In the past year, AI has almost become an unavoidable topic in the FMCG industry. Especially after the explosion of large models like DeepSeek, many distributor owners have felt for the first time that AI is not just for internet companies; it can also enter their warehouses, finance, business, products, customers, and reports. But the hotter it gets, the more chaotic it becomes. AI seems like a suddenly appearing universal tool: it can write copy, make posters, analyze spreadsheets, generate reports, and even help salespeople with visit scripts and help owners spot business problems. But the operational reality of FMCG distributors is far more complex than it appears. A change in gross margin might involve product mix, channel policies, customer payment terms, and salesperson actions; whether a report can be understood depends on whether data definitions, metric definitions, and business scenarios are truly understood. So, the question today is not whether distributors should use AI, but what kind of AI distributors actually need. With this question, New Distribution had an in-depth exchange with Zhou Zhen, Vice President of Zhoupu Data. In his view, distributors' interest in AI is genuine, but behind this interest, it is more about operational anxiety than technical anxiety. The real value of AI is not to create a tool that can chat, but to help distributors reconnect data, metrics, roles, and operational actions, gradually turning what once relied on owner experience and veteran salespeople's judgment into capabilities that the organization can replicate. Why Distributors Care About AI: Not Chasing Trends, But Because Operations Are Getting More Complex Distributors' interest in AI appears to be driven by the tech wave, but looking deeper, what really pushes them toward AI is the pressure of real operations. In recent years, the most direct feeling for many distributors is: business is still being done, goods are still being sold, but the old way of making operational judgments is increasingly insufficient. Previously, owners could look at sales, collections, and inventory to roughly know how the company was running. Now, even if sales remain, profits may be eroded bit by bit by expenses, payment terms, inventory, and pricing systems. What's more troublesome is that problems no longer appear in isolation but are intertwined. For example, a decline in gross margin may seem like prices are too low, but behind it could be changes in product mix, or certain channel policies being breached, or salespeople over-investing expenses in low-value customers to meet sales targets. As the number of customers, products, channel types, and expense items increases, owners cannot monitor every anomaly daily. When problems truly surface, they often have already turned from "small deviations" into "big losses." In Zhou Zhen's view, distributors' interest in AI today is not purely technical anxiety but operational anxiety. Distributors do not lack data. Orders are in the system, inventory is in the system, and a large amount of information about customers, products, collections, expenses, and salesperson actions is generated every day. But the problem is that most of this data is scattered and silent; it only becomes visible when compiled into reports. And from data to judgment, and from judgment to action, there are often many layers in between. Finance can pull reports but may not explain business problems; operations can see product sell-through but may not judge customer quality; salespeople visit stores daily but find it hard to turn their actions into reviewable and replicable experience. This is also why distributors have expectations for AI. On one hand, they hope AI can first help them "see problems," such as which customers' aging receivables are lengthening, which products' distribution rates are declining, and which salespeople have abnormal expense usage. On the other hand, they hope AI can further "explain problems," not just tell the owner that sales have declined, but continue to break it down: which region, which channel, which product. So, when distributors pay attention to AI, what they truly care about is not a new concept but a new operational capability: whether management actions that previously relied on manual monitoring, experience-based judgment, and meeting-driven execution can become more timely, clearer, and more executable. Behind Distributors Using AI Well: Data, Metrics, and Methodology Are Also Needed Distributors' most direct experience with AI often comes from general-purpose large models. You can throw an Excel file into it and ask it to analyze products; input a product selling point and have it generate marketing copy; tell it you want to visit a certain type of customer and have it compile a set of scripts. But in Zhou Zhen's view, the value of general AI cannot be underestimated, but its boundaries cannot be ignored either. It is best at solving expression, generation, and interaction problems, such as writing copy, making summaries, processing tables, and generating preliminary analyses. But the truly complex problems for distributors often lie not in whether an answer can be generated, but whether that answer is based on correct data, correct metrics, and correct business logic. Take a simple example. A distributor tells AI: the gross margin for a certain water product is 8%. Is that high or low? AI can certainly provide an analysis that looks complete. But the key question is whether an 8% gross margin is high or low, which needs to be viewed in the appropriate context. For mass-market beverages, it's one judgment; for premium alcoholic drinks, it might be another; in the KA channel, it's one situation; in the circulation channel, it's another. If peer comparison is needed, there must be a standard product library, standard classification, and unified definitions. Otherwise, AI can only guess based on common sense. Zhou Zhen repeatedly emphasized that the implementation of AI for distributors, on the surface, is about large model capabilities, but behind it, it is about data engineering, metric engineering, and business methodology. Many distributors think that as long as they feed enterprise data to AI and design a few prompts, they can get their own business analysis system. But the actual situation is far more complex. Distributor data comes from multiple systems and processes, such as orders, inventory, customers, products, expenses, collections, and salesperson visits. Different systems may define the same field differently, and different companies may understand the same metric differently. For