How much of your current AI strategy is delivering real, measurable profit, and how much is just…productivity?
For leaders facing margin pressure and operational complexity, the current applications–summarising meetings or drafting emails–are merely the starting point. The real value lies in using AI to solve the industry’s most expensive and complex problems: high return rates, inefficient cross-border scaling, and poor product discovery.
We sat down with Sebastian Spiegler, Senior Director of AI at Rithum, to cut through the hype. With a PhD in Machine Learning and Natural Language Processing and a background leading AI product strategy over the last decade, Sebastian operates at the practical frontier of applied AI.
In the newest episode of Business Casual, he shares how high-growth brands are moving beyond the buzz and why your data, not the AI model, is the single biggest factor for success.
The three tiers of AI: Moving from efficiency to advantage
Most brands are already using AI, but Sebastian frames this adoption as three distinct tiers. While the first two are about efficiency, the third is where competitive advantage is built.
Tier one: Workflow and productivity. This is the AI embedded in your daily tools - like CoPilot or Gemini - helping you draft documents and summarise data. It saves time, but it offers zero differentiation.
Tier two: Third-party services. Here, brands experiment by plugging into subscription-based services like Amazon Bedrock as a low-investment way to test specific use cases.
Tier three: Proprietary AI. This is where the real value lies. "Getting the real competitive advantage, you need to use your own data and build your own models," Spiegler explains. This is about building systems trained on your business data to solve your specific problems.
The push to the third tier is almost always driven by operational complexity. A generic AI can’t solve the “immense challenge of onboarding and categorising products at scale across different languages and market taxonomies”. This is a core bottleneck in European expansion. It’s why Rithum developed its own proprietary AI Magic Mapper, powered by RithumIQ. Magic Mapper uses multi-language models to automate product categorisation and mapping, helping brands “launch in the new channel within minutes rather than days”.
Solving the returns problem with AI
For fashion brands, the clearest path from AI investment to profitability is by tackling returns.
This isn’t just about spotting broad trends. It’s about granular, SKU-level insights. As Spiegler notes, a huge portion of returns is often driven by a very small number of products. The problem is finding them in a catalogue of thousands–which is where AI can help.
A proprietary AI, trained on your own returns data, can provide actionable recommendations: “AI doesn’t just tell you that bikinis return a lot in Germany, but tells you to ship them to the UK where they sell at full margin”.
This is the difference between data (a high-level stat) and insights (an actionable recommendation). An AI can also pinpoint why an item is returned–a misleading image, a poor fit description–and even benchmark your product performance against industry peers by category and country.
We’ve run analysis for some of the big fashion brands where something like 500 SKUs drive 10% of the returns. If you attack those SKUs, you can greatly reduce those returns and increase your profit margins.
Improving discoverability with AI
The way customers are finding products is changing. It’s no longer just a search bar. Spiegler points to the rise of agentic commerce, where customers use AI agents to find products in natural language: “find me the best dry shampoo under €20”.
The new frontier presents a massive challenge. To be discoverable by an AI agent, your brand must have:
A crawlable D2C site.
High-quality, feed-ready product data.
A strategy for new standards, like the emerging llms.txt file that governs how AI models can use your site's content.
"Consumers search more and more for products within LLMs," Sebastian notes. "Brands [must] adapt... making their websites more discoverable". This is why Rithum is already actively integrating its brands' data feeds into new agentic search engines like Perplexity, ensuring they show up where customers are searching next.
Your AI is only as good as your data
The shift to agentic commerce highlights the single biggest hurdle holding most brands back: data readiness.
You can have the most advanced AI model, but it’s useless if it's running on incomplete, inconsistent, or siloed data. Spiegler was clear that Rithum itself had to heavily invest in building its own data lake to create a single source of truth before its AI strategy could succeed.
Preparedness starts with data. Get your data right and consumers will find your products. And this is true for marketplaces today. This is true for the agentic search engines tomorrow.
What's next? Agent-to-system automation
Looking ahead, Spiegler predicts the next major disruption will be "agent-to-system automation".
This is when AI agents move from just finding products to acting on them—connecting directly into retail systems. "They may check your stock... they may order, or they may deal with your returns," Spiegler predicts. This level of automation will "step by step automate those [tasks] for the e-commerce team".
The takeaway is clear: this future requires a clean, connected, and integrated data foundation. AI isn't magic. It's a powerful tool that, when built on a solid data strategy, can move from a simple productivity hack to a core driver of profitability.
