At Snaps, our goal is to make it easy for brands to deliver high-quality user experiences through AI chat automation. While chat has existed for many years on the web and in apps, the addition of AI enabled bots enables us to extend the value proposition by lowering the cost of servicing inbound customer requests while also offering product discovery. Naturally, one flows into the other as customers who need service then look for new products and vice versa. Combining both service and shopping functionality together in a high-quality end-user experience that is easy for brands to deliver is key to our mission at Snaps.
While the service functionality of ingesting FAQ’s and training intents to respond to queries such as ‘I want to return my order’ is more generally understood, NLP-based product search has until recently required a bit more understanding to set up, analyze, and improve. Because at Snaps we have a purview across a variety of brands, we have been able to generalize the product search setup, reducing implementation time from months to days, while eliminating the need for anyone on the brand side to have more than a basic understanding of things like AI and NLP.
In order to allow people to shop for products in conversation by saying things like ‘I want to buy green running shoes for women’ and ‘show me the latest handbags’ we initially hand-trained a Google NLP engine for a major international apparel brand. Using training phrases (examples of what people say) and entities (lists of products, colors, etc) the NLP is able to respond to a wide variety of product phrasings and identify all of the key terms from what someone is saying so:
‘I want to buy green running shoes for women’
Intent – Buy
Parameters – green, running, shoes, women
As you can imagine, even after learning the ins and outs of NLP product training, this can take weeks/months of setup and training, analyzing user input, making sure we cover all the variations of what people want and how they say it. Because this functionality is useful to almost all eCommerce retailers we wanted to find a way to quickly deploy new AI agents for others brands in a way that 1) lets us leverage everything we’ve developed previously 2) improves the experience for all users by incorporating learnings from each deployment.
After we had success with training some product search intents for a few different brands, we were ready to replicate our success for new brands without starting from scratch. We realized that, for each brand, the entities and utterances are very similar, it’s mostly the specific values that change. Therefore, we took to translating the sample utterances from one brand to another through the use of some scripts in the following manner:
As you can see on the left we have samples from an Activewear brand that are output on the right as being ready for a Luxury Brand. The difference is the specific entity values that are applied to the samples which is provided by our script that reads all of the values for Product-Facet and iterates through all samples, replacing the entity values but leaving the samples unchanged. This is essentially our method for starting a new brand with a fully trained product search.
The additional value of being able to monitor and analyze the product searches across multiple brands is what allows us to continually improve the accuracy and functionality of searches. As we find opportunities in one training, we can deploy those updates to all other relevant training sets. In this way, each brand is able to maintain its own training set while leveraging insights from Snaps’ entire customer base on an ongoing basis.
Want to learn more about what Snaps can do for your business? Contact us online to learn more about driving commerce and providing great customer care with conversation.