Logo Lanfrica

mjananjohnson/makititoile

Domain:

socioeconomic
Creator:
mja
Host:
A prototype for understanding African consumer behaviour # Makiti Toile πŸ”¬ **Fashion-Data Lab for African Consumer Behaviour** An experimental ML project that combines behavioural analytics, attitudinal data, and AI-generated synthetic personas to predict fashion demand and understand African consumer preferences. ## πŸ”¬ About Makiti Toile **Makiti Toile** is our experimental fashion-data lab β€” a prototype for modeling African consumer behaviour through language, data, and synthetic simulation. Built on Makiti's real sales data, Toile combines **behavioural analytics** (what shoppers do), **attitudinal data** (what they say), and **AI-generated synthetic personas** (what they might do next) to predict which items will sell in our next restock. ### 🎯 Goal To merge behavioural and attitudinal insights into a single predictive model that helps us: - **Understand** why certain styles and price points perform - **Simulate** realistic customer reactions before launching new stock - **Reduce** over-ordering and waste by forecasting demand across categories ### 🧩 Project Phases | Phase | Focus | Output | |-------|-------|--------| | **1. Behavioural Foundation** | Analyse past Makiti sales to uncover spending patterns, category performance, and linguistic cues in product descriptions | βœ… Complete: Customer segments, product analysis, keyword extraction | | **2. Attitudinal Survey** | Collect short, anonymous responses on style identity, price comfort, and purchase motivation | πŸ”„ In Progress | | **3. Persona Grounding** | Merge behavioural & attitudinal data to build 3-5 realistic customer archetypes | πŸ“‹ Planned | | **4. Synthetic Simulation** | Use LLMs and semantic similarity rating to generate synthetic purchase intent scores for new products | πŸ“‹ Planned | | **5. Validation** | Compare predicted intent scores to actual post-restock sales to measure model reliability | πŸ“‹ Planned | ### 🧠 Core Concepts - **Behavioural Data** β†’ How customers actually shop: average order value, item count, category mix, descriptive language …