AI-Driven Marketing: The E-commerce Revolution and the Dawn of the Agentic Internet (!)
Maciej Lesiak
- 12 minutes read - 2358 words
Ten artykuł jest dostępny również po polsku:
AI-Driven Marketing: rewolucja E-commerce i zapowiedź internetu agentowego (!)
What's in this article
Author’s Note: This text was 90% written in 2024 during an analysis of AI-driven marketing strategies of TEMU and SHEIN, which I did for myself as an after-hours project. I am publishing it after the US elections, which brought victory to Donald Trump and the imposition of tariffs, and after lawsuits were filed by the EU and UOKiK (Polish Office of Competition and Consumer Protection) against these platforms – which only confirms the importance of the issues discussed. This changes the profitability aspect, but the focus here is not on effects, but on mechanisms and strategies. I have completely removed the section on violations and manipulations, where I analyze persuasive mechanisms in more depth (analyzing specific marketing techniques). I encourage you to read it. I believe the text is even more relevant and interesting in the context of the upcoming revolution.
Scott Galloway, when writing his book “The Four,” in which he analyzes the aggressive strategies of major players in e-commerce, marketing, and consumer services, would have to consider a phenomenon that has emerged from China. In recent months, a new pair of leaders has appeared in the e-commerce industry, rapidly changing the face of trade. These are TEMU and SHEIN, two Chinese trading platforms that, thanks to aggressive marketing methods, support from AI behavioral analysis, and dumping prices, are conquering new markets.
Illustration: TEMU - an example of “forced gamification” from the European Commission’s allegations. TEMU forces users to play a “wheel of fortune” to access shopping. Of course, you always win the “highest prize” (100% refund), but the conditions are hidden in the terms and conditions. A classic time pressure technique with a countdown timer and fake discounts.
The Chinese philosophy of “zhà” - data explosion as a foundation
The key to understanding TEMU’s phenomenon is the concept of “zhà” - a Chinese term meaning explosion. As Connie Chan from Andreessen Horowitz explains, companies focused on AI-driven recommendations cannot afford a gradual development of their user base. Their recommendation algorithms do not work on small datasets. They need an immediate explosion of vast amounts of user data, and paid marketing is the fastest way to obtain them.
TikTok provides a perfect precedent for massive investment in user data here. The strategy began with the strategic acquisition of Musical.ly for $1 billion in November 2017 - ByteDance wasn’t just buying an app, but 60-80 million users (mainly from the USA) along with an already formed community creating short videos. This was an investment in a ready-made content consumption culture, not just numbers. Because, you see dear reader, the game being played now is about creating gateways to the internet that will dominate content consumption. In the age of AI, this is a colossal advantage.
As Evan Spiegel, CEO of Snap, noted: “No one in the United States anticipated the level of investment that ByteDance made in the American market… no startup could afford to invest billions of dollars in user acquisition worldwide… This wasn’t a strategy based on innovation; it was really about subsidizing user acquisition on a massive scale.”

The table below shows the dramatic contrast between marketing expenses and revenues during the key expansion period:
Year | Global Revenue | US Revenue (DemandSage) | US Revenue (Business of Apps) | Marketing Expenses | ROI |
---|---|---|---|---|---|
2018 | $0.15 billion | ~$12 million (8%) | N/A | $1.0 billion | -85% |
2019 | $0.34 billion | ~$28 million (8%) | N/A | ~$4.0 billion | -91.5% |
2020 | $2.0 billion | ~$160 million (8%) | $5.0 billion | N/A | Turning point |
2021 | $4.0 billion | $2.1 billion | $10.0 billion | N/A | Positive |
2022 | $11.64 billion | $5.96 billion | $20.0 billion | N/A | High |
2023 | $18.04 billion | $8.75 billion | $30.0 billion | N/A | 1800% vs 2020 |
2024* | $23.58 billion | $11.01 billion | $40.0 billion | N/A | Continuation |
*Estimated data
As the table shows, in 2018 ByteDance spent 6.7 times more on TikTok’s marketing ($1 billion) than the app earned from advertising ($0.15 billion). At one point, the company was spending an estimated $3 million daily just on acquisition marketing in the United States.
Illustration: A classic example of combining forced gamification with pressure selling. The effect of “winning” the wheel of fortune - TEMU shows a “100% back” coupon and immediately presents products with fake discounts.
It is crucial to understand that this was not a strategy of “renting” users, but “buying” them - ByteDance tested retention and engagement before scaling expenses. TikTok became Snap’s largest advertiser in 2018, meaning Snap was de facto selling advertising space to its direct competitor.
This seemingly irrational investment strategy turned out to be a brilliant move - by 2023, TikTok’s advertising revenue grew to $18 billion globally, an 1800% increase within 3 years from the turning point in 2020. This is a fundamental difference between the Western “innovation-led strategy” and the Chinese model of “subsidizing large-scale user acquisition.”
