit margin on private labels is key here, often 25-40% higher than national brands).
  • What are the top-performing private label categories/products in this carousel?
  • Are there any customer satisfaction scores (CSAT) or return/complaint rates specifically linked to these private label upsell purchases?
• Telugu Market Context & Brand Insights:
  • Which specific Telugu household brands (Aashirvaad, Tata Salt, MTR, Priya, etc.) and product categories (e.g., atta, salt, spices, pickles, instant mixes) show high purchase frequency and volume among our target Telugu-speaking customer base in AP/Telangana? This can be derived from overall sales data.
  • What are the typical profit margins for these popular brands? This will be lower than private labels.
  • Are there known cultural preferences or strong brand loyalties for certain essential items in this region?

2. Hypothesis Formulation:

• Primary Hypothesis: Displaying popular Telugu household brands in the checkout carousel will lead to a higher upsell conversion rate compared to private label products, but the lower margin per branded item might result in a lower, similar, or higher overall net profit per order from the carousel. The net effect on total order profit is what we need to determine.
• Secondary Hypotheses:
  • Customers in Telugu-speaking regions exhibit higher trust and familiarity with established regional/national brands for certain categories, leading to more impulse additions.
  • The perceived value or necessity of these popular brands might overcome the "upsell" nature of the carousel better than less familiar private labels.
  • There might be specific categories where brand preference is exceptionally strong (e.g., atta, specific spices) versus categories where private labels are more acceptable (e.g., basic staples if price is a driver).

Experimental Design: Checkout Carousel Optimization

Objective: Maximize Net Profit Per Order from the checkout carousel, with CSAT as a guardrail.

Target Audience for Test: BigBasket mobile app users in AP/Telangana who reach the checkout page.

Test Groups (Variants):

• Group A (Control): Current carousel displaying only BigBasket's private label products.
• Group B (Treatment - Branded): Carousel displaying only popular Telugu household brands (e.g., a curated list of 5-10 top-moving SKUs from Aashirvaad, Tata Salt, MTR, Priya, etc., relevant as impulse/reminder buys).
• Group C (Treatment - Mixed): Carousel displaying a mix of private label products and popular Telugu brands (e.g., 50/50 split, or perhaps alternating).
• (Optional) Group D (Treatment - Personalized/Dynamic): If technically feasible quickly, a carousel that dynamically shows either private label or branded items based on user's past purchase history, items in cart, or segment. This is more advanced. For now, let's focus on A, B, C.

Randomization: Users would be randomly assigned to one of these groups upon reaching the checkout session. The assignment should be sticky for the user for the duration of the test to ensure consistency, or at least for a significant period (e.g., 1 week).

Duration: I'd recommend a minimum of 4 weeks. This allows us to capture weekly purchase cycles and gather enough data for statistical significance. Power analysis based on current conversion rates and expected effect size would refine this. If the current conversion is low, we might need longer or a larger user base.

Primary Metric to Optimize:

• Net Profit per Checkout Session with Carousel Interaction: (Total Profit from items added via carousel) / (Total Checkout Sessions exposed to carousel). This directly measures the profit impact.

Secondary Metrics to Track:

• Carousel Upsell Conversion Rate: % of users adding at least one item from the carousel.
• Average Items Added from Carousel per converting user.
• Average Revenue from Carousel Items per converting user.
• Overall Average Order Value (AOV).
• Overall Net Profit per Order (including main cart and carousel). This is crucial.
• Private Label Sales Share: To see impact on PL promotion.
• Cannibalization Rate: (More on this later).

Guardrail Metrics:

• Customer Satisfaction (CSAT): Measured via post-order surveys, potentially with a question about checkout experience or product suggestions. We need to ensure it doesn't drop significantly for any group.
• Checkout Completion Rate: Ensure the new carousel doesn't negatively impact the overall checkout process.

Segmentation for Analysis: After the test, we'd analyze results not just overall but also by:

• New vs. Returning Customers.
• Customer Value Segments (High, Medium, Low spenders).
• Primary shopping categories of the user (e.g., users who mostly buy staples vs. gourmet).

