This article is a part of Poland Unpacked. Weekly intelligence for decision-makers
One of the biggest shifts in Polish e-commerce in years may be underway. InPost has launched an AI assistant that enables users to shop directly within the InPost Mobile app. The tool already gives access to around 4,000 merchants – and crucially, without charging them commissions. What does this mean for the industry?
InPost’s new solution is designed to reshape how consumers shop, in line with what the company’s CEO, Rafał Brzoska, has been signaling for months.
“Traditional shopping will change as early as this year,” he said in May 2025 during InPost’s earnings call.
In the CEO’s view, a key growth path for Polish e-commerce companies lies in expanding cross-border – international sales.
At the end of 2025, Mr. Brzoska was appointed head of the Business Council established alongside Bielik, a Polish generative AI model. In December 2025, InPost integrated Bielik into its app, allowing InPost Mobile users to interact with the model.
“We want to support Bielik, which is why it will be embedded in the InPost app, helping to train it,” Mr. Brzoska said at the time.
InPost pushes into the tech arena
Now, collaboration is turning into a major challenge for e-commerce platforms such as Allegro, Amazon, and Erli.
“Parcel lockers are not logistics – they were the first technological solution we implemented. The mobile app is truly technology. Many people asked why we created InPost Pay and what the point was. Today, InPost Pay has become one of the key pieces of the puzzle,” said Rafał Brzoska, InPost’s CEO and founder, on March 16 at the Polish History Museum during the presentation of the new service.
Mr. Brzoska emphasized that the company’s goal is to simplify customers’ everyday activities.
“We want to operate in a way that makes customers love the brand. That is the biggest challenge, but it also brings the greatest satisfaction if we succeed,” he said.
“As pioneers, we are building services that no one expects. Did anyone think you could open a parcel locker via CarPlay in your car? No – but today everyone says ‘wow.’ We are constantly looking for new improvements,” Mr. Brzoska added.
The InPost CEO stated that the company is addressing unmet needs – especially in the era of the AI revolution.
“We will either find our place in this, or we will fail. Thanks to us, the way Poles shop online will change. You are witnessing the launch of a product that no one else in the world has created yet,” Mr. Brzoska declared.
InPost expands its app and introduces AI-powered shopping
The company has launched an AI-based shopping assistant. The tool allows users to search for products, compare offers, and complete purchases directly within the app. Delivery and returns are free under the InPost Plus program. The project’s “face” is a dog named Von Halsky.
As Rafał Brzoska explains, InPost revolutionized shopping through parcel lockers. The next step in the company’s evolution was the mobile app. The introduction of this new service aims to take e-commerce to the next level. The tool will not only help users find products but also compare them – and it presents an opportunity for sellers. According to InPost, merchants will pay no commission on purchases completed using the feature.
“Underneath, it’s technology – but on top, there are emotions. The new mobile app is a new home screen. ‘Buy’ will be the starting point of the shopping journey in our project. InPost intends to expand this new functionality. In the coming weeks, the service will reach app users, and by the end of the year, InPost is preparing 100 new features,” Rafał Brzoska said.
InPost AI and Von Halski will help you buy in the internet. Photo: InPost
A refreshed app and an intelligent shopping guide
The service is currently in beta and will soon be available to a trusted group of customers. After the beta tests, InPost plans to roll out a refreshed look for the InPost Mobile app. Updates will include a redesigned home screen to make it easier to navigate features related to sending and receiving parcels, as well as shopping.
The shopping process will be centralized in one place. Users can search for a product, compare offers, pay via InPost Pay, and track their delivery – all without needing to register with multiple stores or fill out additional forms.
The shopping guide will be a virtual assistant named Von Halsky, represented by a dog that symbolically “sniffs out” the best offers for the user.
Once the feature is launched, the customer enters a chat window. They simply type or speak what they are looking for, and within seconds, the system suggests matching products from millions of available listings.
“There are countless use cases. Imagine you’ve run out of a product, but you can’t remember the name or the store. Today, we go to a traditional marketplace and search our purchase history. My assistant remembers all my past purchases and can easily find them and add them to the cart. We are reducing the time needed to complete a purchase even further,” Rafał Brzoska explained.
