王峰 AI × 经营 约一次诊断

总目 / 文章 / WF-0428

文章 · 存档

你最该交给AI的不是搜答案,而是那个卡了你三天的数据问题

财务人用AI用着用着总会有一种说不出的失落。问它概念,它能答;让它润色,它也能写;让它查一段政策、解释一个术语、整理几条会议纪要,它都像模像样。可真到了月底、到关账、到分析、到对账、到老板追着问数字的那一刻,最折磨人的那摊活,往往还在自己手里。那个卡了你三天的数据问题还是没动。真正让人疲惫的从来都不

很多财务人以为AI的价值在搜答案,可国际上的落地路径早就悄悄转向了更务实的方向:先把高频、重复、低附加值、却极其消耗人的数据工作接过去,释放人的判断力和分析力。说到底真正让财务人长时间陷在情绪里的,常常不是难题本身,而是那些看起来“不高级”、却极难脱手的脏活。你明明知道问题大概率出在数据口径、字段格式、历史版本或者跨表关系上,但就是要一行一行翻,一列一列试。最难受的是,这种活干完以后,别人几乎看不见。报表出来了,结论出来了,会议开完了,真正耗掉你两三天的,是那几万行数据背后那团没人愿意承认的混乱。

这篇文章真正想扭转的是很多人对AI的第一反应。说得再直接一点,文章的底层意思其实就是这句:“核心是扭转认知,AI最该干的不是仅仅变成一个搜索引擎,而是财务人每天最头疼的低价值重复劳动。

先说跨表匹配。这个场景几乎每个财务人都熟。客户名称不一致,合同编号有空格,金额字段有文本格式,A表和B表看起来像一回事,真正匹配的时候却总有一堆“差一点”。以前很多人只能在Excel里反复VLOOKUP、XLOOKUP、透视、去重、手工筛,最后把自己搞到怀疑人生。现在更高效的做法是直接让AI参与规则判断和清洗前置。你可以把需求写得非常具体,比如这样交给它:“我有两张表,A表是销售台账,B表是回款台账。请先统一客户名称中的空格、括号、简称和大小写,再按照客户名+合同号+金额区间做分层匹配,输出完全匹配、疑似匹配、未匹配三类结果,并说明每一类的判断依据。”这种提示词的价值不在于让AI替你拍板,而在于它先帮你把原本混在一起的动作拆开:字段清洗、规则定义、分层匹配、异常解释。你会突然发现,原来卡住三天的,不是问题太难,而是动作没有被拆清楚。

再说异常值清洗。很多经营分析做不下去根本不是因为不会分析,而是数据一打开就脏得没法看。负数混进收入,日期列有五种格式,金额里夹着逗号和文字,摘要字段同一件事写出七八种说法,系统导出的空值、重复值和乱码还在表里横着走。人一旦掉进这种泥潭,很容易越做越烦。AI在这里最有用的地方是先帮你把脏乱差变成有结构的问题。你可以这样给它:“请检查这份数据中的异常值和格式问题,识别日期格式不统一、金额字段非数字、重复记录、同义描述不一致、缺失关键字段五类问题,并分别给出修正建议、标准化规则和清洗后的字段结构。”这种做法,本质上是在让AI担任一个耐心极高、不会烦躁的初级数据整理员。清洗完以后,人再去复核重点,效率会有非常明显的变化。

历史数据追溯更是财务人的隐形噩梦。很多问题不是看不懂今天而是找不到昨天。某个费用口径去年怎么定义的,某个客户的历史折扣什么时候变过,某个项目的收入确认规则到底在哪个版本里改过,一旦涉及多期、多版本、多张表,光是翻历史文件就足够把人拖住。这个时候,AI的价值也很直接,它特别适合做“先替你把线索串起来”这件事。提示词完全可以写得像一次正式委托:“请根据我提供的历年台账、制度版本和说明文字,追溯某项指标在不同年份的字段来源、口径变化和计算逻辑,按时间顺序输出差异说明,并标记可能影响同比分析的节点。”它不会替代你的专业判断,但它会先把散落在各处的线头一根根拎出来,帮你节约最难熬的前半程。

