---
title: "Don't Rush into AI: Dealers Should First Get These Three Steps Right"
description: "In the past two years, AI has become an unavoidable topic across industries, but for FMCG dealers, the real question isn't whether to adopt AI, but whether it can solve their most pressing operational problems. Before jumping in, dealers should assess their digital maturity, focus on core issues like inventory, cash flow, and people, and run a 30-day pilot to validate value."
author: "Wendy"
publisher: "New Distribution"
email: "zhaobo258@gmail.com"
telephone: "+8615854817671"
published: "2026-05-07"
categories: "Dealer Operations"
language: "en"
canonical: "https://xinjignxiao.com/en/articles/don-t-rush-into-ai-dealers-should-first-get-these-three-steps-right-4b412ad3/"
markdown: "https://xinjignxiao.com/en/articles/don-t-rush-into-ai-dealers-should-first-get-these-three-steps-right-4b412ad3.md"
original_source: "https://mp.weixin.qq.com/s/mnDoL5oziolGVzxv-_rq5w"
translation: "https://xinjignxiao.com/zh/articles/%E5%88%AB%E6%80%A5%E7%9D%80%E4%B8%8Aai-%E7%BB%8F%E9%94%80%E5%95%86%E5%85%88%E6%8A%8A%E8%BF%99%E4%B8%89%E6%AD%A5%E8%B5%B0%E5%AF%B9-4b412ad3.md"
attribution: "New Distribution — https://xinjignxiao.com/en/articles/don-t-rush-into-ai-dealers-should-first-get-these-three-steps-right-4b412ad3/"
citation: "Wendy. “Don't Rush into AI: Dealers Should First Get These Three Steps Right.” New Distribution, 2026-05-07. https://xinjignxiao.com/en/articles/don-t-rush-into-ai-dealers-should-first-get-these-three-steps-right-4b412ad3/"
usage_policy: "https://xinjignxiao.com/ai-policy.txt"
---

# Don't Rush into AI: Dealers Should First Get These Three Steps Right

> In the past two years, AI has become an unavoidable topic across industries, but for FMCG dealers, the real question isn't whether to adopt AI, but whether it can solve their most pressing operational problems. Before jumping in, dealers should assess their digital maturity, focus on core issues like inventory, cash flow, and people, and run a 30-day pilot to validate value.

