The FMCG industry is often perceived as "traditional," "low-tech," and "digitally lagging." This impression is both right and wrong. It's right because the FMCG industry indeed has a vast network of offline distributors, mom-and-pop stores, and wholesale markets. These capillary channels have extremely low digitalization, with many distributors still managing inventory via spreadsheets or even handwritten ledgers. It's wrong because multinational giants like P&G, Unilever, Nestlé, and Coca-Cola have cumulatively invested over tens of billions of dollars in digitalization over the past decade. P&G's supply chain system is as complex as any top-tier manufacturing company, and Coca-Cola has over 20 million sales points globally, each backed by a complex data stream. The core reason FMCG AI adoption lags far behind verticals like legal and education is that the industry's vast data remains largely untapped. This is precisely where AI opportunity lies. More importantly, the FMCG industry has a structural characteristic absent in other verticals: extremely long decision chains, but extremely high trial-and-error costs at every decision node. A new product takes an average of 18 to 36 months from concept to launch, with R&D costs ranging from millions to tens of millions. Yet the industry's accepted new product failure rate exceeds 70%, meaning 5 out of 7 new products die at launch. A promotional campaign involves millions in subsidies and ad spend, but how much converts to actual sales and how much is intercepted by competitors is only known post-mortem. This review often takes 1-2 months after the campaign ends, by which time it's too late. Checking whether shelf displays in a region meet standards traditionally requires sales reps to visit stores, take photos, upload them to headquarters, and have staff review them. A mid-sized brand may have over 100,000 retail points. Even quarterly inspections per store would be an astronomical workload, let alone issues of data authenticity and timeliness. Behind every pain point lies a scenario where AI can intervene.
Why FMCG AI Is Underrated?
Three systemic reasons explain why the FMCG AI sector is undervalued. First, industry tradition creates cognitive bias among AI entrepreneurs. When we say "AI sector," the first reaction is tech companies: OpenAI, Anthropic, Google DeepMind. When we say "FMCG sector," the first reaction is P&G, Coca-Cola, Nestlé. These two terms occupy different worlds in most people's minds. But in fact, P&G announced AI as a core corporate strategy as early as 2018. Unilever's 2022 annual report dedicated a chapter to AI applications in R&D, supply chain, and marketing. Nestlé established a dedicated AI research center to develop new food formulas. These actions within the FMCG industry haven't received widespread tech media coverage, so most AI practitioners are unaware of the FMCG market. Second, budget allocation leads to market obscurity. FMCG companies' AI spending has never had a dedicated AI procurement budget line. It's scattered across departments: supply chain system upgrades, marketing technology investments, digital transformation budgets, ERP optimization costs... So if you try to estimate the FMCG AI market size, you'll struggle to get a precise number because this spending has never been tracked separately. But ask P&G's CDO, and they'll tell you their annual AI-related spending exceeds the total funding raised by most AI startups. Third, sales cycles are relatively long. AI startup valuations largely depend on rapidly growing ARR. But FMCG procurement decision cycles are far longer than the SaaS industry average. An AI solution entering P&G's supplier system may face a 6-12 month evaluation period, including compliance reviews, IT security audits, and POC validation... This deters many AI startups. But from another angle, high barriers mean that once you're in, the moat is half-built.
AI Opportunity Map Across the FMCG Value Chain
To systematically understand FMCG AI opportunities, we must break down the value chain. From raw material procurement to consumer reach, the FMCG value chain roughly divides into six stages: raw material procurement → R&D formulation → production manufacturing → supply chain logistics → channel sales → consumer reach. Each stage has different AI penetration potential and opportunity windows.
1. Raw Material Procurement: Overlooked Front-End Opportunity
Raw material price volatility is one of the biggest sources of profit uncertainty for FMCG companies. Coffee beans, palm oil, cocoa, sugar... these bulk agricultural commodities are subject to weather, geopolitics, exchange rates, and other factors, causing dramatic fluctuations. Traditional responses rely on procurement teams' experience or hedging via futures institutions. But experience has limited accuracy, and futures hedging is costly. AI's entry point: use machine learning models to integrate climate data, satellite imagery, futures market data, and geopolitical signals to build price prediction models for bulk raw materials, helping procurement teams make better decisions 3-6 months in advance. Additionally, supplier compliance traceability is a significant scenario. The EU's EUDR requires companies to prove their raw material supply chains aren't involved in deforestation, with severe penalties for non-compliance. AI combined with blockchain can build a traceable digital supply chain system, especially critical for categories like palm oil and cocoa with complex sourcing. Representative players: Sourcemap (supply chain transparency), Gro Intelligence (agricultural data and forecasting).
