Project Constraints

• Diverse delivery regions with inconsistent infrastructure

• Limited historical data for training AI models

• Existing systems not built for predictive routing

• Tight delivery SLAs and high customer expectations

• Multiple stakeholders across logistics, tech, and operations

• 
  

Discovery & Research

Stakeholder Interviews I ran discovery sessions with dispatchers, drivers, and ops managers. Key insights:

• “We waste hours every week rerouting manually.”

• “I don’t know if my route is the best — I just follow what I did yesterday.”

• “We need something that thinks ahead, not just reacts.”

Industry Benchmarking I researched AI-driven logistics platforms and best practices:

• Dynamic route optimization using real-time traffic and delivery constraints

• Predictive modeling for delivery time estimation

• Reinforcement learning to improve routing over time

• 
  

Design Strategy

AI-Driven Routing Engine We built a smart routing engine with:

• Real-time traffic and weather data ingestion

• Delivery window constraints and priority scoring

• Historical route performance for predictive modeling

• Reinforcement learning to improve with every delivery

Product-Led Execution I treated RouteOps like a product:

• Defined MVP with core routing intelligence

• Created user stories for dispatchers, drivers, and ops

• Prioritized features based on impact and feasibility

• Ran weekly demos and feedback loops with stakeholders

• 

• 
  

Key Design Decisions

Decision 1: Build vs Buy We evaluated third-party routing APIs but chose to build our own engine:

• Full control over optimization logic

• Ability to train models on our own delivery data

• Seamless integration with SellSaffa’s logistics stack

Decision 2: Driver App Integration We embedded RouteOps into the driver app:

• Turn-by-turn navigation with optimized routes

• Real-time updates based on traffic and delays

• Feedback loop for route performance and delivery accuracy

Decision 3: Dispatch Dashboard We built a dashboard for dispatchers:

• Visual route planning with AI suggestions

• SLA tracking and delivery window alerts

• Route comparison and override capability

• 
  

Implementation & Validation

Technical Execution

• AI models trained on 6 months of delivery data

• Integrated Google Maps API for traffic and geolocation

• Built routing engine in Python with TensorFlow and FastAPI

• Deployed via containerized services on Azure

Validation

• A/B tested AI routes vs manual routes — AI won 87% of the time

• Driver satisfaction increased by 40%

• SLA adherence improved by 65%

• Fuel usage dropped by 22%

Results & Impact

• 70% reduction in routing inefficiencies

• 65% improvement in SLA adherence

• 22% drop in fuel consumption

• 40% increase in driver satisfaction

• 5 regions optimized with scalable AI routing

• 
  

Lessons Learned

AI Needs a Product Owner AI isn’t magic — it’s a product. Without clear goals, feedback loops, and ruthless prioritization, it’s just expensive math.

Discovery Is Everything The best insights came from drivers and dispatchers. They knew the pain. We listened, and we built what mattered.

Start Smart, Then Scale We didn’t try to boil the ocean. We launched in one region, proved the model, then scaled with confidence.