王峰 AI × 经营 约一次诊断

总目 / 文章 / WF-0427

文章 · 存档

财务人真正用AI不必从大项目开始

市面上关于AI的讨论太容易一上来就把人带进“大项目”“系统重构”“全流程自动化”“智能中台”这些庞大叙事里了。

听着很热闹,也很先进,可一旦落到日常工作台面上,很多人就会立刻安静下来:我的确不懂技术,也没啥技术团队,我每天手上的报表、票据、对账、核对、催收、归档都做不完,AI一天一个新花样新名词,我拿什么去做那样的转型?

全球会计行业这两年的调研也很说明问题。

面向全球特许会计师的调查显示,受访者里有85%至少“相当愿意”使用AI,但大家最大的障碍并不在“看不懂未来”,而在数据安全顾虑和技能不足,很多人的使用依然是个人自发,而不是组织化推进。换句话说,财务人并不排斥AI,真正卡住他们的,是不知道第一步该怎么迈。

所以财务人真正用AI最好别从宏大蓝图开始,而要从手边那件最烦、最碎、最重复、最容易量化的小事开始。国际上的成熟实践本来也在朝这个方向收敛。今天真正领先的方向已经不再是先搭一个“巨无霸”,而是让AI直接进到工作的流水线上,先接住那些原来最耗人的动作。

我越来越相信,财务岗位里最容易跑通的不是那些听上去很厉害的应用,而是最朴素的三个入口。

第一个入口,就是批量整理单据。这个场景已经非常普遍了,你发一张手写的单据,AI就会直接识别出来做结构化提取并放入excel表格中,然后再由人做最后核验,整个动作会从“纯体力劳动”变成“半自动审核”。工作没有消失,但人终于从抄写员变回判断者。

第二个入口,是快速清洗脏数据。每个做过经营分析、预算、月报、项目复盘的人都知道,真正吃时间的往往不是分析本身,而是前面那一大坨烂数据:日期格式不统一,客户名称前后不一致,金额字段夹着文字,摘要里充满口语化描述,系统导出的列名每次都不同。很多人一想到“AI做数据”就以为是上平台、建模型,其实最先释放生产力的常常只是让AI帮你统一口径、纠错、归类、补字段、改格式,把一团散沙先变成能用的表。

第三个入口,是自动生成基础报表。这里说的不是给董事会看的年度战略报告,也不是几十页的复杂管理分析,而是那些每天、每周、每月都要重复出现的基础输出:收入汇总、回款跟踪、费用归类、采购波动说明、应收账龄草稿、异常变动初稿。原来很多财务人最怕的,是明明内容高度重复,却还得每次从头来过。AI一旦接手模板化底稿和初稿生成,人就有了余力去看重点、抓异常、做判断,把时间真正花在有价值的地方。

说得更直白一点,AI最适合先接住高频、重复、规则较清楚的任务,然后在人的校验下持续迭代。它不是一把“按下去就万事大吉”的按钮,它更像一个能被训练出来的副手。

所以搞AI真正需要的根本不是一套遥远的方法论,而是一点立刻可用的确定感。

比如第一次把几十张发票交给AI整理,哪怕只省下半小时,那也是一次心理转折;

第一次把一张乱到让人头皮发麻的数据表清洗干净,哪怕只减少一次返工,那也是一次能力确认;

第一次让AI自动吐出一个还不错的基础报表初稿,哪怕自己还要改几轮,那也是一次关系重建。

你会第一次意识到AI是来把你从那些消耗性的动作里往外拽的。

会计与财务行业现在正处在一个很关键的路口。CPA.com在2025年的报告里提得很坦白:最成功的AI落地,通常都不是从“大而全”的改造开始,而是先选择1到3个高价值、可测量、范围可控的试点场景,跑通以后再逐步扩大。

这个判断我非常认同。

因为财务工作天然重准确、重责任、重可追溯,大跃进式推进看起来很猛,实际代价往往很高。小步快跑反而更适合这个行业的节奏:先把一个点做实,再把一个点变成一个面。

真正的变化往往不是某一天突然完成的,而是在很多个“先试一下”的瞬间里慢慢长出来的。今天省下十分钟,明天少返工一次,后天多看懂一个异常,再过一个月,你处理信息的速度、判断问题的方式、安排工作的节奏,都会悄悄变掉。到了那个时候你再回头看,会发现自己并没有参加什么轰轰烈烈的转型工程,却已经实实在在变成了一个会用AI工作的财务人。

别再等那个“万事俱备”的时刻了。

财务人真正用AI从来不用先摆出一个很大的阵仗。先从一张发票开始,先从一份脏数据开始,先从一张基础报表开始。先让它替你接住一个动作,再让它替你节约一段时间,最后你会发现,所谓生产力升级,原来就是这样一点一点长出来的。

而我更想做的,也正是把这件事说清楚、拆细、落到手上。

我的这本新书不是为了制造更大的焦虑,也不是为了让人仰望什么高不可攀的系统,它更像一个真的能放在工位边上的工具箱帮财务人先用起来,再一点点迭代下去,直到AI不再是一个话题,而是你每天都能摸得到、用得顺、离不开的日常能力。

do not need to start with big projects to really use AI

Finance people do not need to start with big projects to really use AI

When people talk about AI, it is very easy to pull everyone into big ideas like “big projects,” “system rebuild,” “full process automation,” and “smart platforms.”

