Olga Romanova: Hello everyone. Good morning, good evening, good afternoon. Welcome to our online panel discussion, dedicated today to the high-profile topic of generative AI in retail, where machines try to mimic human interactions and generate creative content. I must admit, I've never seen software or a robot that could understand all the complexities of the weather in Munich, where I'm based.
Generative technologies are already showing success stories among retailers, with lessons learned and some technological constraints. Today, over the next hour, we hope to explore the most challenging questions we face when speaking to our customers, prospects, and different retailers regarding business and technological constraints, use cases for generative AI in retail, and lessons learned. Please post your questions and comments on LinkedIn.
We'll happily select the most interesting ones and try to answer them at the end of our panel discussion.
Let's begin! Please meet our speakers: Yury Gubin, Chief Innovation Officer at DataArt, based in New York.
Yury Gubin: Hello, everybody.
Olga Romanova: Denis Baranov, Senior Vice President and Head of Retail at DataArt, based in London.
Denis Baranov: Hello.
Olga Romanova: And Claudia Fuchs, Enterprise Sales Account Manager from Google, based in Munich. Together with me, Olga Romanova, Engagement Manager at DataArt, is also based in Munich. During this virtual event, we guarantee our speakers are authentic, not 3D avatars. If, during the conversation, we suddenly start speaking Chinese, please check your language settings.
Let's begin. Some colleagues even refuse to discuss generative AI because it's such a broad and vague topic until a specific use case and business value are discussed. You speak to customers across the globe. Over the last three or four months, or perhaps the first half of this year, what are the most popular or interesting cases, requests, or project ideas you've discussed?
Denis Baranov: Great question. Over the last year, everyone has talked about generative AI because it brings significant change. Machine learning has been around for a while, but generative AI has changed the approach and enables us to do things that previously required much effort.
When discussing real scenarios, I’d divide them into a few groups. First, there are optimization cases, like extracting data from PDFs and using generative AI to add features such as enhanced search and summarization. These cases have existed for a long time, but now we can improve them further.
Second, everyone talks about data. You can buy almost any data you want, but using it for analytics is key. Many clients want to see trends, forecast sales, and dynamically change prices. Prediction algorithms are in high demand, from forecasting popular colors to optimal pricing.
Third, there’s a lot of interest in content generation, such as 3D avatars, text, and videos. Previously, only humans could create these, but now machines can. Everyone wants to experiment with these capabilities but is looking for real business value, not just hype.
Olga Romanova: Thank you, Denis. Claudia, it would be interesting to hear Google’s perspective. Are there already success stories with local and retail companies?
Claudia Fuchs: Absolutely. I can echo what Denis said. There are big names that have succeeded with Google AI technology. For example, through smart analytics and AI, Camilla needed to gain more business value from first- and third-party data. By establishing their data platform on Google Cloud, Camilla developed advanced AI and analytics solutions tailored to their unique business requirements and strategic objectives.
Another example is Tchibo, a leading German retailer. They built a forecasting model with Vertex AI and BigQuery, generating millions of predictions daily to manage customer demand efficiently. Their data scientists and engineers found significant advantages in the flexibility offered by the cloud.
Olga Romanova: Thank you, Claudia. In 2023, everyone was experimenting with generative AI. Boston Consulting Group states that by 2025, generative AI will make up a third of the AI market. Denis, do you see this correlation in your experience with customers?
Denis Baranov: Yes, generative AI is at a hype level right now. Its simplicity attracts attention—anyone can use a chatbot and get immediate answers. Enterprises are starting to use it more because it offers many advantages. It’s the next generation of machine learning, and technology has evolved rapidly. What used to take months now takes days, especially with cloud solutions. Large language models have taken things to another level, providing capabilities that previously required years of education.
For me, one key case is using generative AI for copilots. It doesn’t replace human interaction but makes people more efficient, providing immediate answers and reusable patterns. Many other cases exist, but this is one of the most powerful.
Olga Romanova: Some generative AI applications in retail are textbook cases—chatbots, virtual consultants, and conversational commerce. These approaches boost revenue and provide personalized experiences. Claudia, how is Google shaping the future of shopping with AI-powered conversational tools?