According to Izabela Karolczyk-Szafrańska, InPost’s marketing and ESG director, the project’s goal is to simplify online shopping. The AI assistant is designed to reduce the time spent searching for offers and make the shopping experience more personalized. The character of Von Halsky was created by the agency PZL.
A new sales channel for online stores
The new feature is also set to benefit sellers. The AI assistant in the InPost Mobile app will serve as an additional sales channel. Shops can reach a wide base of app users without investing in their own technological solutions.
Importantly, the platform does not operate like a traditional marketplace. Sellers do not pay InPost a commission on sales and retain full control over their margins and pricing policies.
More than 4,000 merchants have already joined the program, including Modivo, RTV Euro AGD, X-Kom, Komputronik, Superpharm, Sportano, Lancerto, and Bonito. Through partnerships, the app will offer products from well-known brands such as Apple, Samsung, Unilever, Nestlé, Lego, and DeWalt.
Expert's perspective
How technology is reshaping the world of shopping
At the same time, shopping presents a far more complex technological challenge than it often appears. Issues such as user data privacy, massive volumes of product information, constantly changing inventories, and the need to make decisions in a dynamic environment all come into play.
For this reason, I am skeptical of a scenario in which shopping is dominated primarily by enormous, closed models, American chatbots, or AGI. The reason is simple: cost. Large models make sense where the value of decisions is immense, such as in drug discovery or materials design. Using them to pick a shirt for a first date is simply not economically justified.
I find more compelling the approach of developing specialized systems to automate shopping – built from multiple cooperating models, recommendation engines, agents, and integration with external tools.
In this context, InPost’s strategy is particularly interesting. It is not just about a single tool, but a broader direction focused on leveraging the latest models, building a wider technological ecosystem, and educating the market. Such an approach – combining technology development, partnerships (for example, support for Bielik AI), and capability-building – could, over the long term, genuinely reshape how e-commerce operates both in Poland and beyond.
Commission and e-commerce
According to InPost representatives, the AI tool is primarily designed to serve customers. Through the mobile app, users will gain access to fast and easy AI-assisted shopping. Sellers are expected to benefit as well – not only through access to the app’s 16 million users, but also through special conditions, most notably the absence of sales commissions, one of the main pain points of selling via marketplace platforms.
Marketplace models, such as those used by Amazon or Allegro, are built primarily around sales commissions, which sellers pay to gain access to the platform’s infrastructure, traffic, and customers. In practice, these commissions represent the most significant cost for merchants operating on such platforms. Most commissions are calculated as a percentage of the transaction value and vary depending on the product category and additional services, such as advertising or fulfillment.
Commissions generally range from around 5% to as high as 20–25% of the sales value. On Allegro, commissions range from 1% to 17.5% net, depending on the category. The average rate is typically 5–12% net, though in some categories it may be higher. When promotional costs and additional services are included, the effective cost of selling can reach roughly 30% of the product’s value.
On Amazon, the main cost is the so-called referral fee – a commission on each sale – which typically ranges from 8% to 15% depending on the product category. Sellers also pay subscription fees and incur additional logistics costs.
Tech giants target e-commerce – but face challenges
InPost’s move is part of a broader push by tech companies into e-commerce. In November 2025, OpenAI launched a shopping search feature allowing users to describe their purchasing needs in ChatGPT, which then generates a guide to help select products. OpenAI also introduced “instant check-out” from within the chatbot, theoretically enabling a customer to complete the entire purchase process – from search to payment – without leaving the chat window.
However, on March 4, The Information reported that OpenAI is partially scaling back the project. Specifically, the company is revising its strategy for shopping features in ChatGPT. It has abandoned plans to allow users to make payments directly within the chat window. Instead, final transactions will take place in external apps integrated with ChatGPT. The new approach is intended to better align with how users actually use the chatbot when shopping for products.
The decision reflects lessons learned from the project so far. While users often employed ChatGPT for searching and comparing products, they rarely used it to complete purchases. Meanwhile, the number of merchants using the chatbot’s checkout function was very small. Shopify data indicate that only a dozen or so of the millions of stores on its platform were selling products through this feature.
Under the revised model, purchases will be completed in partner apps within the ChatGPT ecosystem, such as Instacart, Target, Expedia, or Booking.com. In practice, once a user finds a product via chat, they are redirected to the seller’s app or website to finalize the transaction. OpenAI plans to continue improving product search and presentation within ChatGPT.