2. R&D Formulation: Least Competition, Deepest Moat
This is the scenario I believe is most worth betting on in the entire FMCG AI sector. The reasons are simple: minimal competition, deep barriers, and extremely clear ROI. First, minimal competition. Companies doing AI-driven formulation generation are few globally: Brightseed, Analytical Flavor Systems, Givaudan's (traditional flavor and fragrance giant) AI department, and some lab-stage startups. This track has no BAT, no Microsoft, no Salesforce, because the scenario is too vertical—big companies overlook it, and small companies lack sufficient food science expertise. Second, deep barriers. What's the core of AI formulation generation? Molecular-level food databases. What taste, mouthfeel, and stability a molecule produces at specific concentrations, temperatures, and media—these data require extensive real-world experiments, not just internet scraping. Whoever accumulates a sufficiently large formulation experiment database first builds an unassailable data moat. Brightseed's model is a great example. They use AI to analyze the bioactivity of plant compounds (phytonutrients), helping food companies discover functional ingredients that exist in nature but have never been commercially developed. Before their AI system, this field relied entirely on scientists' experience and literature searches, with extremely low efficiency. Brightseed compressed discovery cycles from years to months, and each discovery enriches their molecular database. Analytical Flavor Systems took a different path: they built a sensory prediction model that predicts consumer taste preferences before a physical product is made. What does this mean? A food company previously needed to recruit hundreds of consumers for sensory tests, lasting months and costing millions; now, AI delivers predictions in seconds. Moreover, AI prediction accuracy improves continuously with more experimental data input. This is the power of the data flywheel: usage → data accumulation → better models → more usage → more data. How clear is the ROI? A successful new product can bring hundreds of millions or even billions in sales. If AI can raise new product success rates from 30% to 50%, the value to brands is priceless. Compared to that, an AI formulation platform's annual fee is a rounding error. Entry Strategy: Start with a single category (e.g., beverages or snacks), build deep partnerships with 1-2 leading brands, train proprietary models on their historical formulation and new product data, then expand to more brands as the "leading AI formulation platform within the category." Key Risks: Long sales cycles (6-12 months); strict food safety regulations requiring human validation of AI-generated formulas; some large brands may attempt to build in-house AI formulation capabilities.
3. Production Manufacturing: Relatively Mature AI, But Niche Opportunities Remain
AI applications in production are relatively mature, with established solutions for visual inspection, predictive equipment maintenance, and scheduling optimization. The challenge in this segment is the need for extensive on-site implementation and high customization to each factory's processes, making standardization and replication difficult. Additionally, large companies like Hikvision and SenseTime have mature visual AI products, leading to intense price competition. However, one niche scenario deserves attention: food safety AI. Food safety incidents are devastating to FMCG brands. Traditional quality inspection relies on manual sampling and lab testing, with blind spots and long detection cycles. AI visual inspection can achieve 100% online real-time detection, eliminating food safety hazards on the production line. What's special here: food safety is a compliance bottom line, not optional. Once a problem occurs, the cost far exceeds AI system procurement. This makes demand rigid with strong willingness to pay.
4. Supply Chain and Demand Forecasting: Landscape Set, But Startup Opportunities Remain
Supply chain AI is the most competitive segment in FMCG AI. o9 Solutions, Kinaxis, Blue Yonder (acquired by Panasonic), SAP IBP... these players have deeply penetrated large FMCG companies' supply chains, and the landscape is largely established. But there's an overlooked niche: the mid-market. Large FMCG companies can spend millions of dollars on platforms like o9, but a Chinese local brand with annual revenue between 500 million and 5 billion yuan can't afford that price or the IT team to implement such heavy systems. Brands of this size are the main force in China's FMCG market. Their supply chains are complex, but their tool options are far more limited than big brands. Lightweight, SaaS-based, out-of-the-box supply chain AI tools are the right approach for this market.
