I love HN authors, publishing, and talking incessantly about AI, Tech,Startup,Blockchain & etc.
AI-driven commerce is the future. AI knows what products we want and why we like them. Now it’s creating these products for us.
It used to be the case that people invented things, hired mad men to create catchy ad campaigns around them, and sold them to an ill-informed public.
Although everyone knows this is no longer how things work, the true degree of change is more stunning than many people realize.
Today, we have the Commerce.AI “Cai” model. This new tech can “read consumer minds” by quickly ingesting millions of online customer feedback and designing the perfect set of product features to give buyers exactly what they want.
The Cai model has been trained over the last three years by experts from Stanford and MIT, who used over a billion feedback data points within 50k product categories in 9 languages.
The resulting model is groundbreaking: Cai can understand and distill the most important information within customer product feedback that are expressed in text, images and videos. It then creates new product concepts that fulfill the wants and needs of the customer.
“We need something that is fast and accurate,” said Lu Feng, Co-founder of Commerce.AI. “In this respect, Cai outperforms BERT, which is a powerful language model applicable to many tasks but requires a lot of computing resources to fine-tune.” Cai is trained specifically to understand commerce and can do so in less than an hour compared to the few hours that BERT requires at about the same accuracy.
One of Cai's tasks and strengths is to distinguish customer sentiment across vastly different product categories, and this is trickier than it sounds. For example, a microwave getting hot is a good thing, but a phone getting hot is not. Or even worse, some thin jackets are good but some thin jackets are not.
But Cai goes beyond sentiment analysis. It also learns what else matters to the consumers when they purchase a product, such as the reasons for purchasing (or returning) a product, the ways they might use a product, and so on. Cai is able to figure this out for any language with minimal to no human intervention.
Removing human-in-the-loop as much as possible is another one of Cai's strengths. Most neural network models rely on labeled training data, but obtaining this dataset is often very expensive and potentially exploitative. It is not scalable across different domains and languages. Cai receives some human guidance in the beginning for each language but is able to learn on its own.
Another challenge that Cai faces is the problem of interpreting different SKU (individual product) data because products are often talked about interchangeably in customer feedback. Adding to the difficulty is the fact that the same products are often listed in different places under different names within an online catalog. For example, a 32GB phone and a 64GB phone might be listed together, making it hard to determine which consumer feedback are referring to which phone.
Cai is capable of accurately interpreting the mind of the customer even within the vast ocean of online products and product categories.
The Cai model helps product brands know their customers and design better, more relevant products than ever before. As stated by Commerce AI CEO Andy Pandharikar,
“Using AI for market intelligence, you can uncover market opportunities to build the right product. With the number of unstructured product feedback data doubling each year and consumers increasing reliance on this data for their own purchasing decisions, brands have to use a technology-driven solution to understand what product they should be making next and what features it should include or exclude.
We are doing this for Coca Cola with soft drinks, Suzuki for cars, and even for Unilever’s product lines. Progressive brands like Midea and Cisco are also leveraging the Commerce AI platform to expand its existing product offering and identify new opportunities and customer segments.”
If you’re a large enterprise in the Consumer Packaged Goods or Consumer Electronics space with many products in popular SKU’s, there can be millions of feedback points and comments about your products, competitors’ products, and related products around the world.
The number of product feedback and conversations about those products is growing exponentially each day. Understanding the environment for launching new products is impossible with limited data and simple tools. Too many brands are still trying to understand the market by using traditional surveys, focus groups and spreadsheets that need to be reviewed manually.
Research from Neilson has uncovered that 85% of product launches fail. Maybe Cai is the solution.