Claudia Fuchs: Generative AI is changing the game, especially in computational e-commerce. With Vertex AI Agents, you can build sophisticated conversational AI agents using simple prompts—no coding required. These agents understand natural language, access product data, and guide customers through personalized shopping journeys. Dialogflow allows you to craft generative AI and rule-based chatbots for more control.
Our Contact Center AI platform modernizes customer service by combining virtual and human agents, boosting satisfaction, and analyzing calls for insights. Imagine a chatbot that remembers your past purchases, suggests items you’ll love, answers questions 24/7, and processes returns. Our investment in conversational AI is reshaping how we shop and interact.
Olga Romanova: Thank you. As we mentioned, there are many generative AI use cases. What was the most impressive generative AI implementation or project you’ve seen recently?
Claudia Fuchs: I really like solutions that enhance search. Shopping cart abandonment after failing to find the right item is common—three out of four customers will leave if they can’t find what they want. That’s why Google introduced Search for Retail. Imagine a search engine that understands exactly what you’re looking for, even if you don’t describe it perfectly. Retail Search helps customers find what they want quickly, leading to more sales.
Personalized shopping is another highlight—getting recommendations based on your style, budget, and past purchases. Visual search is also powerful: snap a photo of something you like, and Retail Search finds similar products. Google Lens uses similar technology.
Olga Romanova: Are there any metrics or success stories showing that these solutions work?
Claudia Fuchs: Yes, we have many success stories. Macy’s, Benetton, and Snuggie are great examples. Snuggie integrated semantic, contextual, and visual search on their website and increased revenue per search visit by over 3%.
Olga Romanova: Thank you. Yury, what’s the most impressive project or solution you’ve seen recently?
Yury Gubin: There have been many AI solutions in retail, but I really like the idea of precision retail—tailoring responses and recommendations based on my profile, purchase history, and preferences. One example is our prototype: a “buy me bot.” Instead of searching for a specific item, you ask what to purchase for an occasion or project. The AI breaks down your request, creates a plan, and suggests items you might not have considered. For example, if you want to build a greenhouse, it generates a blueprint and recommends tools and components. This conversion of ideas into shopping lists is new and impressive. KPIs include revenue, customer engagement, daily active users, and campaign performance.
Olga Romanova: Thank you. Denis, what are your impressions and perspectives on interesting solutions you’ve seen?
Denis Baranov: For me, prediction is always fascinating—how machines can anticipate what we want to buy or sell. Historically, the first conversations around machine learning were about prediction. Now, we can slice and dice audiences and predict trends, like what will sell in the next few months or the next flavor of ice cream. When we can predict what we want before we even realize it, that’s exciting.
Olga Romanova: I’d also like to mention a solution from DataArt’s team that impressed me: the virtual stylist. Coming from a family of three girls, fashion and dress-up have always been personal. The idea is to have a virtual stylist—a friend who helps you pick the right outfit for any occasion. You can use predefined prompts, type your own, or upload a reference photo. The AI generates recommendations based on your previous purchases and sizes. We even tried integrating a 3D visualizer, but generating precise images based on a few photos is still a technical challenge. What are the current technological constraints of generative AI today?
Yury Gubin: Technically, it’s possible to generate perfect images, videos, and holograms, and the models are evolving quickly. The main constraints are in integration, productization, data maturity, data quality, and cybersecurity. Protecting consumer data, generating reliable and ethical data, and avoiding biases are crucial. Intellectual property protection is another challenge, ensuring your generated content isn’t misused. Connecting structured product data with unstructured consumer requests is also complex. The technology is evolving, but building robust products on top of it is the real challenge.
Olga Romanova: It seems generative AI is still experimental, and only one in ten prototypes makes it to production. Are technological constraints the main reason, or are there other factors?
Yury Gubin: The challenges are mostly outside of technology. You need a strategy, roadmap, and vision. Proofs of concept can be created quickly, but the real question is whether they generate value, either new value or cost savings. Organizational maturity, data readiness, AI strategy, governance, compliance, and infrastructure are all important. It’s not just about implementing a chatbot; it’s about transforming your business. You need resources and partners to help you learn and manage risks.
Olga Romanova: Not all generative AI prototypes are supposed to be successful by design, right?