5. Channel Sales: Fastest Results, Easiest Entry
This is the second most noteworthy scenario after new product R&D AI. The reason: ROI can be quantified within 90 days, contract cycles are short, and workflow integration is deep. The core pain points in channel sales can be summarized in three questions:
- Does my product's shelf display meet standards at retail? (Shelf execution issue)
- Did this promotion actually drive sell-through, or was it just distributor channel stuffing? (Promotion authenticity issue)
- Which outlets might stock out, and which might have excess inventory? (Inventory health issue) Traditional solutions rely on people: sales reps visiting stores, distributor feedback, and market research firm data. But these methods suffer from time lags, data distortion, and incomplete coverage. AI offers two entry paths: Path One: Image recognition + shelf intelligence. Have sales reps take a photo of the shelf with a mobile app, and AI identifies within seconds: which SKUs are on the shelf, inventory levels for each SKU, whether display positions meet brand standards, and competitors' shelf share... The value is immediate. For a mid-sized brand, raising shelf execution compliance from 60% to 80% directly translates to measurable sales increases. Trax is the global leader in this space, serving top brands like Coca-Cola, Budweiser, and Johnson & Johnson. Path Two: Sell-through data analytics + promotion optimization. Integrate POS data (real-time sell-out data from retail terminals) into AI models, combine with external signals like weather, holidays, and competitor promotions, predict sell-through trends for each SKU by region, and automatically generate optimal promotion recommendations. The moat here lies in workflow integration. When sales reps habitually open the app each morning for store visit tasks and alerts, when sales directors use the system's data for weekly regional reviews, and when distributors manage inventory through the platform... replacing the system isn't a cost issue; it's the collapse of the entire sales operation system. Special Opportunity in China: China's FMCG channel structure is extremely complex, with modern channels (supermarkets, convenience stores), traditional channels (grocery stores, mom-and-pop shops), and e-commerce channels (Taobao, Douyin, Meituan Flash Purchase) coexisting, with fragmented data. A platform that integrates all-channel data and provides unified insights has enormous value in China.
6. Consumer Reach: Most Competitive, Most Homogenized
AIGC for ad content, personalized recommendations, social media sentiment monitoring... this direction is already a red ocean. Jasper, Persado, Alimama, ByteDance's Ocean Engine... too many players to count. It's not that there's no opportunity, but this direction has entered a highly competitive phase, making true differentiation difficult. Unless you can do one thing: connect consumer data with upstream value chain stages. For example, if your AI system can automatically transmit consumer discussions on social media about a category (e.g., "many people are discussing low-sugar drinks lately") to R&D for formulation direction ("initiate a low-sugar version project"), and then to supply chain for early material preparation ("increase low-sugar raw material procurement by 30% ahead of time")—that's true AI-driven full-chain decision-making. This is a difficult vision to achieve, but once realized, the moat is very deep.
Three Special Opportunities in the Chinese Market
China's FMCG AI opportunities have unique aspects compared to the global market, warranting separate analysis.
1. Distributor Digitalization
China has over 1 million FMCG distributors, mostly small and medium-sized family businesses. Their digitalization is extremely low, but they hold key data nodes from brands to retail terminals. What goods enter warehouses, what leaves, which outlets are ordering, and which payments are outstanding. Brands can't see this data (or see delayed and distorted versions), but it's crucial for precision marketing and supply chain management. Whoever helps distributors digitalize gains access to the channel data brands crave most. This isn't a pure AI problem but a systematic AI + SaaS + services challenge. Provide distributors with free or low-cost inventory management tools, help them digitalize daily operations, collect channel data, and package it into data insight products sold to brands.