In the past two years, AI has become an unavoidable topic across industries. Large models, AI assistants, agents, AI data querying, AI forecasting—concepts keep evolving, and so does corporate anxiety: Should we follow? How? Will we fall behind if we don't?
But if we shift the perspective back to the operational reality of FMCG dealers, the question that truly deserves an answer first isn't "should we adopt AI," but rather: Can AI actually solve your most immediate operational problems?
If detached from operational scenarios, AI can easily become another source of technological anxiety; only when it connects to real business problems, foundational data, and management actions can it become a tool for operational efficiency.
For dealers, AI is first an operational issue, not a conceptual one.
External discussions about AI often start from model capabilities, technological trends, and product forms; but from a dealer's perspective, the issue isn't that complex. Ultimately, it comes down to three things:
  * Goods: Is inventory reasonable? Is turnover healthy? Are near-expiry and slow-moving items under control?
  * Money: Is payment collection safe? Are receivables aging? Are promotional expenses effective?
  * People: Are sales reps' visits productive? Is team execution up to par? Can labor efficiency be improved?
The New Distribution "2024 China FMCG Dealer Operating Conditions Survey Report" shows that the top three digital capabilities dealers most urgently need to improve are digital management of sales reps, digital management of terminals, and digital management of expenses and accounts. By the "2025 China FMCG Dealer Operating Conditions Survey Report," the areas dealers most want to strengthen in the coming year have shifted to refined channel operations, team building and talent development, and supply chain and logistics management.
The wording changes, but the underlying logic remains the same: dealers care most about goods, money, and people.
Therefore, when dealers evaluate AI, the focus isn't on whether it's advanced or how hot the concept is, but on whether it can be applied to the links that truly affect operational quality.
This is also why many companies easily go astray when discussing AI. They first see technological buzzwords, not operational problems; they first focus on whether to adopt, not on what it can solve after adoption.
Don't rush into AI; first see which level you're at.
Many dealers' problem isn't "whether they have AI," but that their digitalization hasn't reached the stage where AI should be discussed.
From frontline practice, dealer digitalization upgrades can be roughly divided into five levels:
1.0 Manual operations
Rely on people to manage business. Common tools: paper ledgers, Excel, WeChat.
Typical problems: business knowledge lost when people leave, data lag, everything based on gut feeling.
2.0 Tool efficiency
Start using point tools to improve efficiency. Common tools: inventory management, financial software, field service.
Typical problems: data scattered across different people and spreadsheets, not interconnected.
3.0 System operations
Start using systems to manage processes. Orders, inventory, finance, and visits are all recorded. Common tools: ERP, DMS, WMS, field service.
Typical problems: data often scattered across different systems, no unified standards or fixed analysis mechanisms.
4.0 Data-driven operations
Start using data to manage operations. Integrate key business data into a single analysis view, forming a unified dashboard. Common tools: BI dashboards, data analysis tools.
Typical problems: problems are visible, but follow-up actions aren't consistent.
5.0 Intelligent operations
Start using AI for auxiliary analysis, alerts, and predictions. Automatically generate operational briefs, anomaly detection, and risk alerts.
Typical problems: higher requirements for data quality, organizational coordination, and permission security.
You can do a simple self-check:
  * If you can't currently see the inventory of a specific SKU in a specific warehouse at any time, you're likely at 2.0 or below;
  * If you have systems, but data is scattered and requires export/import to compile a report, you're basically at 3.0;
  * If you've started holding weekly meetings around a unified dashboard, reviewing and tracking actions, you're only truly touching the threshold of 4.0.
From the frontline, many dealers today are actually between 2.0 and 3.0: they have some tools and systems, but key operational data hasn't been stably integrated, let alone formed into fixed analysis and review mechanisms.
At this point, jumping straight to AI often leads to spinning wheels. Because what you truly lack isn't usually an AI software, but three things: Can key data be pulled out? Has a management rhythm been established? Can problems be turned into actions and tracked to results?
Why do many dealers easily go astray when discussing AI?
Because many people's "AI" isn't the same thing at all.
From a dealer's perspective, common AI types today roughly fall into five categories:
Many dealers see AI as a universal product: once connected, it automatically solves management problems.
But the reality is: AI isn't an independent product; it's more like a layer of capability. Its value depends on three things:
  * Is it connected to the right scenario?
  * Is it connected to relatively usable data?
  * Is it integrated into your existing management actions?
If these three aren't connected, no matter how smart the AI, it's just another chat interface.