example, the term "collections" in an operational context can be further broken down into concepts like cash collections, credit collections, outstanding amounts, and settled amounts. Similarly, "customer value" cannot be judged by sales alone; it also requires considering multiple dimensions such as gross profit, payment terms, expense occupation, purchase frequency, price sensitivity, and category mix. If these definitions are not clarified first, no matter how beautiful the analysis AI generates, it may just be judgment built on a vague foundation. In Zhou Zhen's view, the real difficulty in AI implementation for distributors is not prompts but three things. First is data cleaning: turning data scattered across different systems into analyzable data. Second is metric governance: turning business issues like gross margin, payment terms, expense efficiency, store value, and bad debt risk into clear metrics. Third is productization design: turning analysis results into actionable actions for the organization through daily reports, alerts, pushes, and one-click forwarding. It is based on this judgment that Zhoupu Data placed AI capabilities in the underlying construction of distributor business data early on. Zhou Zhen mentioned that Zhoupu's "Zhouyi" product did not start intelligent analysis only after the large model craze; it has been iterating around data cleaning, metric governance, and productized reminders for distributors over the past few years. For example, organizing data scattered across orders, inventory, customers, products, collections, and expenses, and then using anomaly alerts, daily business reports, peer comparisons, and key focus reminders to alert owners and management to problems in a timely manner. General AI is like a very capable assistant that can organize information faster and express it more clearly. But operational AI cannot just appear reasonable; it must withstand business validation. This is also the difference between general AI and operational AI. The former lowers the barrier to use, while the latter solves operational system problems. For distributors, the real hurdle to cross is not whether they can ask AI, but whether their data is clean, metrics are clear, scenarios are defined, and whether there is someone to follow up after analysis. AI Entering Distributor Organizations: From Control to Collaboration When discussing what AI has truly changed, Zhou Zhen did not start with distributors but with Zhoupu Data itself. A while ago, Zhoupu conducted an internal review, and overall R&D efficiency improved by about 35%. This number is more restrained than the "several times" often mentioned externally. Zhou Zhen explained that AI did indeed increase coding speed by two to three times, but requirements analysis and code review became slower. Because for AI to work accurately, the context, business rules, and field logic must be clearly explained in advance; after AI writes code, humans also need to re-evaluate quality. One end got faster, two ends got slower, resulting in a net 35%. But in Zhou Zhen's view, this precisely shows that AI brings not just efficiency improvements but a reconstruction of work methods. When this change is applied to distributors, it points not to simple efficiency gains but to shifting from "designing control rules" to "designing collaborative processes." In the past, distributors managed salespeople with one-size-fits-all rules. For example, giving each salesperson a fixed marketing budget monthly to maintain customer relationships, collect payments, and handle damaged goods; or if a customer's debt exceeded a limit, the system would not allow further orders. These rules seem clear, but in real operations, exceptions always arise. A salesperson might use up resources too quickly early in the month, leaving no budget for a key store's anniversary event at month-end; a customer's debt might exceed the limit, but there is a good order on the table. Give it or not? In the end, the problem is pushed back to the owner, who handles it temporarily with a "just this once" attitude. Over time, rules get eroded. "Essentially, it's because the quota design itself is unreasonable," Zhou Zhen believes. Traditional control can only handle standard actions, not the complex exceptions in real business. Where AI can intervene is precisely in breaking down what previously could only be judged by owner experience into computable, remindable, and collaborative factors. For example, based on a store's historical price negotiation frequency, purchase price fluctuations, customer value, and payment term risk, AI can suggest differentiated resource recommendations for different stores and salespeople. When a salesperson places an order, the system directly reminds: "For this store, it's recommended to invest two bottles, not three"; when a customer's debt exceeds the limit, instead of simply locking the order, it prompts to first collect the payment from two months ago, and then this order can proceed. These scenarios are exactly the product direction Zhoupu is advancing. Zhou Zhen mentioned that in the future, AI will not just show the owner a report but will go deeper into specific actions like salesperson order placement, customer collections, expense allocation, and product sell-through. This is what Zhou Zhen calls collaboration. Control is pressing rules onto people; collaboration is breaking rules into actions. In the past, owners would shout "control expenses, speed up turnover, improve collections" in meetings, but after the meeting, salespeople might not know what to do today; the value of AI is to turn judgment into specific reminders at the moment of ordering, collecting, replenishing, and visiting. Final Thoughts AI will not replace distributors' fundamental business skills; instead, it will amplify the gap in those skills. The clearer the data, the more useful AI is; the clearer the metrics, the more accurate the judgment; the clearer the processes, the smoother the collaboration. Conversely, if the company itself has chaotic data, rough rules, and unclear role actions, AI will only generate a bunch of seemingly complete but hard-to-implement answers faster. So, what distributors truly need to supplement is not just AI tool training, but using AI to re-examine their own operational systems. Due to space limitations, the exchange with Mr. Zhou Zhen covers far more than this. For more content, on May 28, Mr. Zhou Zhen will share in depth at the "Tower Alliance Distributor Team AI Hands-On Practical Course," where he will have in-depth exchanges and discussions with everyone.