Measuring and Mitigating Cannibalization:

1. Measurement Strategy:

• Tracking Product Substitutions:
  • We need to log cart modifications after the checkout carousel is displayed. Specifically, track if a user removes an item from their main cart that is similar (same category or a direct substitute) to an item they add from the carousel.
  • Example: User has "BB Royal Atta" in cart, sees "Aashirvaad Atta" in carousel, removes BB Royal and adds Aashirvaad. This is direct cannibalization.
• Category-Level Impact Analysis:
  • Compare the total sales (revenue and profit) of categories featured in the carousel (e.g., atta, salt) across the different test groups. We need to see if the overall category profit increases, not just carousel profit.
  • If Group B (Branded) shows high carousel sales for Aashirvaad atta but a corresponding drop in main cart atta sales (both BB Royal and Aashirvaad added earlier), the net gain might be small or negative.
• Total Basket Profitability (The Ultimate Metric):
  • The most crucial metric is the change in the total net profit of the entire order for users exposed to each carousel variant. This inherently accounts for cannibalization. If Group B's total order profit is higher than Group A's, then even if some cannibalization occurred, the overall effect was positive.
• Items Per Order & Basket Diversity:
  • Track if the carousel leads to an increase in the total number of unique items or categories per order, or if it just shuffles items around.

2. Mitigation Strategies (Design of Carousel Content):

• Focus on Complementary Products: Prioritize showing branded items in the carousel that are complementary to what's typically in a user's cart, rather than direct substitutes for high-volume items. E.g., if user has rice and dal, show ghee or pickles instead of another brand of rice.
• Offer Different Pack Sizes/Variants: If Aashirvaad 5kg atta is a common main cart item, the carousel could feature Aashirvaad 1kg atta (for smaller top-up needs) or a specialized variant like "Aashirvaad Select Atta."
• Introduce Impulse or Discovery Categories: Use the carousel for items users might not actively search for but would add on impulse if reminded – e.g., MTR ready-to-eat mixes, new snack brands, or seasonal specialties.
• Smart Carousel Logic (If using Group D - Personalized):
  • The recommendation engine should be designed to explicitly AVOID showing items identical or very similar to what's already in the user's cart in a higher quantity/value.
  • It could prioritize items from categories the user frequently buys but hasn't added in this session, or items commonly bought together with what's in their cart (market basket analysis).

By measuring the total basket impact and designing the carousel content thoughtfully, we can get a true picture of whether popular brands are additive or merely substitutive.

Cultural Intelligence Framework for Carousel Product Selection:

1. Festival-Driven Assortment:

• Sankranti: Feature items like good quality ghee (for sweets), jaggery, sesame seeds, pooja items, new apparel (if BigBasket sells them), or even specific traditional snack ingredients.
• Ugadi: Ingredients for Ugadi Pachadi (jaggery, neem flowers, tamarind, raw mango), new clothes, items for home cleaning/decoration.
• Varalakshmi Vratam / Other Poojas: Pooja samagri, flowers, specific fruits, coconuts, traditional sweets or ingredients for them.
• Ramadan/Eid: Dates, seviyan (vermicelli), dry fruits, ingredients for biryani or haleem.

2. Regional Staple Variations & Preferences:

• Rice: While Sona Masoori is common, certain regions might have preferences for specific local varieties or aged rice for better taste. The carousel could feature smaller packs of these "connoisseur" choices.
• Oils: Groundnut oil, sunflower oil are prevalent. Sesame oil (nuvvula nune) and Ghee are culturally important for certain dishes and occasions. Consider offering trusted local brands of these.
• Pickles (Ooragayalu): Extreme brand loyalty here. Priya is dominant, but also consider popular local pickle makers if BigBasket can source them. Offer seasonal pickle varieties (e.g., mango, gongura).
• Spices & Powders (Podulu): Beyond generic Tata Salt or MTR masalas, consider popular local "podi" varieties (kandi podi, karivepaku podi) or regional spice blends.
• Breakfast Items: MTR is pan-South, but are there specific Telugu brands for idli/dosa batter, upma rava, or traditional breakfast accompaniments (e.g., specific chutney powders)?

3. Consumption Occasions & Lifestyle:

• Summer: Items for making cooling drinks (nannari शरबत, raw mango for panakam), buttermilk ingredients, ORS.
• School Season: Healthy snack box items, breakfast cereals popular in the region, stationery (if applicable).
• Weekend/Family Meals: Ingredients for special weekend dishes, perhaps a popular local brand of ice cream or sweets.

