How AI Orchestration Transformed Afrimart's Catalogue Management
How integrating OpenAI and Google Gemini into Afrimart's workflow reduced content turnaround time by 70% and enabled governance of over 95,000 product SKUs.
The Scale Problem
Managing 95,000+ product listings manually is not just inefficient — it's impossible to do well. Afrimart's catalogue had inconsistent descriptions, poor SEO taxonomy, and irregular pricing logic across categories.
The AI Orchestration Approach
Rather than treating AI as a single tool, we built a multi-model orchestration framework using OpenAI GPT models for content generation and Google Gemini for classification and quality validation.
Workflow Overview
- Input: Raw supplier data (often incomplete or inconsistent)
- Stage 1 — Enrichment: GPT generates product titles, descriptions, and feature bullets from sparse input
- Stage 2 — Classification: Gemini classifies products into the correct category taxonomy
- Stage 3 — Validation: Rule-based checks verify pricing logic, SEO keyword density, and completeness
- Output: Publish-ready product listings
Prompt Engineering at Scale
The quality of AI output is entirely dependent on prompt quality. We developed a library of structured prompts tailored per category — electronics prompts differ significantly from fashion or homeware prompts.
Results
- Content turnaround reduced by 70%
- Catalogue consistency score improved from 43% to 91%
- SEO organic traffic increased by 38% within 6 months
Key Takeaway
AI is not a magic button — it's a system you architect. The investment is in the framework, prompts, and validation logic, not the model itself.