2. Private Domain Operations Combined with AI
China is the world's most developed private domain e-commerce market. Brands have accumulated large user bases in the WeChat ecosystem (official accounts, mini-programs, WeCom), but most brands still run private domain operations manually: posting on Moments, replying to customer inquiries, and designing member activities by hand... AI has three very practical entry points here:
- First, intelligent customer service and sales assistants, using AI to replace repetitive customer inquiries while identifying high-value users and proactively triggering sales conversations;
- Second, personalized content generation, automatically creating personalized product recommendations and marketing copy based on purchase history and behavioral preferences;
- Third, repurchase prediction and churn warning, predicting which users are about to churn (e.g., no purchase for over 60 days and no message opens), triggering retention actions in advance. Competition in this scenario is relatively fragmented, with no dominant player yet.
3. Douyin E-commerce Data Intelligence
Over the past three years, Douyin e-commerce has become a major incremental channel for Chinese FMCG. But brands' operations on Douyin remain largely manual: selecting influencers by experience, setting ad spend by intuition, and adjusting strategies based on post-hoc reviews. What can AI do here? Predict ROI for influencer collaborations (telling you before the campaign how many conversions a given influencer might bring), automatically optimize ad strategies (dynamically adjusting bids and budget allocation based on real-time data), and monitor competitors and trends (which new category is exploding, which competitor is gaining momentum). Data access is somewhat challenging (Douyin's API openness is limited), but once the data barrier is broken, the value is enormous.
Moat Determines Survival:
Why Most FMCG AI Companies Will Die
Having a scenario isn't enough. The most common reason FMCG AI startups fail isn't technology but failing to build a moat, eventually becoming an easily replaceable tool. I categorize FMCG AI moats into five levels, from shallow to deep: Level 1: Feature moat (shallowest, inevitably caught up within 6-12 months) Your AI accuracy is high, but competitors' AI will also improve. Your interface is user-friendly, but competitors will copy it. Relying purely on feature differences is the weakest moat and what most FMCG AI companies currently depend on. Level 2: Product experience moat (medium) Workflow design is smooth enough that users develop habits, and switching creates friction costs. This is stronger than Level 1 but still replaceable. Level 3: Workflow integration moat (strong) Your product is embedded in customers' core business processes. Sales reps use your app daily for store visits, sales directors use your data system weekly for reviews, and procurement teams use your prediction model for material planning. Replacing you means rebuilding entire business processes, retraining staff, and re-accumulating data. This cost is real. Level 4: Data flywheel moat (extremely strong) Your product accumulates proprietary data assets competitors can't replicate. After 100 brand formulation projects, you have 100,000 formulation-market performance training data points; after serving 500 distributors, you have three years of channel sell-through history... How many years would competitors need to start from zero? Level 5: Ecosystem and industry standard moat (ultimate moat) You become industry infrastructure. Not just one brand uses you, but the entire industry does; your platform becomes the standard data interface between brands and distributors. Think of Veeva's position in life sciences—the top 20 pharma companies all use it, and no one dares not to, because not using it means disconnecting from the industry ecosystem. So, the key question for FMCG AI startups is only one: what level of moat can you achieve? Most teams, honestly, are still oscillating between Levels 1 and 2, busy running sales, optimizing features, and responding to customer demands. Levels 3 and above require strategic design from day one.
Competitive Landscape: Who Are the Real Threats?
Regarding risks, the FMCG AI sector faces three potential competitive threats that need identification. Threat One: Will foundational model vendors enter the market? Will OpenAI, Anthropic, or Google directly enter FMCG vertical scenarios? In the short term, unlikely. Two reasons:
First, FMCG AI requires extensive industry data accumulation and workflow customization, which isn't foundational model vendors' strength;
Second, the FMCG market is too fragmented for them, less attractive than high-ticket verticals like healthcare or legal. But in the medium to long term, if AI Agent technology matures to automatically handle highly customized industry tasks, the threat becomes real. The response strategy: before the AI Agent era arrives, complete workflow integration and data flywheel construction so your moat can't be replaced by general AI. Threat Two: Traditional ERP/SaaS vendors' AI upgrades SAP has deeply integrated AI into its S/4HANA system; Salesforce launched Consumer Goods Cloud specifically for FMCG; Oracle added AI prediction capabilities to its supply chain modules. These giants' advantage is existing system and service contracts with large FMCG companies, allowing seamless cross-selling of AI modules. But their disadvantage is being general solutions with insufficient industry depth, and their product iteration speed is far slower than focused startups. Startups' response: establish strongholds in niche scenarios and mid-market customers that giants don't cover, while making data assets deep enough that giants can't quickly replicate. Threat Three: FMCG giants building in-house AI capabilities P&G, Unilever, Nestlé, etc., have enough capital to build in-house AI teams. But there's a structural issue: FMCG companies' core competencies are brands, channels, and consumer insights, not AI technology. They can hire AI engineers, but it's hard to build a stronger AI team than specialized AI companies, and even harder to achieve scale effects across multiple clients' data (since their data is only their own). In-house AI is an inevitable trend for big brands, but it doesn't affect vertical AI companies' survival. P&G building its own AI formulation tool doesn't mean they don't need external consumer insight AI or external channel sales AI. Big brands' in-house efforts often focus on core strategic scenarios, while peripheral scenarios still procure external solutions.