还有一个特别常见也特别容易被低估的场景,就是格式统一。做过集团汇总、预算填报、经营周报的人都知道,这件事看起来像小事,实际上最吃人。十几家单位、几十个项目、几百个字段,一会儿有人把日期写成2025/3/1,一会儿有人写成2025年3月1日,一会儿列名是“客户名称”,一会儿又变成“客户全称”,最后花最多时间的,常常不是分析,而是把大家拽回一个标准格式。你完全可以这样交给AI:“请把这批来自不同部门的数据统一成同一模板,字段名按主模板映射,日期全部转成YYYY-MM-DD,金额保留两位小数,空值统一标记规则,客户字段按主数据字典校正,最后输出标准化后的表和未能自动纠正的问题清单。”当AI开始承担这种格式警察、口径协调员、初步核对员的角色时,财务人的脑力终于能从琐碎里抽出来。

这也是为什么越来越多领先的业财实践开始把AI放在流程前半段的脏活累活上,而不是只把它当一个会聊天的搜索框。

它先替你把单据结构化,把表格标准化,把异常挑出来,把疑点归类,把历史口径串起来。等这些动作稳定下来,后面的分析、判断、经营解释、决策支持才真的有空间长出来。说到底很多人迟迟没把AI用深是因为过去的讨论总喜欢围着概念打转,却很少有人愿意认真处理那些真正消耗人的细节。可财务工作的真实重量恰恰就压在这些细节里。那些看不见的苦活,那些没人夸奖的整理、核对、追溯、修正、统一,才是最先该被替代、被减轻、被重做流程的地方。

我的这本新书真正想做的也正是这件事:把这些长期躲在报表背后的脏活累活,一步一步拆成可复制、可照着做、可反复调用的流程,让AI别再停留在会回答,而是真正开始会干活。当它替你省下时间,你才有机会把精力留给分析,留给经营,留给判断,留给那些本来就该由财务人亲自完成的决策时刻。

What you should give to AI is not searching for answers

What you should give to AI is not searching for answers, but that data problem that has stuck you for three days

People in finance often feel a kind of quiet disappointment when they use AI for a while. You ask it for ideas, it can answer. You ask it to polish writing, it can do that too. You ask it to check a policy, explain a term, or organize some meeting notes, it all looks pretty good. But when it really comes to the end of the month, closing the books, doing analysis, reconciling numbers, or when the boss keeps asking about the numbers, the most painful part of the work is still in your own hands. That data problem that has stuck you for three days is still not solved. What really makes people tired is never that they don't know how to ask AI, but the endless work of pulling tables again and again, matching back and forth, cleaning bad data, checking old rules, making formats the same, fixing one round and then another problem shows up.

Many finance people think the value of AI is in searching for answers, but the real way people use it in the world has already quietly changed to something more practical: first take over the high-frequency, repetitive, low-value but very tiring data work, and free up human thinking and analysis. In the end, what keeps finance people stuck in bad feelings for a long time is often not the hard problem itself, but those “not fancy” jobs that are very hard to hand off. You clearly know the problem is most likely in data rules, field formats, old versions, or links between tables, but you still have to check row by row and try column by column. The hardest part is that after you finish this kind of work, almost no one can see it. The report is done, the conclusion is done, the meeting is over, but what really took your two or three days is the messy chaos behind tens of thousands of rows of data that no one wants to admit.

What this article really wants to change is many people's first reaction to AI. To say it more directly, the deeper meaning of this article is this sentence: “The core is to change your thinking. What AI should really do is not just become a search engine, but take over the low-value repetitive work that gives finance people the most headaches every day.