4. Sourcing & Data Validation:

• Analyze Sales Data: Identify which specific SKUs of Aashirvaad, MTR, Priya, etc., are top sellers in AP/Telangana. The carousel should feature these proven winners.
• Local Team Insights: Work with BigBasket's regional procurement and marketing teams in AP/Telangana. They will have invaluable on-the-ground knowledge of emerging local brands or specific preferences.
• Customer Surveys/Feedback: Periodically survey Telugu customers about brands they'd like to see or items they often forget and would appreciate a reminder for at checkout.

5. Presentation in Telugu (Optional but impactful):

• If the app supports localization, ensuring the product names in the carousel are accurately displayed in Telugu script for these culturally specific items could enhance appeal.

The carousel should feel like a helpful, culturally aware reminder service, not just a random assortment of products. This deep localization can build significant customer affinity.

Strategy for Handling Conflicting Results:

My goal remains to maximize net profit per order while protecting CSAT.

1. Deep Dive Segmentation Analysis:

• The first step is to rigorously analyze the A/B test results across multiple dimensions:
  • Customer Segments: Price-sensitive vs. quality-conscious (can be proxied by past purchase behavior, AOV, discount redemption), new vs. loyal, urban vs. tier-2/3 cities within AP/Telangana.
  • Product Categories: Staples (atta, rice, salt), impulse buys (snacks, beverages), high-margin private label strongholds vs. categories with strong brand dominance.
  • Time of Month: Behavior might differ early month (post-salary) vs. end-of-month.

2. Develop a Decision Matrix or Rule-Based System:

Based on the segmented results, we can create rules for carousel content. For example:

IF User_Segment = 'Price_Sensitive' AND Category_In_Carousel = 'Salt'
THEN Show 'BB Royal Salt' (if it won for this segment)
ELSE IF User_Segment = 'Quality_Conscious' AND Category_In_Carousel = 'Atta'
THEN Show 'Aashirvaad Atta' (if it won for this segment)
ELSE // Default or other rules

3. Category-Specific Strategy Rollout:

• Branded Wins (High Conversion & Acceptable Profit): For categories where popular Telugu brands significantly outperform private labels in conversion and the net profit impact is positive or acceptably neutral (due to much higher volume offsetting lower margin per item), switch to branded products in the carousel for the general AP/Telangana audience.
  Example: If Aashirvaad atta has 3x conversion of BB Royal atta, even at half the margin, it might be more profitable.
• Private Label Wins (Good Profit, Acceptable Conversion): For categories where private labels perform well (decent conversion, excellent margin), continue featuring them. This might be true for items where brand loyalty is lower or price is the main driver.
  Example: Basic sugar or certain dals.
• Mixed Results / Segment Dependency: This is where personalization (Group D from my earlier design) becomes powerful.
  • If technically feasible, implement a personalized carousel that shows branded items to segments that prefer them and private labels to segments that respond better to them or are more price-sensitive.
  • If full personalization isn't immediately possible, we might use a hybrid carousel (Group C) that features both, perhaps prioritizing the display order based on overall segment performance or AOV.

4. Strategic Trade-offs (Profit vs. Long-term Value):

• If branded products increase CSAT and potentially long-term retention/LTV, but slightly decrease immediate profit per order, the business needs to make a strategic call. We might accept a small dip in immediate carousel profit if it demonstrably improves customer loyalty in a key market. This requires tracking LTV impact for cohorts exposed to different carousels over a longer period (3-6 months post-test).

5. Continuous Learning & Iteration:

• The carousel content shouldn't be static. Regularly re-evaluate based on new sales data, emerging brands, changing customer preferences, and ongoing A/B tests for specific categories or segments.

The key is to move away from a one-size-fits-all approach and towards a more dynamic, segmented, and eventually personalized strategy for the checkout carousel, always guided by the net profit per order and CSAT guardrails.

Framework for Ensuring Statistical Rigor & Isolating Impact:

1. Rigorous Experimental Design (Foundation):

• Proper Randomization: Ensure users are truly randomly assigned to control and treatment groups at the point of exposure (e.g., checkout page load). This is the most fundamental way to balance out known and unknown confounding variables between groups.
• Sufficient Sample Size & Duration: Use power analysis to determine the sample size needed to detect a statistically significant effect, accounting for baseline conversion rates and desired Minimum Detectable Effect (MDE). A longer duration helps average out short-term fluctuations.
• Concurrent Execution: All test groups must run simultaneously to ensure they are exposed to the same external conditions (seasonality, competitor actions, macro trends).

