At last week's StoreAI launch, our CEO Andreas Hassellöf showed one way to use ChatGPT and similar Large Language Models (LLMs) in retail using Ombori Grid.
To start with, Andreas told the bot he was looking for a phone charger. The bot asked him what kind of phone he was using, and then recommended suitable product options, with additional information to help him make a decision.
Once Andreas chose which charger he wanted, the bot offered him a number of purchase options. He could have had it delivered. or else he could go to a nearby store and pick it up right away. To help him make this decision, he was shown a map to the nearest available store. Once he selected the pickup option, he was given a detailed store map showing him exactly where his purchase would be located.
To conclude the demo, Andreas asked about sustainability, and was told about the retailer’s sustainability initiatives and principles.
The important thing to note is that this is not possible using a standard ChatGPT implementation. It needs to be customized for use in retail. Let’s look at what we did.
Public vs private data
The secret behind this is that retail chatbots can’t just use generic, publicly available information. Most of the time, LLMs pull data from all over the Web. That doesn’t work in this context for two reasons. First, much of the information needed to answer a customer’s question isn’t actually available on the Web. And second, much of what’s on the Web is inaccurate, out of date, or misleading. Using generic data leads to so-called “hallucinations” or misinformation, which is not acceptable in a commercial context.
Instead, retail chatbots need to work from curated private data. They need access to data that’s not publicly available, such as current inventory or a customer’s purchase history. They also need to know that the data they’re using is up to date and accurate. To do this, the retailer needs complete control over the data the bot is using.
In the example above:
- The product information is sourced from the retailer’s product and inventory data held in Grid Products, which includes pricing and availability. This ensures Andreas won’t be shown products he can’t buy, and will get the current prices, as well as any discounts or promotions.
- Purchase options are generated by linking in with the retailer’s omnichannel and e-commerce setup in Grid.
- The store map is pulled from mapping data generated by Pointr and held in Grid. This ensures Andreas is directed to the precise location where he can pick up his charger.
- The bot referenced the retailer’s private knowledgebase on Grid and extracted the relevant data about sustainability. This ensures that Andreas is being given the very latest information.
This is all made possible by using a customized AI and private data.
Creating a private database
We’ve made it super easy to create a private AI database inside Grid. Simply drag and drop documents into StoreAI, and the bot will automatically parse them. In the demo I dropped a PDF about the retailer’s sustainability programs into StoreAI, and right away, the bot was able to answer Andreas’s question about their sustainability initiatives and principles.
The StoreAI knowledgebase ties in with existing private real-time Grid data, including products, pricing, discounts, inventory, delivery options, loyalty programs, locations, maps, and marketing materials. StoreAI has access to the entirety of this private information, and can construct its responses based on detailed, up to date information that is not publicly available.
Using a private database to power an LLM chatbot ensures that customers are getting accurate information. It completely removes many of the concerns with AI, and gives customers a reliable, enjoyable experience that’s fun and easy to use.