First, let's talk about matching across tables. Almost every finance person knows this situation. Customer names are not the same, contract numbers have spaces, amount fields are in text format, table A and table B look like they match, but when you really try to match them, there are always many “almost but not quite” cases. Before, many people could only keep using VLOOKUP, XLOOKUP, pivot tables, removing duplicates, and manual filtering in Excel, until they almost doubted themselves. Now a more efficient way is to let AI join in the rule checking and data cleaning first. You can write your request very clearly, for example: “I have two tables. Table A is a sales record, and Table B is a payment record. First, please make customer names the same by fixing spaces, brackets, short names, and upper/lower case. Then match by customer name + contract number + amount range, and give three results: fully matched, possibly matched, and not matched, and explain why for each group.” The value of this kind of prompt is not that AI makes the final decision for you, but that it helps you break down the mixed actions: cleaning fields, setting rules, matching in layers, and explaining exceptions. You will suddenly see that what stuck you for three days was not that the problem was too hard, but that the steps were not clearly broken down.

Next, let's talk about cleaning abnormal data. Many business analyses cannot move forward not because people cannot analyze, but because the data is too messy to even look at. Negative numbers appear in revenue, date columns have five different formats, amounts include commas and text, the same item is described in many different ways, and the system output still has empty values, duplicates, and strange codes everywhere. Once people fall into this mess, they get more and more frustrated. The most useful thing AI can do here is to turn messy data into structured problems. You can ask it like this: “Please check this data for abnormal values and format problems. Find five types of issues: inconsistent date formats, non-numeric amount fields, duplicate records, inconsistent descriptions, and missing key fields. Then give fix suggestions, standard rules, and a cleaned data structure.” This method is basically letting AI act like a very patient junior data helper who never gets tired. After cleaning, people can focus on checking the key points, and efficiency will improve a lot.

Tracking old data is an invisible nightmare for finance people. Many problems are not that you don't understand today, but that you cannot find yesterday. How a cost rule was defined last year, when a customer's discount changed, where a project's revenue rule was updated—once it involves many periods, many versions, and many tables, just reading old files can trap you for a long time. At this moment, the value of AI is very clear. It is very good at “connecting the clues for you first.” You can write a prompt like a real task request: “Based on the past records, rule documents, and explanations I provide, trace how a certain metric's data source, rules, and calculation logic changed over different years. Output the differences in time order, and mark points that may affect year-over-year analysis.” It will not replace your professional judgment, but it will first pull together the scattered pieces and save you from the hardest early stage.

There is also a very common but often underestimated situation: making formats the same. Anyone who has done group reporting, budgeting, or weekly business reports knows this looks like a small task but actually takes the most time. More than ten teams, dozens of projects, hundreds of fields—one person writes the date as 2025/3/1, another writes March 1, 2025, one column is called “Customer Name,” another is “Full Customer Name.” In the end, the most time is not spent on analysis, but on forcing everything back into one standard format. You can fully give this to AI: “Please standardize this data from different departments into one template. Map field names to the main template, convert all dates to YYYY-MM-DD, keep two decimal places for amounts, set rules for empty values, fix customer names using the master data list, and output the standardized table and a list of issues that could not be fixed automatically.” When AI starts to take the role of a format checker, rule coordinator, and first reviewer, finance people can finally free their minds from these small details.

This is why more and more leading business-finance practices start to use AI in the early steps of the process for these heavy and messy tasks, instead of only using it as a chat-based search tool.

It first structures documents, standardizes tables, finds abnormal data, groups suspicious points, and connects old rules. When these steps become stable, then real analysis, judgment, business explanation, and decision support can finally grow. In the end, many people have not used AI deeply because past discussions always stayed at the level of concepts, and very few people are willing to seriously deal with the details that truly take time. But the real weight of finance work lies exactly in these details. Those invisible hard tasks, those unpraised jobs of sorting, checking, tracing, fixing, and unifying, are the first things that should be replaced, reduced, and redesigned.

What my new book really wants to do is exactly this: take these hidden messy tasks behind reports and break them step by step into repeatable, easy-to-follow, reusable processes, so AI does not just “give answers,” but really starts to “do work.” When it saves your time, you can finally keep your energy for analysis, for business thinking, for judgment, and for those decisions that finance people should make themselves.