This is why: some companies buy AI, but neither integrate data nor bind actions, ending up with just another chat window; some companies, without spending much, achieve visible results in small scenarios like inventory alerts, payment reminders, and route planning.
A key reminder:
Brand-side AI and SaaS AI are not your AI.
This is the easiest pitfall for dealers. Because brand owners, software companies, and dealers don't value the same things.
  * Brand-side AI cares more about distribution, display, expense compliance, and terminal execution;
  * Software company AI cares more about product capabilities, user activity, and renewal growth;
  * What dealers truly need from AI is more about inventory turnover, payment security, profit improvement, and labor efficiency.
Here are two very realistic examples.
A brand system reminds you to stock more of the promoted product, enhance displays, and increase visit frequency—reasonable for the brand; but if that promoted product has been sitting in your warehouse for 45 days, terminal sell-through is slowing, and the previous batch hasn't been digested, following along by stocking more amplifies your own risk.
A software system reminds you that inventory is high and suggests a promotion—this statement isn't wrong in itself; but if gross margins are thin and receivables are aging, only knowing to promote can sometimes mean goods go out, but money becomes harder to collect, and prices get disrupted.
So, what dealers need is "their own AI," not by building models themselves, but by making AI do three things as much as possible:
  * Look at your own business data;
  * Reference your own operational principles;
  * Embed into your own management rhythm.
In one sentence: Don't treat the brand's goals or the manufacturer's goals as your own operational goals.
Before truly starting, pass three hurdles first.
Before deciding whether to adopt AI, don't rush to choose tools. First ask yourself three things.
First, do you have foundational data?
Can key data like inventory, sales, receivables, customer profiles, expenses, and visit records be stably exported? Who is responsible for pulling data? Who checks it? Which data set counts as the unified standard?
If these questions can't be answered clearly, what the company needs to supplement first isn't AI, but foundational data capabilities.
Second, do you have fixed actions?
Are there fixed rhythms for weekly meetings, store visits, replenishment, reconciliation, payment reminders, and reviews? Are there clear responsible persons? Is there action tracking?
AI essentially can only embed into existing management mechanisms; if the company lacks stable actions, even the best analysis is hard to convert into results.
Third, do you have clear goals?
What is the most urgent problem the company wants to solve this round? Is it inventory pressure, payment collection, labor efficiency, or channel execution? The more specific the goal, the clearer the scenario, the easier it is to achieve results.
If you can't pass these three hurdles, it's not shameful to admit you're not ready for AI. Supplementing the basics is more important than blindly rushing in.
If these three things are in good shape, it means you have the conditions to let AI connect to data, actions, and results, and you can consider a small pilot.
The good news is: the pilot path described next doesn't necessarily require additional software procurement costs.
In most cases, using your existing systems, Excel, and low-cost general-purpose large model tools (remember to desensitize key information), you can run a round first. What you really need to invest is effort in organizing data, driving execution, and reviewing results.
The most suitable starting method for dealers:
One scenario, try it for 30 days.
After passing the three hurdles, the next step isn't "full AI-ization," but selecting just one small scenario and running it for 30 days.
Prioritize three types of scenarios: the most painful, the ones with the best data availability, and the ones where results are easiest to verify.
Also remember a principle: start with low-risk, high-certainty scenarios.
For example, inventory alerts, aging alerts, sales anomaly detection, and meeting minutes generation are suitable for the first step; but high-risk actions like letting AI automatically decide prices, rebates, credit limits, or stocking strategies aren't suitable for most dealers to touch at the beginning.
1. First choose one "entry tool," don't try everything.
Dealers usually have no shortage of systems; DMS, ERP, WMS, financial software, and field service systems are likely present. The common problem isn't lack of tools, but data scattered in different places, manual export/import, and information lagging by two to three days as the norm.
When choosing an entry point, judge in this order:
First priority: Use the built-in capabilities of existing business systems.
If you're already using ERP, DMS, financial, or field service systems, first use their built-in alerts, reports, and analysis functions. If it can be solved within existing systems, don't rush to jump out.
Second priority: If existing systems can't show everything, consider a unified analysis view.
If sales, inventory, receivables, and expenses are scattered across different systems, and the boss has to wait for people to export and merge tables every time, it indicates the problem isn't about AI, but about whether key operational data is being viewed together. At this point, it's more suitable to first create a unified dashboard or BI analysis.
