Executive Perspective

Improving Cart Sizes with Conversational AI Assistants: A Conversation with Blutag's CEO Shilp Agarwal

Grocery Doppio spoke with Blutag CEO Shilp Agarwal about the company's revolutionary generative AI conversational intelligence platform and its ability to optimize cart sizes for grocers while harnessing the potential of impulse purchases.
Neha Ghai
July 14, 2023

Shilp Agarwal


Revolutionizing Online Grocery Shopping with AI

In the evolving landscape of online grocery, consumers often need more support due to the overwhelming number of options available. One solution is a recommendation engine that offers valuable insights based on real-time conversations, leveraging the advancements of the post-ChatGPT digital revolution. 

To compete effectively with the next generation of shoppers who have access to vast amounts of information but may need help with information overload and variety, grocers require a verticalized conversational AI assistant. Grocers can retain future shoppers and maximize customer satisfaction by harnessing the power of large language models and integrating real commerce datasets specific to the grocery domain. This approach enables a personalized and streamlined shopping experience that meets each customer's unique needs.

Blutag's Impact on Grocery Retail

Blutag, an intelligent conversational commerce platform, leverages the capabilities of large language models and domain-specific conversational commerce datasets to provide real-time insights that effectively increase the average cart size for grocery retailers. By enhancing and maximizing cart sizes, e-commerce grocers can gain a strategic advantage in customer retention and profitability. Blutag's solution is designed to capture additional impulse purchases, thereby driving up the average order value and delivering tangible results for grocers.

In an exclusive Grocery Doppio interview, Blutag CEO Shilp Agarwal shares insights into the company's vision and objectives. The discussion highlights how their innovative approach is transforming the retail grocery industry by providing in-depth insights through the interpretation of customer data. By leveraging this approach, grocers have experienced significant increases in their cart sizes, with improvements ranging from 12% to 18%.

Grocery Doppio: Tell us about yourself and what inspired you to start the company.

Agarwal: My career has primarily revolved around digital retail. Before starting Blutag, I founded an online jewelry business during the early days of e-commerce. We didn't have any venture backing for that venture. Instead, I approached a prominent diamond manufacturer in New York's diamond district and took a leap of faith. This gave us access to a vast inventory of diamonds and other materials. However, we soon realized customer acquisition costs were relatively high, especially in luxury retail. As a result, our focus shifted to increasing customer retention. We wanted to determine how to maximize revenue from each customer after acquiring them. Our main strategies involved capturing impulse purchases and increasing the average order value. When voice assistants like Alexa and Google Assistant started gaining popularity, I found them to be an intriguing opportunity. These devices were becoming ubiquitous, connected to the Internet, and an excellent channel to capture user impulse and enhance customer retention. This led to the inception of Blutag, with the idea of leveraging voice assistants as a platform for our vision.

Grocery Doppio: Can you explain the architecture of your generative AI system?

Agarwal: Our generative AI system is a multi-model setup that combines multiple LLMs (large language models) with a recommendation engine. We view our language model as a reinforcement learning-trained model, which distinguishes it from general LLMs like ChatGPT, which are primarily trained using supervised learning. The distinction lies in our goal-oriented approach. Our language model is specifically trained to direct users toward purchasing products. This emphasis on achieving a specific objective differentiates it from more generalized language models. By integrating this goal-based model with a recommendation engine, we create a powerful combination ideally suited to our mission of building and delivering a comprehensive solution for our customers.

Grocery Doppio:  Can you provide an overview of the different user experience approaches available for grocers to enhance their website?

Agarwal: In a couple of ways, e-commerce-enabled grocers can benefit from large language models. The first approach is integrating a language model directly onto their website to assist users in navigating the site more efficiently. This can be done through text or voice buttons, allowing visitors to ask questions and receive guidance. The AI assistant helps users navigate the website more efficiently, provides personalized recommendations, and facilitates quick additions to the shopping cart, improving the website's overall experience.

The second approach involves using plugins designed explicitly for ChatGPT. In this scenario, users initiate their journey within ChatGPT itself. For example, someone might seek dietary recommendations for managing hypertension. They can start their query within ChatGPT, asking about suitable foods and meal options. ChatGPT can then provide guidance and relevant suggestions. Plugins come into play by seamlessly guiding users from ChatGPT to the e-commerce website, offering recommended products, and simplifying the checkout process. This streamlines the path from inquiry to the shopping cart.