Denis Baranov: Exactly. Proofs of concept are experiments—they may or may not work. You need clear KPIs and red lines to know when to stop. Sometimes, the technology isn’t the issue; it could be data quality or interpretation. For example, we struggled to generate product descriptions from photos a few years ago. Now, generative AI solves this in weeks. Technology evolves, and sometimes you have to stop and try again later.
Olga Romanova: Prioritization and strategy are key. Yury, any comments?
Yury Gubin: Yes, we start with workshops and assessments to outline use cases across departments. We might identify 50 or 60 use cases, but the challenge is on the product development side. We measure and score each use case by cost, return, risk, and strategic fit. Resources are limited, so we shortlist and prototype quickly—fail fast. If something doesn’t work, move on. The goal is to keep experimenting and learning.
Olga Romanova: Speaking of risks, how do you create individual consumer profiles for generative AI solutions while protecting privacy?
Denis Baranov: It’s a great question. With these technologies, we must be careful about what user data we use. Regulations are evolving, and many governments are working on solutions with enterprise players. The golden rule is to use anonymized data.
Personalization is possible without exposing individual identities. Generative AI can provide direct consumer predictions, but it’s important to use categories and anonymized information.
Claudia Fuchs: We do not use customer data for training. Generative support on Vertex AI and other Google Cloud products provides the same security controls as BigQuery. Customer data is always encrypted, and access is traceable. No model is trained on customer data.
Olga Romanova: Thank you. Another question: regarding product discovery, how can generative AI guide users who don’t know exactly what they want?
Yury Gubin: Imagine an interface where you can ask for a business casual outfit for an interview. The AI may ask clarifying questions and, using structured product data and generative language models, recommend items that fit your needs, even if you didn’t specify them. This bridges structured data and unstructured requests, helping customers discover products through conversation. This applies to groceries, DIY projects, and outfits.
Olga Romanova: That’s the magic of conversational commerce—guiding users through dialogue. Claudia, AI is becoming more accessible, but when should companies bring in a partner?
Claudia Fuchs: Even with no-code platforms and user-friendly interfaces, partners are important for strategic guidance, customization, integration, and ongoing support. Partners help align AI goals with business strategy, tailor models to unique workflows, integrate AI seamlessly, and provide updates and training. To scale and optimize AI solutions, consider partnering with experts.
Olga Romanova: Thank you. As a Google partner, DataArt already has accelerators and prototypes to demonstrate these solutions. If you're interested, please post your questions or request a demo. Another question: Where is the memory of past purchases stored?
Denis Baranov: It depends on where you store your data—on-premises or in the cloud. Cloud solutions like Google Cloud make storing large amounts of data easy and affordable. We try to store as much as possible and reuse it as needed.
Yury Gubin: You can store every search request and data point about a customer. For generative AI, you must also vectorize this data—turn it into vectors for comparison and search. This bridges structured data and unstructured requests, all within one platform like GCP.
Olga Romanova: What are the main challenges for the retail sector, and do you have examples of generative AI use cases that solve them?
Claudia Fuchs: There are many use cases, including sales and marketing, content generation, internal knowledge bases, and more. Our team has prepared a deck of potential use cases for retailers and CPG companies, even tailored by role. If you’re interested, let us know in the comments.
Olga Romanova: A question about internal procedures: how can generative AI help with merchandising, operations, inventory management, promotion management, and pricing?
Denis Baranov: Custom pricing is a major topic. We can adjust prices dynamically to achieve different goals, like selling faster or hitting volume targets. Machine learning helps suggest optimal prices and predict trends. Merchandising is another area—predicting what to buy for the next year, when to move from shop to boutique, and how to run promotions. Machine learning and analytics help staff make data-driven decisions, finding trends that might be missed otherwise.
Olga Romanova: One last quick question for Claudia: Can you deploy custom, non-Google models on Google Cloud?
Claudia Fuchs: Yes, you can deploy custom models on Google Cloud. This is a unique feature of Google’s platform.
Olga Romanova: Thank you. We hope this panel discussion was useful and insightful. Thank you for spending this hour with us. If you have more questions or comments, please leave them or message any of our experts on LinkedIn. Enjoy the rest of your day.
Denis Baranov: Thank you so much for your time.
Yury Gubin: Thank you.
Olga Romanova: Have a good day. See you next time.