The shift in strategy underscores how difficult it is to integrate AI technology with the realities of online retail.
Expert's perspective
The future of AI in shopping
An AI agent can perform tasks on behalf of a user using the functions available to it. Most people using solutions like ChatGPT already do this, even if they don’t realize it. Commands such as “generate an image” or “find something online” are precisely a form of task delegation to an AI agent like ChatGPT.
One day, a fully functional “buy” feature will emerge. An agent, equipped with our delivery address and access to payment methods, will simply execute transactions according to the user’s expectations. Already, heavy users of AI chat tools treat them as a primary channel for decision-making. The agent will search the internet, identify an offer from a sufficiently trusted seller, with the best price and an acceptable delivery time, and then purchase the item.
This decision-making mechanism poses a serious challenge for online retailers. Users, bypassing individual websites, may stop perceiving specific stores as entities in their own right. The model favors large, trusted brands – trusted both by the buyer and the AI. An agent seeking to act responsibly will select the best offer, meaning price will become the dominant factor. It will not matter whether the product is available directly from the retailer or via a marketplace. Marketing, as we know it, will practically cease to exist, because the agent’s decision cannot be influenced.
Google and Amazon battle over technology
OpenAI is not alone. A true race is underway among OpenAI, Google, Amazon, PayPal, Visa, Mastercard, Klarna, and other e-commerce platforms.
For example, in May 2025, Google announced the development of e-commerce services in so-called AI Mode. This included not only intelligent search but also special product presentations, virtual try-ons, and agent-assisted checkout. In January 2026, Google introduced the Universal Commerce Protocol (UCP) and a toolkit for retailers to operate more effectively in the era of agent-driven shopping.
The system allows sellers to launch sales through AI tools via AI Mode. Dozens of entities have joined the initiative, including payment operators, Shopify, and Etsy.
Amazon, by contrast, has pursued a different strategy than OpenAI and Google. It is not building a new external discovery layer; instead, it is strengthening its own shopping ecosystem through the Rufus system. Rufus can search for products based on activity, deals, or purchase intent, automatically add items to the cart, identify the best price, find top deals, and even execute auto-buy when a target price is met.
At the same time, Amazon is defending its turf. Reuters reported that in March 2026, a California court temporarily blocked the agent-based shopping tool Perplexity on Amazon. This followed a lawsuit filed by Amazon in November 2025, in which the company accused Perplexity of illegally accessing user accounts through the Comet browser and AI agents.
The future of AI and technological possibilities
Expert's perspective
Potential lies in system cooperation
As consumers, we are far more willing to trust AI with price comparison and product discovery than with placing an order on our behalf. A U.S. YouGov survey found that 65% of respondents trust AI for price comparisons, 59% for product discovery, but only 14% for completing a purchase. This highlights where excitement ends and real customer behavior begins.
Today, I see the greatest potential for LLMs in e-commerce in three areas:
- Intelligent shopping assistants as a new interface layer – a conversation driven by intent (“I’m looking for trail running shoes; I have knee issues”).
- Deep personalization across the entire journey – dynamic content, recommendations, emails, Q&A, all consistent and based on customer data.
- Operational backend support – automating product descriptions, categorization, customer service, demand forecasting; areas where there is real margin impact, not visible in flashy presentations about “chat-window shopping.”
In other words, agent-driven LLMs have enormous potential in retail, but primarily as an intelligent layer on top of existing e-commerce rather than an attempt to wrest checkout from sellers and centralize it in a single chatbot. Anyone concluding from recent experiments that “AI conversations in sales don’t work” is simply overlooking the full picture. It is not the concept of a “shopping assistant” that failed – it is the overly aggressive idea of who should own the cart and customer data that stumbled.
Others are joining the race
The competition for technological development also involves e-commerce platform owners and the tools that support them. For example, Shopify is developing Shopify Magic and Sidekick, an AI suite for store management, marketing, catalog creation, administration, and e-commerce workflow support. Magic is a free set of AI functions integrated directly with the platform’s products. Salesforce, meanwhile, is developing Agentforce Commerce, which combines e-commerce, POS, and order management.