The first approach focuses on providing an AI assistant on the website to facilitate browsing and quick product addition. The second approach capitalizes on capturing users already engaged in ChatGPT and smoothly directing them to the e-commerce site. These innovative plugins cater to the evolving expectations of users who seek more intelligent, efficient interactions. People expect technology to be quick and competent, leveraging the vast data available. The aim is to eliminate the need for users to start their journey on one platform, search for information elsewhere, and then navigate to an e-commerce website separately. By connecting the dots, plugins enable a smoother and more integrated user experience.

Grocery Doppio: How do you ensure the accuracy and reliability of your system's recommendations?

Agarwal: Our model operates on a reinforcement learning framework, which can be likened to a model that receives rewards for guiding users towards a desired outcome, known as the happy path. In our case, the happy path involves users adding items to their shopping cart. The system receives a reward whenever a user successfully adds an item to the cart. This reward mechanism forms the basis of our reinforcement learning approach, and constant reinforcement helps improve the model's accuracy.

As more users interact with the system and add items to their carts, the model collects more data. With this growing dataset, the system's accuracy in predicting and suggesting things that align with users' preferences keeps improving. It becomes a self-learning system, continuously refining its recommendations based on user interactions.

Grocery Doppio: What are the specific pain points in the grocery industry that these plugins aim to address?

Agarwal: The grocery industry has recognized that many customers now begin shopping online, seeking information and guidance. This shift is evident even in simple Google searches, which can translate to interactions with ChatGPT, where users ask for help planning meals or finding specific products. BluTag's model can respond by suggesting gluten-free meal options, providing menu ideas, and even adding all the necessary items to the user's cart. 

Before the advent of the ChatGPT revolution, AI-powered Chat solutions were considered a value-driven benefit for grocers, enhancing the overall customer experience. To illustrate this shift, we can compare it to the concept of vitamins versus painkillers. Previously, AI assistance was seen as a "vitamin," something extra that provided value but wasn't perceived as essential.

However, user expectations are rapidly changing, transforming AI solutions into a "painkiller" for grocers. Customers now consider this level of AI-powered assistance to be a necessity rather than a luxury. They anticipate personalized recommendations, streamlined experiences, and seamless interactions as a standard offering from grocers.

Amazon has been significantly doubling the grocery business every two years; this exponential growth rate poses a challenge for grocery retailers who want to remain competitive and retain customer loyalty.

A key strategy to meet customers' changing needs in this dynamic and evolving landscape is AI-driven solutions that meet customers in their preferred digital space and enhance the shopping experience. 

Our solution addresses these pain points by providing an AI-powered platform that assists users in their online grocery journey, from meal planning to adding items to the cart. Grocers can effectively compete and retain their customer base by meeting customer expectations in the digital realm.

Grocery Doppio: What are the challenges and integration capabilities of your technology in the grocery sector?

Agarwal: The biggest challenge we encounter is the disconnect between the product teams, who are enthusiastic about implementing our solution, and the overwhelmed tech teams. While the product teams understand the value and importance of AI-powered assistance, the tech teams are often burdened with backlogs due to the rapid changes in the grocery landscape, mainly driven by the pandemic. Many grocers are currently focused on establishing the basics of their online presence. Grocers are currently focusing on establishing the basics of their digital operations, which creates a delay in adopting new technologies like ours.

Previously, convincing product teams to embrace AI solutions posed a challenge, but now the focus has shifted to align with the tech teams' capacity and priorities. While the product teams recognize the value of our solution, integrating it into existing technology becomes a crucial consideration.

To address this, we have developed various integration approaches. For instance, if a grocer already has a recommendation engine that caters to individual preferences, we can tap into that system. This ensures a seamless experience for users, whether they search for products online or use AI systems for recommendations. We aim to avoid creating any surprises or disruptions for users already familiar with the grocers' platforms, ensuring a smooth integration of our solution.

Grocery Doppio: How can your company boost online grocers' digital profitability?

Agarwal: As the grocery industry shifts online and competition intensifies, one of the significant challenges for grocers is to increase their basket size. This is especially crucial when considering additional costs associated with curbside pickup or home delivery. Having a larger basket size helps offset these costs and make these services economically viable. Our focus has been twofold: capturing impulse purchases and increasing the average order value. We have been successful in achieving both objectives effectively.

