.png){kind=logo}
- Insights
- Cryptocurrencies
- Stocks
- White Papers
- Industry
- Geography
- Insights
- Cryptocurrencies
- Stocks
- White Papers
- Industry
- Geography
Harnessing the power of technologies like artificial intelligence, machine learning and deep learning has enabled businesses to evolve with the ever-changing market conditions. Digitization helps eCommerce companies sophisticate their services while also maintaining the level of relevance. In the era of big data, eCommerce platforms have access to more information than ever before. Making sense of customer data to better tailor product development and marketing campaigns is the best way to streamline business perspectives in the modern ecosystem.
Sayurbox is a tech-supported fresh produce distribution platform based in Indonesia. Founded in 2016, its customers can order organic, hydroponic, and conventional fresh produce, which is sourced directly from farmers and producers as well as suppliers. The company offers deliveries seven days a week – with a focus on digitisation of the supply chain, waste elimination and helping farmers, producers, and their communities to achieve better access to their markets using fair trade principles to solve issues of unfair pricing and market access.
With 15+ years of rich experience in advancing Fortune 100 and Startup companies, Dr. Angshuman Ghosh emerges as a front-running Data Science, Engineering, and Business leader in the tech ecosystem. He has held leadership roles at top global companies including Disney, Sony, Target, Grab, and Wipro.
He is the winner of the 'Top 100 AI Leader Award' and 'LinkedIn Spotlight' award. Dr. Angshuman is a member of the prestigious Forbes Technology Council and AIM Leaders Council. He is a visiting professor at top institutes such as IITs, IIMs, and IISc. He has also published 12 research papers, 1 book, and 3 US patents.
Dr. Angshuman Ghosh is currently the VP and Head of Data at Sayurbox. As an eCommerce startup from South East Asia, Sayurbox is leading the digital transformation of the end-to-end supply chain from farmers to end consumers and has raised US$120 million in a Series C funding round recently. Dr. Angshu heads all the data functions of Sayurbox including Data Science, Data Engineering, and Business Intelligence.
Before joining Sayurbox, he worked as the Head of Data Science at Sony Research India where he helped set up Sony R&D Center India and its Data Science team from scratch. Dr. Angshuman led the Data Science team, collaborated with top researchers, and contributed to cutting-edge research and innovation in the areas of Data Science and AI-ML before moving to Sayurbox.
Both forecasting and recommendation are core technologies that are in high demand across industries. Forecasting technology helps predict the future. In the retail and eCommerce industries, predicting the future demand is highly important. If it is possible to predict the future demand well in advance, then anybody can plan all the supply chain initiatives in a structured and planned manner. Forecasting helps in purchasing, pricing, inventory planning, and even promotion planning activities.
Recommendation technology also has wide-scale implications. Most online media and technology platforms depend on recommendation technology to deliver personalized products and services to consumers. Some applications of recommendation in the online media context include Facebook Newsfeed, Netflix video recommendations, YouTube video feed, Instagram content explore page, Twitter News Feed, etc. Product recommendations on eCommerce platforms like Amazon are also an example of recommendation systems.
Forecasting plays an extremely important role for both offline retail and eCommerce companies. Forecasting models help predict future sales and thus help in supply planning, purchasing, pricing, and other activities. Dr. Angshu's team first used ARIMA-based models to create a baseline forecasting model and then utilized state-of-the-art models based on advanced Deep Learning techniques to improve forecasting accuracy by 30% over baseline.
Recommendation is a key technology in both eCommerce and media contexts. The goal of a recommendation system is to suggest personalized content or product to individual consumers based on their profile, behavior, and preferences. The team has created state-of-the-art recommendation systems using multi-modal inputs and neural networks.
Last year, Dr. Angshu and his team published 4 research papers in top international publications and 3 US patents based on research and innovation in Data Science and AI-ML areas.
Dr. Angshu's team used advanced AI-ML based methodologies to create best-in-class solutions for forecasting and recommendation. His teams at both Sony and Sayurbox developed novel technologies in Data Science and AI-ML areas. Last year, his team published 3 US patents and 4 international research papers based on its research and innovations. Dr. Angshu has also given talks at international conferences and institutions across the world.
Innovations by his team helped push the envelope on cutting-edge multi-modal recommendation and forecasting technology. The team also published its data and research in top international publications, making its data and research available to the wider research and innovation community. The team beat the state-of-the-art performance for the recommendation system, pushing the innovation frontier forward.
Sayurbox developed automated forecasting models using state-of-the-art AI-ML technologies. The company first developed ARIMA-based forecasting models. Then it implemented advanced Deep Learning models such as Prophet, CNN-QR, and Deep-AR-Plus. Using ARIMA and Prophet, Sayurbox was able to improve forecasting accuracy by 10% over baseline, and CNN-QR and Deep-AR-Plus to improve forecasting accuracy by 30% over baseline. With such advanced AI-ML based forecasting, the company saved lots of time and effort required for manual forecasting and improved forecasting accuracy significantly. The company also connected forecasting output to demand planning, supply planning, and purchase ordering system to optimize supply chain efficiency.
Both Sony and SayurBox developed several recommendation models for different use cases. It developed popularity-based and segment-based recommendation models to solve cold-start problems. Also, used content based on recommendation models to recommend content and products similar to the ones viewed or bought by users. They created collaborative filtering-based algorithms for item-to-item and user-to-user collaborative filtering. Finally, Sony developed multi-modal recommendation and graph neural network-based recommendation to beat the state-of-the-art performance of recommendation systems. AI-ML-based recommendation models provided automated personalized recommendations at scale for millions of users with best-in-class performance and saved huge time and effort.
Sayurbox developed the forecasting and recommendation models for the eCommerce use cases and Sony created an end-to-end recommendation system for the online video consumption use case. However, both forecasting and recommendation are common problems and may have other use cases in other industries. Models are optimized for industry and use cases for which they intended to use them, and hence same models may not give a similar performance for other industry use cases.
Sayurbox trained and implemented deep learning-based forecasting for a few thousand SKUs and it cost them a few thousand dollars. If the company needs to train and deploy the models for more SKUs, then the cost can go up to millions of dollars. Such high costs may be prohibitive for certain companies and use cases, and they may need to use simpler ARIMA-based models to save costs.
Sony's multi-modal recommendation system required huge efforts and costs in tagging thousands of videos with multi-modal information. Generating knowledge graphs and audio-visual embeddings also required huge efforts and costs. The company also protected its innovations with patents and anyone willing to use similar innovations may need to license the technology.
Both Sony and Sayurbox hired top talents in the fields of data science and machine learning for both full-time and intern roles to drive the innovation agenda. Sony also collaborated closely with top university professors and researchers in the related areas to foster research and innovation. The entire team dedicated significant time and efforts regularly, studying state-of-the-art research, conducting numerous experiments, and writing papers and patents. They also shared their research and innovations with the world using research papers, conference presentations and online articles.
