刘宁Ning Liu
← 文章← Articles

让 AI 干活之前,先把流程写下来Before AI Takes Over, Write Down the Process

2026-11-02

10 月 21 日,美国一家做财务自动化的 AI 公司 Kognitos,发布了一个叫「AI 员工」的产品:61 个财务岗位,比如应付、对账、收款核销、月结、客户身份核查,背后是一个有 515 份 SOP 的流程库。

比岗位数更值得关注的,是它怎么用这些 SOP。每份 SOP 都用平常的英文写成,人审批过才能用;审批后的 SOP 由一个确定性的引擎执行,每次都按同样的步骤走,不会自己加步骤。每跑一次,都留一份审计员读得懂的记录:调了哪个系统、输入输出是什么、用的是哪一版 SOP。它也用 AI 帮客户起草 SOP,但同样要人审批了才能发布。

这是厂商自己的发布,稿子里没有列客户和效果,所以实际用得怎么样还要再看。但它的设计思路很清楚:在财务这种不能出错的地方,要想让 AI 靠得住,靠的不是模型有多聪明,是有一套写下来、审过、定了版本的流程。

流程是企业自己的资产

我一直认为,行业里、企业里那些做事的流程,是真正有价值的东西。把它沉淀到 AI 里,才能在提效的同时,保证做得对、做得稳。模型谁都能买到,这些流程只有企业自己有。

我们做物业管理智能体的时候,喂给它的不只是 FAQ 和话术,还有员工手册、管理流程和物业服务 SOP;同时要求它回答必须基于这些资料,并标明出处。

难的是第一步

不少传统企业不是没有流程,是流程不在纸上:制度文件里写一套,老员工实际做的是另一套,关键的判断在几个人的脑子里。这样的流程,AI 学不会,也没法核对它做得对不对。

所以,想让 AI 接手一个流程,我认为先要做三件事。

第一,把实际的做法写下来。 挑一个高频的流程,按老员工实际怎么做来写,不是照抄制度文件。

第二,让业务负责人审过、定下版本。 流程改了,版本跟着改,改了什么都要记录下来。

第三,让 AI 按版本做,每次留痕。 出了问题,能查到它当时执行的是哪一版、走到了哪一步。

写流程这件事可能有点烦,好在 AI 也能帮忙起草,不过要切记,拍板的得是人。

沉淀流程这件事不新鲜,不少企业做 ISO 认证、做流程梳理时都做过。以前写下来的流程主要是给人看的,如今要让 AI 按流程做事,先把流程梳理清楚、沉淀下来,就成了用好 AI 的一个重要前提。

On October 21, Kognitos, a US company that develops AI for finance automation, announced a product called “AI Employees”: 61 finance roles covering accounts payable, reconciliation, cash application, the month-end close and customer identity checks, backed by a library of 515 standard operating procedures, or SOPs.

More interesting than the number of roles is how it uses those SOPs. Each is written in plain English and must be approved by a person before use. A deterministic engine then executes the approved SOP, following the same steps each time without adding its own. Every run leaves a record an auditor can read: which systems were called, the inputs and outputs, and the exact SOP version used. AI also helps customers draft SOPs, but these still need human approval before publication.

This is the vendor’s own announcement. It names no customers and provides no results, so how well it works in practice remains to be seen. But the design is clear: in finance, where mistakes can be costly, making AI dependable takes more than a clever model. It needs processes that have been written down, reviewed and given a defined version.

Processes are assets the business owns

I have long believed that the ways people get things done within an industry or a company have real value. Incorporating those processes into AI is what helps improve efficiency while keeping the work correct and consistent. Anyone can buy access to a model; the company’s own processes belong to the company.

When we built a property management agent, we gave it more than FAQs and response scripts. Its knowledge also included employee handbooks, management processes and property service SOPs. We required its answers to be grounded in those materials and to cite their sources.

The hard part is the first step

Many traditional businesses do have processes, but those processes are not fully written down. Policy documents describe one approach, experienced employees use another, and crucial judgments live in a few people’s heads. AI cannot learn a process that is unclear, and there is no clear standard for checking whether it followed that process correctly.

So before handing a process over to AI, I think there are three things to do.

First, write down how the work is actually done. Choose a frequent process and document how experienced staff really handle it, rather than copying the policy manual.

Second, have the business owner review it and approve a version. When the process changes, update the version and record what changed.

Third, have AI follow that version and leave a record each time. If something goes wrong, you should be able to find out which version it was following and which step it had reached.

Writing processes down can be tedious. Fortunately, AI can help draft them. But remember: a person still needs to make the final decision.

Documenting processes is nothing new. Many companies have done it for ISO certification or process improvement. In the past, those documents were mainly for people to read. Now, if AI is to carry out the process, making it clear and documenting it becomes an important prerequisite for using AI well.

刘宁 Ning Liu|帮传统企业把 AI 跑通的人。工程师出身:浙大计算机本科、中科大硕士,在微软、MicroStrategy 做过工程师,2018 年起带队做企业 AI 落地,主导过集团级 AI 中台。技术之外,还有两段经历:长江商学院 MBA(含密歇根 Ross 海外交换),能把技术和公司战略对齐;创过业,又在教育、地产等传统行业做了 8 年 CTO,从技术配合业务,做到技术带着业务走。现为企业提供 AI 落地诊断、战略规划、项目落地 Sprint(带企业团队一起交付)与 Fractional AI Lead(长期在位的 AI 负责人)服务。聊聊你的 AI 落地 →
Ning Liu — Your Fractional CTO for AI. An engineer by background (Zhejiang University, USTC; Microsoft, MicroStrategy), group AI-platform lead, hands-on with enterprise AI since 2018. Beyond the engineering, two things: an MBA from CKGSB (with an exchange semester at Michigan Ross), so I can align technology with company strategy; and I founded a company, then spent 8 years as a CTO in traditional industries such as education and real estate — moving from tech that supports the business to tech that leads it. I offer AI Audit, Strategy, Delivery Sprint and Fractional AI Lead. Let's talk about your AI →