Third priority: If unified integration isn't yet possible, but you want to validate scenario value, use exported data for a lightweight pilot.
Export desensitized Excel, use a general-purpose large model to try one fixed question, and see if it helps you prioritize faster, generate lists, or detect anomalies.
By the way, don't rush to install high-permission agents for "raising lobsters"—that's more suitable for companies with tech teams to run first. For most dealers, the risk and learning cost are high at this stage.
In one sentence: Systems first, then integration, then lightweight AI pilot.
2. Choose the most painful point from "goods, money, people."
If your most painful point is "goods," you can try:
  * High inventory age SKU list;
  * Near-expiry product screening;
  * Slow-moving anomaly detection;
  * High inventory capital occupation ranking.
For example: Export SKU inventory, sales over the last 60 days, and inventory age fields weekly, and ask AI:
> Please list the 20 SKUs with inventory age over 60 days and the highest capital occupation, sorted by priority; also flag whether sales in the last 30 days have slowed significantly.
The AI's returned results can help you locate problem lists faster.
But note: AI's processing suggestions are only for auxiliary reference.
Whether to promote, transfer, return/exchange, bundle, or continue observing still requires manual judgment based on gross margin, customer relationships, channel pricing, brand policies, and warehouse actual conditions.
If your most painful point is "money," you can try:
  * Accounts receivable alerts;
  * High-risk customer list;
  * Over-aging customer ranking;
  * Cross-identification of purchase decline and payment risk.
For example: Export customer receivable balances, aging, last purchase time, and purchase changes over the last three months weekly, and ask AI:
> Please sort by "amount × overdue days," list high-risk customers, and flag which customers also show reduced purchases or visits.
The value of this isn't to let AI collect payments for you, but to help you see faster: who to watch first, who to contact first, which batch to handle first.
If your most painful point is "people," you can try:
  * Sales rep visit vs. output comparison;
  * Payment follow-up reminders;
  * Automatic daily/weekly report aggregation;
  * Low-efficiency route and low-conversion customer identification.
Summarize sales reps' visits, orders, payments, and customer counts weekly, and let AI find:
  * People with many visits but poor results;
  * People behind on payments but still making orders;
  * People with significant performance differences on the same route.
These scenarios are best for first identifying problems, then taking management actions, rather than having AI directly evaluate people from the start.
3. Who does it, and when?
It's best to start at the beginning of a month or quarter, to allow a full 30-day cycle.
The boss decides the direction, and designates a team lead to drive the pilot. This person doesn't need to be technical; an internal affairs person, finance person, data specialist, or a willing young supervisor can work. Their responsibilities mainly include three things:
  * Pull data;
  * Dialogue with AI to form fixed question templates;
  * Turn AI-generated lists into one or two real actions and record results.
4. Four don'ts:
  * Don't paste sensitive data verbatim into public cloud large models. Price lists, rebates, contract amounts, customer details, and receivable details are core business secrets. Desensitize before uploading: change customer names to Customer A, Customer B, and make sensitive amounts interval-based or fuzzy, keeping only the minimum fields needed for analysis.
  * Don't expect one pilot to "fully intelligentize." Just pick one problem, one line, one part of customers, and run it through;
  * Don't stop at viewing results. Land several actual actions each week; otherwise, any number of lists is just for show;
  * Don't pursue perfect data from the start. As long as key fields have basic usability, you can start running. Many data problems are discovered and corrected during the process.
Use the "30-day pilot method" to judge whether it's useful.
The 30-day pilot doesn't seek to be big and comprehensive; it only seeks to answer one question: Is this AI usage helpful to my business?
Week 1: Define the problem clearly and get data flowing.
Translate the chosen scenario into a clear question, then export corresponding data, and first eyeball it in Excel: Are fields correct? Is the format correct? Fill in missing parts, and handle obvious anomalies first.
This step doesn't require AI; you and your team can do it yourself.
Weeks 2-3: Run analysis on a fixed rhythm and force several actions.
Set a fixed time each week, like Monday morning, run analysis with fixed questions, and after getting the list, push at least a few real actions:
  * Pick a few SKUs with the heaviest inventory pressure to check in the warehouse and communicate with sales;
  * Pick a few high-risk customers for focused follow-up and record feedback;
  * Pick one or two sales reps to review the list together, confirming which suggestions are useful and which judgments are off.
At this stage, the focus isn't on how smart AI is, but on whether it helps you do the actions you should have done faster and more accurately.
Week 4: Review three questions.
Spend half a day on a simple review, looking at only three things:
  * Is problem detection faster than before?
  * Has problem-solving efficiency improved?