In December 2025, Klarna also announced its push into AI. The company created its own Agentic Product Protocol, providing AI agents with standardized access to 100 million products and 400 million prices. PayPal, in October 2025, launched Agentic Commerce Services, promising integrations with platforms such as Wix, Cymbio, Commerce, and Shopware, enabling sellers to make their products available on AI platforms including Perplexity.
The trend is clear. In early 2025, Capgemini reported that 71% of consumers wanted GenAI integrated into the shopping experience. Its 2026 report noted that 25% of consumers had already used GenAI tools for shopping in 2025, with another 31% planning to do so. At the same time, 71% of respondents expressed concerns about data usage.
Expert's perspective
The greatest potential lies in assistance
The more we “converse” with an LLM, the better it gets to know us. Over time, it understands context, remembers preferences and interaction history, so recommendations and discovery support simply improve. A good example is my own experience building a house: the AI already knows what tools I have, what I’m doing, and at what stage I am, so instead of a random product list, I receive sensible suggestions that actually match what I’m working on.
As for purchases executed directly by LLMs, that will likely happen sooner or later. Infrastructure for such scenarios is already emerging – for example, Stripe offers single-use payment cards that AI agents can utilize.
It is primarily a question of which categories will work best. Grocery or repeat-purchase items? Absolutely – this is easy to imagine today. But for more personal items, such as new shoes, where style, fit, or brand matter, it is much harder to trust an AI agent with the decision – though it can assist, for instance, through a virtual fitting room.
What it means for sellers
In Poland, roughly 150,000 companies sell online. A defining feature of the Polish market is the strong role of marketplaces, particularly Allegro. The platform remains the country’s main online sales channel, accounting for around 38.8% of the e-commerce market, clearly ahead of competitors including Amazon and AliExpress.
At the same time, the sector of independent online stores is expanding, operating across various technological platforms. BuiltWith data show that over 26,000 stores in Poland use WooCommerce, more than 20,000 run on Shopify, and over 17,000 are on Shoper.
Despite the dominance of marketplaces, independent stores are playing an increasingly important role – especially in terms of profitability. One of the sector’s fundamental challenges remains high competition and the significant commission fees charged by marketplace platforms.
Expert's perspective
The key to success in AI
A project like this must integrate demand with shelf availability, payments, taxes, and – most importantly – people’s trust that a chatbot is reliable enough to handle their credit card information.
That moment has not yet arrived. I have no doubt that millions of consumers are already using LLMs to make everyday shopping decisions – I do so myself every day – but not yet directly. I am confident that AI capable of completing purchases with integrated payment and delivery will become a reality. It will transform product searches into contextual conversations, turning queries like “I’m looking for running shoes in size X” into a discussion about me as a runner. The bond between the user and their agent will be crucial.
Even today, LLMs know more about us than our partners do. In the future, this relationship will supplant all existing relationships we have with brands. The shopping funnel of tomorrow will look completely different from today – moving from awareness to purchase in a single conversation.
Key Takeaways
- InPost enters e-commerce with an AI-powered shopping ecosystem. InPost has launched an AI shopping assistant within the InPost Mobile app, enabling users to search for products, compare offers, and complete purchases in one place. Users interact with a virtual assistant named Von Halsky, which recommends tailored products from millions of listings. The entire process requires no registration across multiple stores. Payments are processed via InPost Pay, while delivery and returns are free under the InPost Plus program. The service is currently in beta and will be rolled out more widely after testing concludes.
- A new sales channel for online merchants. InPost’s shopping assistant does not operate as a traditional marketplace. Sellers pay no commission on sales and retain full control over pricing and margins. For many businesses, this offers a way to reach a large user base without building their own technological infrastructure. Over 4,000 merchants have already joined the initiative, including major brands and retail chains. The app features products from manufacturers such as Apple, Samsung, and Lego.
- AI-assisted shopping is becoming a global technological trend. InPost is part of a broader technology race involving leading companies. OpenAI, Google, Amazon, Shopify, and Klarna are developing tools to help users search for products, compare offers, and even complete purchases via AI agents. Experts note that the advancement of large language models and recommendation systems could transform how purchasing decisions are made, potentially leading to a future where some transactions are executed directly by intelligent systems acting on behalf of users.
------
We are reporting this because we consider the information important and interesting. In the interest of full transparency, we note that RiO Fund, owned by Rafał Brzoska, CEO and shareholder of InPost, is an investor in XYZ.