For example, one of our partner grocers shared that their customers spend an average of 13 minutes shopping for groceries online when placing an order. Whether they add five items or 50 items, they invest the same amount of time. This highlights the importance of enabling customers to add items to their carts quickly. Our AI system significantly accelerates this process. For instance, if someone wants to shop for a Mexican dinner for six people, including vegetarian options, our system can help them build their shopping cart approximately 20 to 25 times faster than traditional methods of browsing and filtering.

By reducing the time spent on basic shopping tasks, we provide customers with an extra five minutes for impulse purchases. This is significant because impulse buys typically constitute around 65% of a traditional grocery shopping cart. By allowing customers to indulge in impulse purchases online, we contribute to replicating the in-store experience. Our solution empowers grocers to capture these additional purchases and consistently helps them enhance their online shopping experience. It's a vital aspect that grocers must consider in this competitive landscape.

Grocery Doppio: What is your target market? How do you ensure your products meet the needs of the end user? 

Agarwal: Our target market includes any grocer that has an e-commerce presence. This is because any grocer with an online platform faces similar challenges. 

Customers engage in two main types of online shopping: buying online and picking up at the store or buying online and getting items delivered. In both cases, it is crucial to have additional items in the cart to offset the costs associated with delivery and pickup.

Our product is designed to meet these needs, providing grocers with the means to offset delivery and pickup costs by encouraging customers to add extra items to their carts. We provide analytics and data to ensure that grocers can track and measure the impact of our solution. When grocers see the potential ROI and how our solution effectively addresses the pain point of increasing basket size, it becomes a no-brainer choice for them.

Grocery Doppio: What unique proposition sets your company apart in the competitive industry?

Agarwal: Our most significant competitive advantage in this ever-evolving digital landscape is our foresight regarding the future of LLMs. Initially, our product focused on integrating with voice assistants like Alexa and Google Assistant, enabling users to perform basic shopping tasks. However, we quickly recognized the limitations of these voice assistants when it came to engaging in natural and intelligent conversations. 

Around three years ago, we came across GPT-2, an open-source language model, and were fascinated by its potential. We conducted experiments by training the model on various product datasets, including the beauty industry. Even with minimal training, we were astounded by the model's ability to provide detailed answers to product-related queries, offer opinions, and even provide pricing insights. This realization cemented our belief that building an effective conversational AI system necessitated harnessing the power of a robust LLM. Since then, we have embarked on our journey to develop real conversational commerce datasets, a process that began nearly three years ago. 

Our early adoption of large language models and the extensive training we have conducted have positioned us ahead of the competition in the fiercely competitive landscape of ChatGPT solutions. By leveraging real-world conversational commerce data, our system has already demonstrated its ability to impact numerous shopping carts significantly. Today, we are proud to possess one of the most extensive datasets outside of Amazon, strengthening our position in the market and reinforcing our commitment to delivering high-quality conversational AI experiences.

Grocery Doppio: Can you detail how BluTagBlutag's AI solutions have helped grocers and the metrics showing their effectiveness? 

Agarwal: The primary focus of our solution is to increase the basket size for grocers. In the grocery space, we have achieved substantial results, with average increases ranging from 12% to 18% in basket sizes. This is a critical metric that we track and quantify for our clients. To ensure transparency and accountability, we provide grocers with a dashboard and analytics that showcase accurate data. This allows them to see the impact on their cart sizes and hold us accountable for the promised results. By setting specific key performance indicators (KPIs) related to the cart size increase, we work closely with grocers to achieve these goals.

Even when we initially integrated with voice assistants like Alexa and Google Assistant, we could boost grocers' cart sizes by around 11% to 12%. This was achieved by leveraging the impulse buying behavior captured through these voice assistants. By combining this capability with our new generative AI solution, we have witnessed a significant increase in basket sizes.

While we address other metrics and areas of impact, such as customer engagement and time spent with the brand, our primary focus remains on increasing cart sizes. By concentrating our efforts on this specific metric, we ensure that grocers experience tangible and quantifiable results from our solution.

Grocery Doppio: How do you ensure your AI system respects customers' privacy while providing personalized recommendations?

Agarwal: One of the primary reasons brands, e-commerce companies, and grocers should consider leveraging our platform is because of the concerns surrounding data privacy and security regarding large language models. Data privacy is a critical aspect of our approach. We recognize its importance in a rapidly evolving landscape where data security concerns are paramount. 