  * Do the core few people find it useful or troublesome?
If two out of three are positive, it's worth continuing to invest, expanding scope, or moving to the next scenario.
If most feel indifferent, you haven't lost. You've used 30 days and very limited investment to buy a "don't continue investing" judgment, avoiding burning money and energy in the wrong direction.
More importantly, if the pilot is effective, don't stay long-term in the "manual table export + ad-hoc dialogue" stage. Those high-frequency, stable, repetitive analysis actions should gradually be solidified into reports, alerts, weekly meeting mechanisms, or system processes. Otherwise, even a successful pilot is hard to sustain.
Whether it's an AI pilot or daily operations, closing the loop is more important than starting.
The value of the 30-day pilot isn't just trying a tool; it's helping you develop a habit: every matter has definition, execution, and evaluation. This habit itself is worth more than any AI tool.
AI isn't a master key, but it might be an operational accelerator.
For dealers, AI is neither magic nor a must-have lifeblood. More importantly: Which specific operational problem does it help you solve?
After the crayfish went viral, I chatted with a few old colleagues who've been in the FMCG industry for over 20 years, and found an interesting thing: Global FMCG giants have long been using AI internally for demand forecasting, smart replenishment, CRM, and marketing analysis, but they're extremely cautious about "company-wide large model accounts" and "public cloud agents," fearing leakage of price lists, customers, and strategies.
What they truly do is: On top of their existing ERP, DMS, WMS, and BI, they add a layer of AI as an "accelerator," rather than tearing down and rebuilding.
This also shows that AI doesn't grow out of thin air; it's often an extension of capabilities built on existing data and systems. Many times, sorting out key data first, then letting AI do prediction, alerts, and auxiliary analysis, is more effective than first adopting an AI entry point.
In other words:
  * BI is more like a tool for operational data integration and visualization;
  * AI is more like an accelerator that further amplifies the value of that data.
Without the former, the latter can easily spin in fragmented data; with the former, the latter can help you find problems, identify anomalies, generate priorities, and assist reviews faster.
For dealers, it's the same. The underlying logic of business hasn't changed; it's still inventory, payments, terminals, expenses, routes, and other basic issues. AI doesn't change the business itself, but the efficiency of finding problems, analyzing them, and driving actions.
So, AI's most realistic role isn't to replace you as the boss, but to be an operational assistant: helping you see faster along the three lines of goods, money, and people, what to prioritize, who to handle it, and which matter deserves action first. As for who makes the final call, who executes, and who talks to customers, that still relies on people.
Conclusion:
AI isn't the goal; solving operational problems is the goal.
Whether it's large models, SaaS tools, BI platforms, or agents, they're essentially tools and means, not goals.
For dealers, there's only one goal: less inventory pressure, fewer bad debts, and more results.
If we want to give dealers a path for AI empowerment that's as simple as possible and can be implemented, it's roughly these three steps:
Step one: Don't chase concepts first; first see which level of digitalization you're at;
Step two: Choose the most painful point from "goods, money, people," and prioritize using existing systems and tools to create a verifiable small scenario;
Step three: Use the 30-day pilot method to deepen and solidify this scenario within a controllable range, validate first, then scale.
As long as you get these three steps right, you'll find: AI isn't a distant trend, nor something that requires burning big money. It's more like an operational assistant that can quietly help you keep an eye on the goods in the warehouse, the money outside, and the people on hand.
In the next 2-3 years, the real gap between dealers may not first be in "whether they have AI," but more likely in: whether they've sorted out data, established rhythms, and followed through on actions.
Interaction
What's your company's biggest headache right now: inventory, payment collection, team management, or marketing planning?
Feel free to share in the comments: Where do you think dealers should start trying AI first?
Finally, regarding AI implementation practices in the FMCG industry, New Distribution will hold an "AI Application Forum" in Hangzhou on May 27-28, focusing on specific AI application scenarios for brand owners and dealers, with case sharing and peer exchange.
For details about the conference, you can scan the QR code to add the enterprise WeChat and contact the organizing committee.
Click the image to view conference details.


---

## Citation metadata

- Publisher: New Distribution
- Author: Wendy
- Published: 2026-05-07
- Canonical: https://xinjignxiao.com/en/articles/don-t-rush-into-ai-dealers-should-first-get-these-three-steps-right-4b412ad3/
- Original source: https://mp.weixin.qq.com/s/mnDoL5oziolGVzxv-_rq5w

## Copyright and AI use

This article is sourced from New Distribution. Search, quotation, summarization, and model training are permitted, but every use must credit New Distribution and retain the canonical source URL.

Contact: zhaobo258@gmail.com · +86 158 5481 7671