It sounds exciting and advanced, but when it comes to daily work, many people become quiet right away: I really don't understand technology, I don't have a tech team, I already cannot finish my reports, invoices, reconciliations, checks, collections, and filing every day, AI has new tricks and new words every day, so how can I do that kind of big change?

Research in the global accounting industry over the past two years also shows this clearly.

A survey of chartered accountants around the world shows that 85% of people are at least “quite willing” to use AI, but the biggest problem is not that they “don't understand the future,” it is worry about data safety and lack of skills. Many people still use AI by themselves, not as part of a company plan. In simple words, finance people do not reject AI, what really stops them is that they don't know how to take the first step.

So when finance people really use AI, it is better not to start from big plans, but to start from the small, annoying, repetitive tasks right in front of them that are easy to measure.

In fact, good practice around the world is moving in this direction.

Today, the real leading direction is no longer to build a huge system first, but to let AI enter the workflow and take over the tasks that used to take the most time.

I more and more believe that the easiest things to make work in finance are not those “cool” applications, but three very simple starting points.

The first one is batch organizing documents. This is already very common. You upload a handwritten document, AI can read it, turn it into structured data, and put it into an Excel table, and then a person does the final check. The whole process changes from “pure manual work” to “semi-automatic review.” The work is still there, but people move from copy workers back to decision makers.

The second one is quickly cleaning messy data. Anyone who has done business analysis, budgeting, monthly reports, or project reviews knows that what really takes time is not the analysis itself, but the messy data before it: dates are in different formats, customer names are not the same, amount fields include text, notes are written in many different ways, and column names from the system are different every time. When people hear “AI for data,” they often think about big platforms or models, but the first real improvement often just comes from letting AI help you make rules the same, fix errors, group data, fill missing fields, and change formats, turning messy data into a usable table.

The third one is automatically creating basic reports. This does not mean big yearly strategy reports for the board, or long and complex analysis, but the simple outputs that repeat every day, every week, every month: revenue summaries, payment tracking, expense grouping, purchase change notes, accounts receivable aging drafts, and first drafts of abnormal changes. Before, many finance people were most afraid that even though the content was very similar, they still had to start from zero every time. Once AI takes over templates and first drafts, people have more time to focus on key points, find problems, and make judgments, and spend time on what really matters.

In simple words, AI is best at taking over tasks that are frequent, repetitive, and have clear rules, and then improving step by step with human checking. It is not a magic button that fixes everything at once, it is more like a helper that can be trained.

So what people really need for AI is not some far-away theory, but a small feeling that “this works right now.”

For example, the first time you let AI organize dozens of invoices, even if it only saves you 30 minutes, that is already a change in how you feel;

The first time you clean a messy table that used to make your head hurt, even if it only reduces one round of rework, that is a sign of real ability;

The first time AI gives you a decent first draft of a report, even if you still need to edit it, that is a new kind of working relationship.

You will realize for the first time that AI is here to pull you out of those tiring tasks.

The accounting and finance industry is now at a very important point.

CPA.com said clearly in its 2025 report: the most successful AI use does not start from big and complete changes, but from choosing 1 to 3 high-value, measurable, and controllable pilot tasks, and then growing from there.

I strongly agree with this.

Because finance work needs high accuracy, strong responsibility, and clear tracking, big fast changes may look powerful, but the real cost is often high. Small and fast steps fit this industry better: first make one point work well, then grow it into a larger area.

Real change usually does not happen in one big moment, but grows slowly through many small “let's try this” moments.

Today you save 10 minutes, tomorrow you reduce one rework, the day after you understand one more problem, and after one month, your speed, your way of thinking, and your work rhythm will quietly change. When you look back, you will see that you did not join some big transformation project, but you have truly become a finance person who can work with AI.

So don't wait for the “perfect time.”

Finance people do not need to make a big show to start using AI. Start with one invoice, one messy dataset, one simple report. Let AI take over one task, save you a bit of time, and step by step you will see that real productivity growth comes little by little.

And what I really want to do is exactly to explain this clearly, break it down, and make it usable.

My new book is not here to create more stress, and not to make people look up to some unreachable system. It is more like a real toolbox you can put next to your desk, helping finance people start using AI, and then improve step by step, until AI is no longer just a topic, but something you can use every day, easily and naturally, and cannot work without.