Product Search and Recommendation

Most people use the online E-commerce search engines for exploration and product discovery. The relevance of products displayed by product search engines varies considerably across different platforms.

A customer interacts with a product search engine by typing a string in the search box often known as the query. Most product search engines rely on matching user-typed keywords and product titles. However, it is well known that simple keyword-based filtering is unlikely to find semantically related products. For illustration, consider a user-supplied query “backyard sitting.” A simple keyword-matching system will ignore products such as “outdoor chair” or “outdoor bench” as they are not keyword matches. Even after missing these relevant products, the search engine will likely find hundreds of products for user-supplied keywords. However, showing all those results can frustrate the customer by showing a lot of irrelevant items.

Fortunately, with recent advances in AI, this problem is seeing a solution.  AI mines patterns in the user behavior or past behavior of similar customers and identifies relevant products enhancing user experience and making the product discovery process easy and enjoyable.  Research shows AI can produce relevant search results and increase conversion rates by 915% and average order values by 3%.

ThirdAI’s BOLT unlocks the power of AI and NLP (Natural Language Processing) in any product search engine without any effort.  Our push-button solution integrates with existing search engines without any engineer or data scientist’s effort. BOLT consumes historical data of past customer interactions and automatically builds and deploys the AI model with the highest seen accuracy and at 100x less cost for AI.

BOLT gives you a complete handle on training, retraining, and deployment in any given environment or infrastructure of your choice. Our scalable AI can be easily extended to provide personalization. In addition,  BOLT can leverage and mine known relationships between entities in the form of a graph.  Furthermore, BOLT provides a continually learning solution, which means that the model evolves automatically with more usage.

In addition to autotuning all the hyperparameters, BOLT also autotunes for the available resources and latency budget.  Therefore there is no need to spend on expensive data scientists and ML engineers. Overall, BOLT takes care of the complete AI cycle on any infrastructure.

1. Improved customer retention and sales.

2. A higher number of returning customers.

3. Reduced cart abandonment.

4. Accurate AI/NLP-based semantic search that works on any infrastructure.

5. Reduced operating cost

• • –No need for expensive hardware for AI.
  • –No overhead for data scientists and ML engineers.BOLT takes care of the complete lifecycle of AI (training, retraining, deployment)

6. BOLT gives you the complete ownership of your model. You have full control of the AI model from building to deployment on  any infrastructure of your choice.

Almost all AI and NLP systems out there are either complicated and require expensive hardwares and data scientists. Even after all the headache, they are likely going to provide a response time of over 200ms.  A study done by Amazon found that every 100ms of latency Cost them 1% in Sales. Clearly 300ms is prohibitive.

Our evaluations of an e-commerce giant show that BOLT results in a 4% uptick in customer engagement, with the displayed results leading to proportional growth in sales.  In a rigorous evaluation BOLT provided less than 1ms latency where the competition was more than 200ms.

ThirdAI is a software-only solution to accelerate AI on any CPU. Our scientifically proven hash-based “sparse” solutions can reduce the computing, energy, and carbon footprint of deep learning engines by 100x. As a result, we can provide GPU level or better performance on simple commodity computers. More information can be found at www.thirdai.com/bolt