Measurement Chaos: When AI Eludes Traditional Methods
The data in the table above reveal a fascinating problem of the modern digital economy - dramatic discrepancies between sources. TikTok’s US revenue for 2023 ranges from $8.75 billion (DemandSage) to $30 billion (Business of Apps) - a 3.4x difference! This uncertainty is not accidental.
AI-driven platforms operate in a way that eludes traditional measurement methods. Is revenue from TikTok Shop “advertising revenue”? How to classify commissions from influencers? Where does advertising end and discovery-based commerce begin? These questions will become even more pressing in the era of the agentic internet, where the boundaries between advertising, recommendation, and transaction will completely blur.
TEMU is following an identical path - in 2023, PDD Holdings allocated nearly $2 billion for advertising on platforms like Meta and Google, knowing that it needed an addictive site with high-quality recommendations from the very beginning of its development.
TEMU’s complete arsenal of manipulative techniques in one interface. At the top, a “Discount Card” with a 65% discount and a countdown timer “Expires in 00:29:09:8”. Products have various pressure labels: “Special Offer” with its own timer, “Flash Deal”, “Bestseller” labels, fake reviews (3173, 3661, 940, 2687 reviews), and suggesting you won’t get a specific product (random color). Also visible is “just bought” - imitation pressure.
Unit Economics as an Investment in the Future
TEMU’s strategy seems financially irrational at first glance. Goldman Sachs reports that the company incurs total costs of about $5 to acquire each order worth $39 in the US, but after accounting for all operational costs (product, logistics, fulfillment), it loses about $6 per order in the US. In other markets, losses reach up to $18 per order. This is a significant improvement compared to 2022, when the company spent $16 to acquire each order worth $29.
However, this seemingly reckless spending strategy has a deeper justification. Each new user provides valuable data on purchasing behavior, which allows algorithms to be refined, optimal prices to be identified, and inventory needs to be forecasted. Moreover, attracting users to the platform creates a snowball effect - repeat visits or app referrals can trigger discounts and coupons that encourage further purchases.
Discovery-based vs Search-first: A Revolution in the Shopping Paradigm
TEMU introduces a fundamental change in the way we think about online shopping. While Amazon represents the “search-first” model - consciously entering a category and searching for a specific product - TEMU promotes “discovery-based shopping.” It’s the digital equivalent of entering a store for one item and leaving with a dozen purchases.
Just as TikTok offers an endless stream of videos, TEMU’s AI algorithm provides an endless “discovery” of new products that we didn’t need or never thought about, but which suddenly become necessary. This shopping model is much closer to the traditional shopping experience in a brick-and-mortar store and, importantly, is simply fun.
Illustration: TEMU transforms shopping into an online casino through gamification in push communication. PUSH notifications present a “Mystery Gift” to unwrap, a card game (“In appreciation of your support 💖 We offer a very low price on the item you added!”) and a “Gift Box from Temu🎁”. At the top, a disclaimer about SMS/email scams - ironic in the context of its own manipulative practices. Each notification has elements of gambling: uncertainty of reward, immediate gratification, time pressure.
Gamification as a Psychological Weapon
TEMU doesn’t limit itself to product recommendations. The platform implements advanced gamification systems that transform shopping into a gambling game. Users receive super offers, can spin a wheel of fortune (always winning the highest pool of coupons), and the system encourages sharing coupons with friends.
The platform deliberately creates time pressure through dedicated discounts with a one-hour timer, “ending soon” offers, or information about “low stock.” These techniques trigger a spiral of compulsive shopping, where every purchase or word-of-mouth marketing is a win for the platform.
Permanent AI-driven Adaptation: Massive Data Fueling Product Innovation
SHEIN demonstrates another dimension of AI-driven commerce’s power. The platform is capable of generating nearly 6,000 new products daily, testing through user behavior analysis what is most popular and desired. Based on AI recommendations, new designs are added, and the next day a trial batch is produced and shown to selected customers.
This continuous adaptation to the customer, fueled by the aggregation of unimaginable amounts of data and AI analysis of buyer behavior, allows new companies to adjust to market needs in real-time. No traditional company without gigantic budgets in the hundreds of millions of dollars can afford such a model of continuous adaptation.
Illustration: Continued gamification by TEMU in push notifications. Three different manipulative techniques: 1) “Items are waiting in your cart” with an animated dog and the text “It won’t happen again” - pressure selling through FOMO, 2) “Low stock” with a tied package and a warning “Your favorite items might be taken by others!” - fake scarcity, threat of stock depletion, pressure to act before others, 3) “Oh my! You’ve been selected💖” with an “extra discount on an item added to your cart” - pseudo-personalization and false urgency.
