Skip to main content
Home
AboutExperienceProjectsCase StudiesBlogContact
Home
Saswata

Product Manager shipping outcomes that move revenue.

Pages

  • Home
  • About
  • Experience
  • Projects
  • Case Studies
  • Blog
  • Contact
  • Pay

Connect

  • LinkedIn
  • GitHub
  • Twitter

Resources

  • Resume
  • saswatasg@gmail.com

© 2026 Saswata S. Sengupta. All rights reserved.

    Back to all case studies
    LiveKeeping · Pro+ Feature · Send GreetingsJan–Mar 2026AI Integration·UX Redesign

    A dormant Pro+ feature. Rebuilt with AI. +168% engagement.

    Integrated Google's Nano Banana (Gemini Flash) image model so SMB owners could generate custom AI greeting cards for their customers — in seconds, from their phone.

    +168%

    Feature Engagement

    Pro+ Send Greetings module

    Nano Banana

    AI Model Integrated

    Google Gemini Flash Image

    27 occasions

    Festival Calendar Built

    Mar–Sep 2026, geo-segmented

    50,000+

    Platform Users

    LiveKeeping SMB base

    Role: Associate PM — End-to-end feature ownership|Team: Engineering, Design, Customer Success|AI Stack: Nano Banana (Google Gemini 3.1 Flash Image)

    BEFORE STATE

    LiveKeeping's Send Greetings module let Pro+ users send festival and occasion messages to their customers. The idea was right. Execution was static: pre-written templates, generic images, no personalisation. Indian SMB owners — shop owners, traders, manufacturers — couldn't customise the greeting to feel like it came from their business. Usage was low. Most Pro+ users had never touched it.

    GENERIC TEMPLATES

    Users were sending the same Diwali image as every other business on LiveKeeping. No brand identity. No personalisation. The greeting looked like spam.

    ZERO CUSTOMISATION

    Business name auto-filled, but the greeting image itself was fixed. Users couldn't change the colour, the style, the language register, or the visual tone.

    MISSED OCCASIONS

    The module covered major national festivals but ignored regional ones. A Ganesh Chaturthi greeting is irrelevant to a trader in West Bengal. A Vishwakarma Puja greeting is deeply relevant. One-size-fits-all = low resonance everywhere.

    WHAT NANO BANANA DOES

    Nano Banana is Google's AI image generation model, powered by Gemini 3.1 Flash Image. It generates photorealistic, text-accurate visuals from prompts — with precise text rendering across fonts, languages, and calligraphy styles. Specifically, it was designed for marketing mockups, greeting cards, and branded content with in-image text translation. That made it the right model for this use case: Indian SMB owners sending festival greetings in Hindi, Bengali, Tamil, Gujarati — not just English.

    Precise text rendering on imagesMulti-language / regional script supportGreeting card & marketing asset generation4K output · Lightning-fast (Flash model)

    SOLUTION

    PART 1

    AI Greeting Generator

    Users select an occasion, input their business name and customer name (pre-filled from ledger contact), choose a language, and Nano Banana generates a custom greeting card — unique to their business and their customer. 15 seconds. No design skills needed.

    PART 2

    Geo-Segmented Festival Calendar

    Built a 27-occasion calendar from March to September 2026 with geo-segmentation across Pan-India, South India, Maharashtra/Goa, East India, and Gujarat. A West Bengal trader sees Ratha Yatra and Vishwakarma Puja. A Maharashtra shopkeeper sees Ganesh Chaturthi and Gudi Padwa.

    PART 3

    Evergreen Greeting Library

    13 on-demand types available any day — Good Morning, Good Night, Thank You (post-payment + post-order variants), Motivation (Monday and general), Payment Reminder (3 urgency tiers: soft/firm/urgent), Sale Promotion, and 4 discount tiers (5%/10%/15%/20%).

    March is the busiest month (8 occasions: Ugadi, Gudi Padwa, Navratri, Eid, Navroz, Cheti Chand, Rama Navami, Hanuman Jayanti). March 19 and 20 each have 3 simultaneous occasions — requiring geo-segmentation logic to prevent a single user from receiving 3 identical-timing greetings. Independence Day (August 15) is the single highest-reach notification of the full calendar — mandatory Pan-India, all plans.

    WHAT I GOT SPECIFIC ABOUT

    1

    AI prompt construction — the Nano Banana prompt is built dynamically from: occasion name, business type (from user's LiveKeeping account category), customer name (from ledger), language preference, and region. A Kerala shop owner sending a Vishu greeting to a customer named Rajesh gets a different image than a Punjab trader sending the same Vaisakhi greeting. This level of specificity was not present in V1.

    2

    Template IDs — every greeting type has a unique template ID (e.g. GRT_EG_GOODMORNING_01, GRT_FEST_DIWALI_SOUTH_01) stored in our backend. This allows A/B testing between AI-generated and template-generated greetings at the same occasion — and gives customer success the ability to reference specific templates when investigating complaints.

    3

    The June 21 problem — Yoga Day and Father's Day fall on the same date. Two separate occasion push notifications on the same day would feel spammy. Solution: send only one, or schedule one for June 20 and one for June 21. Documented in the conflict resolution rules as a named exception.

    OUTCOME

    +168%

    Feature engagement uplift · Pro+ Send Greetings module

    27 occasions covered

    Mar–Sep 2026

    Geo-segmented

    Pan-India + 4 regional zones

    13 evergreen types

    Available any day, user-initiated

    15 seconds

    AI greeting generation time (target)

    RETROSPECTIVE

    1

    AI integration for a use case where AI genuinely adds value — personalised, localised visual content at scale — is fundamentally different from AI as a buzzword feature. The +168% engagement was driven by relevance improvement, not novelty. Relevance compounds. Novelty doesn't.

    2

    The evergreen library (Good Morning, Thank You, Motivation) was underbuilt in V1. Users wanted to stay in touch with customers between festivals. Building these as user-initiated on-demand greetings (not scheduled pushes) was the right call — they're a relationship tool, not a notification.

    3

    Measuring downstream business impact would have made this much stronger: do businesses whose owners send greetings have higher payment collection rates? Higher repeat orders? LiveKeeping's ledger data was right there to answer this. I'd build that measurement into V2 from day one.

    Compliance GapAll Case StudiesNotifications