SF Maps aims to improve navigation for delivery partners, thereby assisting e-commerce platforms and Direct-to-Consumer (D2C) brands in reducing cancellations due to address inaccuracies on their platforms.
In contrast to standardised address formats prevalent in Western countries, Indian addresses often lack structure, are susceptible to language and understanding based gaps and are highly prone to input error posing challenges in pinpointing exact locations during delivery. SF Maps addresses these complexities by leveraging a AI/ML model trained on a vast dataset of Shadowfax’s past deliveries and pickups exceeding 1.5 billion data points. This model handles incomplete addresses, ambiguous area names, reliance on distant landmarks, and inaccurate pincodes, ensuring smoother operations. Further, precise navigation to customer addresses enables seamless deliveries without the need for additional calls, effectively reducing instances of missed deliveries and improving overall efficiency. Since the introduction of SF Maps, Shadowfax has achieved a reduction in customer cancellations or RTOs (Return to Origin) by almost 10% and boosted customer Net Promoter Score (NPS) by 25%.
Vaibhav Khandelwal, Chief Technology Officer at Shadowfax, said in a statement, “SF Maps represents a significant leap forward in our mission to optimise the delivery speed and elevate customer experience while solving fundamental problems in last-mile logistics. This innovative AI model trained on our vast set of historical delivery data drives significant operational efficiencies for us. We deeply understand the problems that arise due to incomplete addresses and how it hinders further innovation and hence we aim to make this AI model generally available for research in future.”
SF Maps uses an in-house Artificial Neural Network (ANN)-based embedding model, trained using a Siamese Network architecture. The generated embeddings are fed into VectorDB and the extracted locations are passed through H3 geospatial indexing, further fine-tuning location intelligence. This custom-built model captures complex contextual relationships between address components and their geographical associations, leveraging deep learning algorithms to discern intricate patterns for more accurate location-based intelligence.
The underlying algorithms and design architecture allows SF Maps to be a self-correcting engine which captures changing ground operations realities. The feedback loop is powered by real-time delivery partner geolocations that Shadowfax captures every 5 seconds while they are on their platform. Shadowfax also deploys a WhatsApp based conversational bot which interacts with customers and gathers address information for improved results and error correction.
