Project Management

Context: 70% of small businesses struggle with tracking inventory effectively, resulting in lost revenue and unhappy customers.

My Role: Led 01POS development with 3 engineers and 2 sales personnel.

Approach:

• Ran 47 user interviews across Nigeria, Ghana, and South Africa to understand business owner pain points.
• Shipped MVP in 6 months using simplified UX.
• Ran controlled rollout: 5% → 25% → 100% over 24 weeks.

Impact:

• Reduced stock losses by 34% for 100+ businesses (₦780K monthly savings).
• Increased business performance analysis speed from 1hour → 5min.

What I Learned: Every business is unique and simple solutions beat sophisticated ones. Our initial onboarding design required us to create user accounts. However, this approach made conversions difficult. To simplify the process, we introduced a sign-up page, reducing the required information to just two fields.

AI Engineer

Problem: 01POS needed an intelligent forecasting engine to help merchants predict next-week and next-month sales, and automatically detect low-stock risks with suggested reorder quantity.

Technical Challenge:

• Feature wasn’t planned initially; had to restructure data and build new endpoints.
• Cost of new VPS for machine learning was becoming a concern.
• Integrate a Python ML pipeline into an existing Laravel + React ecosystem without disruption.

Architecture:

1. Data Layer: Pulls structured data from 01POS backend API:
   • 60–90 days sales history
   • Low stock data
   • Product image, name, current stock, price, category
   • Average daily sales (computed)
   • Stock limit (merchant-defined)
2. API Layer: Responsible for fetching datasets, packaging JSON payload, sending to ML model, and receiving forecast results (7-day & 30-day sales, stockout risk level).
3. Model Layer (XGBoost): Generates weekly/monthly sales forecasts and computes low-stock risk using historical sales, stock limits, and product metadata.
4. Prediction Engine: Cleans/transforms API data, generates model-ready features, runs inference, and outputs forecast numbers, risk scores, and recommendations.
5. Frontend Integration: React dashboard displays:
   • Next week/month sales forecast
   • Low-stock alert list
   • Product risk level indicators
   • Restocking recommendations

Model Performance:

• Strong short-term accuracy (7-day).
• Robust on noisy data.
• Slightly weaker for low-frequency sales.

Head of Product

The Bet: Aligning marketing, sales, and product into a single continuous feedback loop would accelerate growth, reduce guesswork, and help us build exactly what customers need.

Execution: Sales and development teams shared the same communication channels, speaking with one unified voice. Every complaint became a signal to investigate, and every sale became product feedback that guided what we built next.

Results: We moved faster, responded to market needs with precision, and shipped improvements matching real customer behavior. Conversion improved because messaging, product updates, and sales insights were always in sync.

Trade-offs Made: Centralizing the loop required more coordination, tighter communication, and the discipline to adjust quickly. We prioritized features that directly moved the product forward, saying no to anything outside the feedback cycle.

What I’d Do Differently:

• Validate willingness-to-pay earlier — We built for 6 months before discussing pricing. Should’ve run pricing experiments at month 2 with mockups. Cost us 3 months repositioning.
• Start with one vertical — We tried to serve all retail businesses. Should’ve focused on pharmacies or grocery stores first, then expanded. Generic features meant slower adoption.
• Ship mobile app in month 3, not 1 year later — 100% of our merchants use mobile phones. We built desktop-first because traditional POS terminal were PC. Mobile app took 12 months to retrofit. Should’ve built responsive mobile web first, or gone mobile-native from the start. Cost us estimated 40 potential customers who couldnt afford a new terminal.