2. Control for Internal Confounding Factors:

• Holdout on Major Campaigns: If possible, try to schedule the A/B test during a period with no other major overlapping A/B tests or large-scale marketing campaigns targeting the same user base or checkout flow. If unavoidable, ensure these campaigns are applied uniformly across all test groups.
• Track App Updates: Log any significant app updates deployed during the test period. If an update impacts the checkout flow or overall app performance, its effect should ideally be uniform if randomization is working. Analyze pre/post update within each group if a major change occurs mid-test.
• Inventory Consistency: Ensure that products featured in any carousel variant (private label or branded) have consistent stock availability. Out-of-stock issues in one group can skew results.

3. Statistical Techniques for Adjusting for Covariates (Post-Hoc Analysis if needed):

• Difference-in-Differences (DiD): This is powerful if we have pre-test data for the metrics.
  • Compare the change in the outcome metric (e.g., profit per order) from the pre-test period to the test period for the treatment group, versus the same change for the control group.
  • (Change_Treatment) - (Change_Control) = Estimated Impact.
  • This helps control for time-based trends that would have affected both groups anyway.
• Regression Adjustment / CUPED (Controlled-experiment Using Pre-Experiment Data):
  • Use pre-experiment data (e.g., user's past AOV, purchase frequency) as covariates in a regression model to reduce variance and increase the precision of the treatment effect estimate. This makes it easier to detect smaller true effects.
  • The model would look something like: `Outcome_Metric ~ Treatment_Group + Pre_Experiment_Metric_Value + Other_Covariates`. The coefficient for `Treatment_Group` gives the adjusted impact.

4. Segmentation & Subgroup Analysis:

• Analyzing results by segments (new vs. old users, different cities within AP/TG) can sometimes reveal if external factors disproportionately affected a particular subgroup, helping to understand variability.

5. Qualitative Insights:

• Monitor customer feedback channels during the test. Any unusual spikes in complaints or comments related to pricing, product availability, or marketing campaigns can provide context if results are unexpected.

No method is perfect, but a combination of a strong upfront experimental design and appropriate statistical adjustments post-test gives us the best chance of isolating the true impact of the carousel change with a high degree of confidence.

Decision Framework for Mixed Results:

1. Quantify the Trade-offs Clearly:

• Positive Impact:
  • +15% CSAT (Statistically Significant - S.S.)
  • +10% Upsell Conversion Rate (S.S.)
• Negative Impact:
  • -8% Net Profit from Carousel Items (S.S.)
• Neutral Impact (Key Metric):
  • -1% Overall Total Order Profit (Not S.S. - this means we can't confidently say it decreased total profit; it could be noise around zero).

The fact that overall total order profit did not significantly decrease is very important. It suggests that the 8% profit drop from carousel items might be relatively small in the context of the entire order, or there were minor positive compensatory effects in the main basket that offset it (though we can't prove that with a non-significant result).

2. Estimate Potential Long-Term Value of Increased CSAT:

• A 15% increase in CSAT is substantial. We need to hypothesize its potential impact on:
  • Customer Retention / Reduced Churn: Satisfied customers are more likely to stay. Can we model the LTV impact if this CSAT improvement translates to even a small (e.g., 0.5-1%) reduction in monthly churn for the AP/Telangana cohort?
  • Increased Purchase Frequency or AOV over time.
  • Positive Word-of-Mouth & Referrals.
• Break-Even Analysis for LTV: How much does retention need to improve for the long-term LTV gain to offset the 8% immediate carousel profit dip (assuming that -1% overall profit is effectively zero impact for now)?
  • Let's say the average order has a profit of ₹P, and the carousel items on average contributed ₹C_profit_old with private labels. With branded, it contributes 0.92 * ₹C_profit_old. The dip is 0.08 * ₹C_profit_old per order.
  • We need to see if the increased LTV from better retention (due to higher CSAT) for customers exposed to Treatment B outweighs this small per-order dip from carousel items over their lifetime.