Sustainability Challenges and the Future of the Model
But does it add up? Despite impressive growth, TEMU’s business model raises questions about long-term sustainability. This problem is particularly visible in the era of trade wars and problems with carriers (subsidized flights from China). An aggressive pricing strategy, though effective in gaining market share, can lead to diminishing returns. Logistics also pose a challenge - most products shipped from China reach American consumers in over 10 days, significantly longer than Amazon’s two-day delivery.
Concerns about consumer trust are also emerging. Reports of low and inconsistent product quality, especially at such low prices, may deter customers from purchasing more expensive items in the future, potentially limiting the platform’s market scope and growth trajectory.
What does this mean for Poland and the EU?
The regulatory response to the strategy of Chinese platforms has already begun. In November 2024, the European Commission, together with the network of national consumer protection authorities (CPC Network), officially notified TEMU of a number of practices violating European consumer law. The allegations include: false discounts, pressure selling (“limited time offer”), forced gamification (necessity to play a “wheel of fortune”), misleading information about returns, false reviews, and hidden contact details.
- Fake discounts: Giving the false impression that products are offered with a discount where there is none.
- Pressure selling: Putting consumers under pressure to complete purchases using tactics like false claims about limited supplies or false purchase deadlines.
- Forced gamification: Forcing consumers to play a ‘spin the fortune wheel’ game to access the online marketplace, while hiding essential information about the conditions of use linked to the rewards of the game.
- Missing and misleading information: Displaying incomplete and incorrect information about consumers’ legal rights to return goods and receive refunds. Temu also fails to inform consumers in advance that their order needs to reach a certain minimum value before they can complete their purchase.
- Fake reviews: Giving inadequate information about how Temu ensures the authenticity of reviews published on its website. National authorities found reviews which they suspect to be unauthentic.
- Hidden contact details: Consumers cannot easily contact Temu for questions or complaints. European Commission
Trump’s tariffs, EU lawsuits, and UOKiK’s actions against TEMU and SHEIN are just the beginning. Europe faces a dilemma: how to protect local players from “data dumping” without stifling innovation? Polish e-commerce companies must quickly learn AI-driven strategies or risk complete marginalization.
The Agentic Internet: What’s Coming
The success of TEMU and SHEIN is just a foretaste of the revolution awaiting us. We are currently observing AI in the role of an advanced recommendation system and behavioral analysis. However, the upcoming agentic internet will introduce a qualitatively new level of automation and personalization. I already see preparations for a rapid change through modifications to frontends and website source code for more complex, semantic indexing. In a separate article, I will discuss ai.txt - attempts to manage the chaos we currently observe on servers, the Internet, and in e-commerce business.
We can realistically imagine AI agents that will:
- Proactively identify our needs before we realize them
- Negotiate prices in real-time on our behalf, search for promotions and best offers
- Automatically order products based on predictions of our future needs (integration with smart homes, further subscriptions, etc.)
- Coordinate purchases across different platforms to optimize costs and delivery times
In this context, the current strategies of TEMU and SHEIN, though revolutionary, are merely the first step towards a fully automated, AI-driven consumer economy. Companies that are not investing in understanding and implementing AI-driven marketing today risk complete marginalization in the upcoming era of the agentic internet.
The key question is: are we ready for a world where our purchasing decisions will increasingly be made by algorithms, and the traditional boundaries between need and desire will be completely blurred by intelligent recommendation systems? Completely abstracting from the cost that fast fashion represents for the environment in an era of climate crisis…
Sources
How Temu is taking on the giants of e-commerce - Maddyness UK
Temu: What It Is, and Why It Matters | Andreessen Horowitz
Temu’s Strategies for Explosive Growth - Alphabridge
How AI and Deep Learning Power Temu’s Explosive Growth - LinkedIn
TikTok Ad Revenue (2018–2027): Detailed Analysis - DemandSage
TikTok’s billion-dollar secret that wasn’t - Mobile Dev Memo
Related
- #2520 Manipulating Recommendation Systems: GROK, White Genocide, and Musk's Racist Conspiracy Theories
- Phatic Function in Practice: How ChatGPT's Conversation Maintenance Generates Millions in Losses
- AI Series: The final warning – AI's Self-Reflection on Its own development
- Case Study: Leak of sensitive airport data due to email configuration error (2007)
- Phatic Function in Practice: How ChatGPT's Conversation Maintenance Generates Millions in Losses
- GPTBot Is Scanning The Internet: How OpenAI Will Change Content Consumption and the Future of Search
- Bypassing Security Filters in ChatGPT's SVG Generation
- The Illusory Security of BIP: A Brief Technical Analysis of Security Measures
Support Independent Research

If you find this research valuable, consider supporting through Ko-fi. Your contribution helps maintain this project as an ad-free, independent resource.
COFFEE NOT BULLETSDirect support means more research, better content, and no ads.