Three Pieces of Advice for FMCG AI Entrepreneurs
If you're starting or considering entering the FMCG AI sector, here are three pieces of advice.
- Enter from the depths of pain points; don't build a big, all-encompassing platform. Every niche scenario in FMCG has enough market capacity for a company to reach unicorn status. Building a comprehensive FMCG AI platform sounds appealing, but you'd have to sell to procurement, R&D, and sales departments, each with different decision chains, making your sales efficiency extremely low. The right approach is to excel in one scenario, become the de facto standard there, then expand to adjacent scenarios. Trax only did shelf recognition and became the global leader. Then they expanded to sell-through data analytics and distributor management. That's the right path.
- Design data assets as your most important strategic asset from day one. From the first day, think: what data will my product accumulate during use? What value does this data have? How can I build network effects from data? Not all AI products can form data flywheels. If you're building a pure process automation tool (e.g., automatic report generation), the data generated during use doesn't help model improvement, so you have no data flywheel, only feature moat. Deliberately design product usage to generate valuable training data. Each shelf recognition improves the recognition model; each formulation test enriches the formulation database; each promotion feedback optimizes the promotion prediction model.
- Find strategic customers who can give you data. FMCG AI's cold start problem is more severe than other sectors. Because proprietary FMCG data is scattered across each company's systems, you can't access any public data to train models. The only way to solve cold start is to find 1-2 leading brand customers willing to co-build with you, exchanging data for customized AI solutions that solve a specific, painful problem they can't ignore. In the process, you gain valuable industry data, they get solutions—a win-win. Such strategic customers aren't necessarily the largest companies but those with sufficient data volume, openness to new technology, and real pain points in your target scenario. Finding them is more important than fundraising.
Conclusion: How Long Is the Window?
Any systematically undervalued sector will eventually be revalued; it's just a matter of time. In 2021-2022, no one thought AI code generation was a big market, but GitHub Copilot's release redefined developer tools. In 2022-2023, AI customer service was considered "technology not good enough, hard to implement," but now many companies use AI to handle 80% of customer conversations. FMCG AI is at a similar inflection point: technology is sufficient (multimodal AI, LLMs have provided good foundation models for shelf recognition, formulation generation, and text analysis), industry acceptance is high enough, but the competitive landscape isn't established (most scenarios lack clear leaders). The window is roughly 2-3 years. After this window, either clear leaders emerge in scenarios, making it hard for latecomers to disrupt, or AI Agent technology matures to automate most scenarios, fundamentally changing the competitive logic. Entering now is perfectly timed. Waiting longer might be too late. The window is open. The next critical question is: where should FMCG companies start? Which scenarios are worth doing first? How can AI truly enter new product development, marketing, sales, supply chain, and organizational management? On May 27-28, in Hangzhou, the 2026 China FMCG Conference AI Application Forum and Baidu NARA AGENT Launch will systematically dissect FMCG AI cognition, scenarios, cases, tools, and organizational implementation. From trend judgments to enterprise cases, from large model applications to practical operations for brands and distributors, this forum aims to explore with the industry how AI can truly rewrite FMCG growth. For event details, scan the QR code to add the WeChat account and contact the organizing committee.