3. My Recommendation: Phased Rollout with Continued Monitoring & Optimization.

Given that overall order profit wasn't significantly harmed, and CSAT (a leading indicator of retention) improved significantly, I would recommend the following:

• Phase 1: Limited Rollout to a Segment (e.g., 10-20% of AP/Telangana users) of Treatment B (Telugu Brands).
  • Why: To gather more data over a longer period (e.g., 2-3 months) to specifically track cohort retention and LTV for users exposed to the branded carousel vs. control. This helps validate the hypothesis that higher CSAT leads to better LTV.
  • Monitor the -8% carousel profit and -1% overall profit closely. If overall profit starts showing a statistically significant negative trend, we may need to pause.
• Phase 2: Explore Hybrid Model (Treatment C) Optimization.
  • Simultaneously, iterate on Treatment C (Mixed Carousel). Can we find a mix of high-margin private labels and select high-converting popular brands that balances CSAT, conversion, and profit optimally? A/B test different mixes within this framework.
  • Perhaps use popular brands for "hero" categories where brand trust is paramount (e.g., Aashirvaad Atta) and private labels for less brand-sensitive impulse items.
• Phase 3: Develop Personalization (Treatment D).
  • Long-term, the goal should be to personalize the carousel. Use the learnings from Phase 1 & 2 to build rules or an ML model that shows the most relevant (and profitable while satisfying) items to each user based on their profile, cart content, and segment.

Justification for the CEO:

"The initial test indicates that featuring popular Telugu brands significantly boosts customer satisfaction and carousel engagement without a statistically significant negative impact on overall order profitability. While items from the carousel itself are less profitable, the strong positive customer sentiment is a valuable asset in a key strategic market like AP/Telangana.

We propose a cautious, phased rollout to a limited user segment to validate the expected long-term benefits on customer retention and LTV. Concurrently, we will optimize a hybrid model and invest in personalization capabilities. This approach allows us to capture the benefits of increased customer happiness while carefully managing profit implications and iteratively finding the optimal balance."

1. Operational & Supply Chain Complexity:
   • My analysis focused primarily on the digital/analytical aspects. Introducing more branded items, especially potentially fast-moving ones in a checkout carousel, has implications for inventory management, sourcing from multiple vendors, maintaining stock levels for these specific upsell SKUs, and potentially different negotiation terms with brands for items featured so prominently. This could affect overall operational costs not captured in simple item margins.
2. Competitive Landscape & Reactions:
   • I haven't deeply considered how competitors (e.g., other e-grocers, local supermarkets with delivery) might react if BigBasket makes a visible shift in its checkout strategy. If this strategy is successful, it could be copied, eroding any unique advantage. We also need to understand if competitors are already doing something similar effectively.
3. Long-Term Impact on Private Label Strategy:
   • There's a strategic tension. While the goal was profit per order, BigBasket also has a long-term interest in growing its higher-margin private labels. Over-emphasizing popular brands at a key touchpoint like checkout might inadvertently slow down the adoption or visibility of private labels. This needs to be balanced with the overall company strategy for private brands. Perhaps the mixed carousel (Treatment C) is the best long-term compromise.
4. Scalability & Maintainability of Personalization:
   • While personalization (Treatment D) is often the ideal, implementing and maintaining a robust, real-time personalization engine for the checkout carousel across diverse user segments and a vast product catalog is technically challenging and resource-intensive. The ROI of this complexity needs to be justified.
5. Definition & Measurement of "Customer Satisfaction":
   • CSAT from post-order surveys is one measure. However, a 15% increase is a relative term. We'd need to understand the baseline, the absolute scores, and correlate this CSAT with more tangible business outcomes like actual retention, repeat purchase frequency, and reduced complaints over time. The link between stated satisfaction and actual behavior isn't always 1:1.
6. Novelty Effect:
   • The initial positive results (higher conversion, CSAT) for branded items might be partly due to a novelty effect – users reacting positively to a change. We'd need to monitor if these effects sustain over a longer period beyond the initial 4-week test. This is partly addressed by the phased rollout and longer monitoring.
7. Cross-Device & Cross-Platform Experience:
   • My focus was on the mobile app. We'd need to consider if the strategy and its performance would be consistent on the web platform or other BigBasket interfaces. User behavior can differ.

Addressing these would involve close collaboration with operations, supply chain, technology, marketing, and finance teams to ensure a truly holistic and sustainable solution.