When businesses contemplate using these models, questions arise about the collection and handling of data, including personally identifiable information (PII) and customer information. Our platform acts as a firewall between the generalized models and the e-commerce platforms, ensuring that a grocer's data remains their own. We prioritize data privacy and work closely with organizations like the Open Voice Network to address these concerns. Data privacy must be at the forefront of this technology because many of the benefits become irrelevant without it.

By implementing our platform, grocers can have peace of mind knowing that their data remains their own. If a grocer decides to disconnect from our platform or a specific large language model, they can rest assured that all their data will be transferred with them, leaving nothing behind. We have carefully considered data privacy concerns and are committed to providing our clients a secure and trusted environment.

Grocery Doppio: Please provide insight into any upcoming products or services your company plans to release.

Agarwal: Our LLM has been in the market for some time, proving its effectiveness in increasing engagement and cart sizes. Now, we are taking it a step further by integrating our LLM with generalized language models like ChatGPT. The future lies in the collaboration between these two types of models. 

We are excited to announce the upcoming release of SDKs for web and mobile platforms to facilitate this integration. These SDKs will enable seamless integration of our LLM and ChatGPT, providing grocers with a unified model that combines the strengths of both verticalized and generalized models. With our SDKs, grocers will no longer need to integrate their own LLM separately or navigate between different models. Instead, they can access a comprehensive solution that leverages the capabilities of our verticalized model while harnessing the power of generalized models. This integration allows for a more holistic approach, where the vertical models enhance and complement the functionality of the generalized models. We anticipate the release of these SDKs later this year, and we are thrilled about the possibilities they will unlock for our customers and partners.

Grocery Doppio: How do you see the market for AI-driven conversational assistants in the grocery industry in the coming years?

Agarwal:  In today's fast-paced AI world, developments can occur rapidly. Looking ahead to the next five to 10 months, the advancements in generative AI have been monumental. We believe that, eventually, everyone will have their personalized AI-driven agent. This agent will be capable of performing various tasks, such as booking tickets, ordering food, and even groceries. The key focus will be enabling a connection between generalized AI and vertical AI models. Building vertical models is becoming increasingly relevant because it provides a platform for AI personal assistants to interact with different specialized assistants. This integration will allow for a seamless flow of information and capabilities. 

In the context of grocery stores, we are already exploring capabilities that can be realized shortly rather than a distant five-year timeline. For example, our AI agent can currently assist users in building shopping carts based on their needs. However, with the advent of intelligent refrigerators equipped with cameras, we envision a scenario where our system can access the refrigerator's contents and avoid duplicating orders for ingredients already available, streamlining the shopper experience.

These advancements will lead to the development of personalized digital agents with a comprehensive understanding of individuals' preferences, knowledge about different stores and vertical models, and even awareness of their refrigerator inventory. The goal is to empower users by automating various tasks and streamlining their daily routines. Given the rapid progress, we will likely achieve this vision sooner than anticipated. The rapid pace of advancement is genuinely remarkable and exciting.


Grocery Doppio: Tell us about your creative side when you are not ideating for Blutag.

Agarwal: In the past, I had a strong passion for playing musical instruments, although I haven't had the opportunity to pursue it recently. However, one activity that still brings me joy, particularly during weekends, is photography. I find great pleasure in capturing moments through the lens of my camera. One area of photography I particularly enjoy is capturing action shots, as I have children who participate in sports. Currently, photography is one of the main hobbies I actively engage in.

Grocery Doppio: What advice helped you persevere through tough times and could inspire new entrepreneurs?

Agarwal: Being an entrepreneur is an incredibly challenging journey, and it's essential to acknowledge its difficulties. One valuable advice I received from an advisor during a particularly tough period was to take a few days off from work and reflect on the purpose and direction of my venture. This advice had a significant impact on me. During those days away, I was encouraged to think deeply about where I envisioned my business in the next three years. It wasn't just a matter of thinking about it; I was urged to write it down. Putting my thoughts into writing made the vision more tangible and forced me to commit to it on a different level. 

By taking this step, I was able to assess whether my passion was merely a personal interest or if it had the potential to address a genuine need in the market. It prompted me to reevaluate my approach to the business and assess if it had a viable future. Upon returning from my break, armed with a renewed perspective and a clearer sense of purpose, I approached my work differently. I focused on building things that made sense and aligned with my vision.

I highly recommend this practice to fellow entrepreneurs in a similar phase of doubt and uncertainty. Taking time off to reflect and write down your aspirations can provide valuable insights and help reignite your passion and belief in your venture. It certainly made a difference for me, and I hope it can do the same for others.