65–70% recommendation adoption
5–10% revenue uplift within 3 months
~20,000 outlet UAT scale
64+ outlet-level attributes
India to Malaysia scale-up
Business problem
Sales teams were relying heavily on manual SKU push lists such as MSBF, UF and Core SKUs to drive product penetration
across retail outlets. These lists helped standardise execution, but they did not fully account for outlet-level differences
such as purchase behaviour, category affinity, channel type, sales pattern, portfolio gap or growth potential.
- Frontline users needed more relevant SKU recommendations instead of static product lists.
- Retail outlets had different growth opportunities, but recommendations were not sufficiently personalised by outlet cohort.
- Business teams needed a scalable way to improve SKU penetration, cross-sell and basket expansion without increasing manual planning effort.
Product strategy
The strategy was to build a recommendation engine that could convert multiple outlet, SKU and commercial signals into actionable
product recommendations for frontline sales teams. The objective was not only to predict which SKUs could be sold, but to make the
recommendations usable in the field.
- Personalised SKU discovery: recommend the right products for the right outlet based on outlet behaviour and cohort.
- Cross-sell and basket expansion: move beyond core SKU lists into predicted cross-sell opportunities.
- Field adoption: keep recommendation outputs simple enough for frontline users to act on.
- Scalable rollout: create a model and operating playbook reusable across markets with optimised cost and performance.
AI and data logic
The product used outlet, SKU, purchase history, channel, sales behaviour and 64+ outlet-level attributes as input signals to understand
buying pattern, product affinity and growth opportunity. Outlet characterisation models were used to strengthen personalisation and cohort logic.
- RFM model: to understand outlet recency, frequency and monetary behaviour.
- Headroom portfolio model: to identify portfolio growth opportunity.
- Outlet profiling model: to group outlets based on attributes and business context.
- Ranking and thresholding: generated around 50–60 predicted SKUs, then surfaced only recommendations above the relevance threshold.
This moved the product from “many possible recommendations” to “fewer, better, actionable recommendations.”
Branch A — AI Recommendation System
Model-to-product translation across outlet signals, cohort logic, SKU affinity, ranking thresholds, recommendation outputs and frontline usability.
- Moved execution from static SKU push lists to AI-led next-best-SKU recommendations.
- Balanced model relevance with sales usability and adoption.
- Converted AI outputs into practical sales workflows.
Branch B — AI Product Scale-Up Playbook
Enterprise operating model across market readiness, data validation, model refresh, UAT, stakeholder accountability, adoption tracking and BAU handover.
- Created repeatable rollout logic from initial implementation to scale-up market adoption.
- Defined validation, issue-resolution, training and handover routines.
- Made the AI product scalable, not just technically deployed.
Product execution and stakeholder leadership
I led the recommendation product from concept to rollout, owning product strategy, roadmap, user stories, UAT, business case,
adoption planning and scale-up approach. The work required aligning business, data science, engineering, vendor, market and leadership stakeholders.
- Worked with business and sales teams to understand frontline execution pain points.
- Partnered with data science teams on model logic, ranking, cohorting and thresholding.
- Worked with engineering and vendor teams to convert model outputs into usable workflows.
- Aligned India and Malaysia market stakeholders on feasibility, rollout and operating model.
- Supported leadership and SteerCo conversations around ROI, prioritisation, budget and scale-up planning.
Rollout model and experimentation
India acted as the initial implementation environment where the recommendation engine was developed and validated. Malaysia was then taken forward
as a scale-up market, with a plug-and-play mindset while optimising rollout complexity, cost and performance.
- Covered market readiness assessment, data availability, signal validation, recommendation output testing and threshold calibration.
- Conducted frontline usability checks, UAT with business users, adoption tracking, ROI review and uplift analysis.
- Used A/B testing with comparable retailer groups to compare AI-led recommendations against existing sales execution logic.
Business impact
- Achieved 65–70% recommendation adoption.
- Delivered 5–10% revenue uplift within three months.
- Improved next-best-SKU relevance and cross-sell opportunities.
- Strengthened SKU penetration and basket expansion by outlet cohort.
- Created a scalable way to personalise sales execution in FMCG and B2B retail environments.
What I would scale next
LLM-led activation
Convert recommendation outputs into personalised retailer journeys, offer messages, campaign creatives and B2B app nudges.
Agentic recommendation workflows
Build agents that monitor outlet performance, detect gaps, recommend interventions and trigger next-best actions.
Real-time signal refresh
Move from periodic recommendation cycles toward dynamic updates using recent purchase, inventory and engagement signals.
Advanced experimentation
Expand from A/B testing to continuous experimentation across cohorts, channels, recommendation ranking and offer design.
Multi-market operating model
Create a reusable deployment playbook so each market can plug in local data, calibrate thresholds and scale faster with lower cost.
