AI Won't Break SAP, But It Will Break Per-Seat Pricing
AI will not flatten what enterprise software has accumulated, but it will break the per-seat pricing model. SAP's defense is not training its own foundation model — it is open partnership plus business understanding, and the real battlefield is the last mile: process, data and organizational inertia.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Pricing, not code, is where AI actually hits SAP
AI's impact on SAP is not at the code layer but on the business model. SAP's share price has come back down from a high of $300 to $160, but stretched back to 2021 it is still up more than 20%; Adobe has fallen from around 480 to roughly 220, and Zoom has fallen to almost nothing. What actually gets replaced is single-point SaaS. The core of the impact is the traditional per-seat pricing model, which will shift toward charging by consumption and even by outcome — a challenge every enterprise software company has to face.
— Yuan XinSAP won't train a foundation model; it partners openly instead
SAP does not train a general-purpose large model of its own, because foundation models are evolving too fast, and because what SAP is good at is enterprise process, data and interaction. Its approach is to stay open to all third-party models and, on model selection, to fit the most suitable model to the most suitable scenario. In parallel, it trains proprietary small models on its own structured table data — RPT 1.5 — for finance scenarios such as collections forecasting, where the accuracy requirement is close to 100%, because in finance 99% accuracy is still unusable.
— Yuan XinThe boss wants AI, but the dirty data is still there
The biggest headache for CIOs at large enterprises is the boss reading some article on WeChat and tossing it over with ‘I want that too’. The gap is that everyone assumes that now AI has arrived, the dirty, grinding work no longer has to be done. AI can speed up data cleaning, but it cannot make that process disappear. Getting AI to land still requires FDEs to comb through historical data and organize it by business object, and at the same time to distill the unstructured judgment of the veteran hands, settling it into enterprise memory — only then does the share of work that can be executed automatically gradually rise.
— Yuan XinFDEs and ERP consultants will merge into one hybrid role
The FDE is not a relabeled on-site engineer. A traditional ERP consultant is an industry expert who enters from the business side and ultimately lands requirements inside a standard product; an FDE is a product builder deployed to the customer early, and the capability that accumulates flows back into the product. Early FDEs leaned toward low-level engineering and lacked a translator to turn business into coding requirements, which is why Anthropic and OpenAI have started working with consulting firms like McKinsey and Accenture. The two groups will merge into a composite role that understands business and technology at once — exactly the route that spillover from a small circle into the enterprise market has to take.
— Yuan XinCustoms agent accuracy flipped once old process assets were reused
One large innovative company built its own agent to handle customs declarations for 33 countries and got only 60-70% accuracy; people then had to re-check the remaining errors, which produced widespread complaints. After SAP came in, relying on GTC, its global tax platform, for a basic understanding of tax in those 33 countries, plus template configuration in document AI to replace long-context tokens, accuracy jumped immediately to ninety-something percent, and what had started as a pilot in one country turned into more than thirty countries competing to go live. This shows that exploring AI's business value does not require discarding what was built before: layering incremental value on top of standard processes is the path that works.
— Yuan Xin28% have tried AI, only 3% see real business returns
McKinsey interviewed more than 2,000 companies globally: 28% have already run AI projects, but only 3% believe there is genuine business return. So very few projects actually reach production; most are still at POC and prototype stage. Large customers have the money to absorb trial and error and move fast; small companies run fast on open-source tools; the ones struggling most are mid-sized companies in the few-hundred-million to roughly ten-billion range, where whether they embrace AI correlates strongly with the founder's own understanding — and Chinese companies are more willing to put money into the product side than the management side.
— Yuan XinGoing global works by borrowing SAP's localization base
SAP's globalization platform itself covers operations in more than 150 countries, and the product architecture includes a localization team dedicated to localizing tax compliance country by country; Chinese companies expanding abroad can borrow this base directly and only have to care about their own business. Lenovo has worked with SAP from the 20-billion scale all the way to several hundred billion, and now uses process mining tools to comb through more than 2,000 processes, looking for breakpoints and room for efficiency gains. When it acquires European companies, the target is naturally already an SAP user, so system integration carries a natural advantage too.
— Yuan XinIn their own words · checked verbatim
You can't escape handling an enterprise's processes, organization and data, you can't escape understanding the customer's business, you can't escape the question of how I turn business into my code — and it still needs a human in the loop to intervene.
你逃不出来处理企业的流程组织和数据，逃不掉对客户业务的理解，逃不掉我怎么把业务转化成我的代码，他还是要有human in the loop去做干预的。
Yuan Xin0:00
I don't think it doesn't exist. It's just that the pattern of history tells us spillover takes time.
我觉得不是不存在，只是历史的规律告诉我们，它外溢是需要时间的。
Yuan Xin30:51
The boss is always reading something on WeChat about what AI can do, and then he tosses it over and says I want that too — that is the single biggest headache right now, especially for CIOs at large enterprises.
老板经常在微信上读了一个什么什么什么AI能干啥，然后丢给他说我也要这是现在就是尤其大企业的CIO最头疼的事。
Yuan Xin44:47
Because after all, most of your business today still follows a set of rules. All you have to do, on top of a well-laid foundation, is fill in the parts that used to be less standardized, or that you couldn't read, or where collecting that kind of data was very expensive — and that's actually a very good state to be in.
因为毕竟你现在的业务的大部分，其实还是遵循一定之规的。 你只需要在打底打的好的基础上，把那些以前不太规范。 和你不能读懂的或者收集那样的数据代价很大的那一部分补充上，其实就是很好的一个状态。
Yuan Xin56:30
Right now the whole AI world is really one big co-opetition free-for-all. Right — nobody can do without anybody else, but at the same time everyone wants to fence off a piece of land of their own.
现在大家其实整个AI的世界就是一个大的竞和的世界大混战啊。 对，就是谁也离不开谁，但同时大家都想去圈出来自己的一块地。
Yuan Xin58:40
For enterprises it's a relatively complicated thing: once you're out there in the game, sooner or later you have to pay it back.
对企业其实相对来讲还是比较复杂的一个东西，就是出来混，总是要还的。
Yuan Xin1:11:28
Figures
| Companies in the global top 100 that use SAP | 99 | 3:15 |
| SAP cloud users | more than 300 million | 3:15 |
| SAP employees worldwide | 110,000 | 3:15 |
| Countries SAP's business covers | more than 150 | 3:15 |
| Share of SAP customers that are small and mid-sized businesses | 80% | 27:28 |
| Customs agent accuracy | from 60-70% up to ninety-something percent | 56:30 |
| Companies that have already run AI projects | 28% | 1:06:03 |
| Companies that believe there is genuine business return | 3% | 1:06:03 |
| SAP lines of code | several hundred million | 18:03 |
| Lenovo processes | more than 2,000 | 1:24:03 |
Glossary
- ERP / Enterprise Resource Planning
- The core system SAP was built on, managing end-to-end enterprise processes such as finance, procurement and production.
- FDE / Forward Deployed Engineer
- Goes on site at the customer to get AI to land, and settles the results back into the product.
- Human in the loop
- Even after AI executes automatically, people still have to review the critical judgments.
- Process mining
- Using logs to analyze how the business actually flows, and to find breakpoints and room for efficiency gains.
- Physical AI
- AI combined with physical things like robots and electric vehicles, seen as a direction where China has the edge.
How to listen
Enterprise-software founders, investors, technology leaders at companies expanding abroad, and engineers trying to work out whether FDE is a good career direction.