Original · Unique Research · 2026-03-23 · Shanghai
Editor's note: This is the complete English rendition of Unique Research's March 23, 2026 interview-based article. First-person judgments belong to the original author; commercial and technical assertions are attributed to that source and Liao Can, not independently verified findings. YUHE.AI is the spelling supplied in the separate October 31, 2025 Chinese interview; Liao Can and Fanruan are romanized renderings. The source does not specify the currency of the annual pay of 120,000 or revenue exceeding ten million, nor the measurement period and customer cohort behind the 100% renewal rate. The quoted 4-day-to-20-minute workflow and 20% customer-value increase should not be combined with differing shipping-workflow figures in the earlier interview: an identical case, scope, or before-and-after progression has not been established. The digital-employee and salary language is a commercial analogy; the article itself identifies the product as software. Its descriptions of memory, permissions and accountability are not an independent audit of privacy, compliance or reliability.
Unique Research · Guest Interview
Why Do More and More Enterprises No Longer Want to Buy AI Tools?
Liao Can of YUHE.AI: What enterprises really lack is not smarter software, but a digital employee that can commit to KPIs, deliver results, and understand the business better the longer it works.
Interviewee: Liao Can (Marketing Partner, YUHE.AI)
Interview date: March 2026
Over the past two years, almost every enterprise has been talking about AI.
Some have bought knowledge bases; some have connected large models; some have deployed assistants; some have even run internal agent pilots. On the surface, everyone seems to be moving forward. But if you really ask business owners and business-unit leaders—especially a sales VP, a presales manager, or an IT leader in manufacturing—you discover a subtle reality:
They do not lack "tools that know a little AI."
What they really lack is a role that can enter their workflows, take responsibility for results, and be accountable to the business.
That was also my strongest impression after speaking with Liao Can, marketing partner at YUHE.AI.
He is not someone who begins by talking about model parameters, agent architecture, or an Agent workflow. On the contrary, his judgments almost always come from the business side. He keeps returning to one question:
"Can this thing actually do real work for an enterprise?"
If not, it is just a tool that looks new.
If it can, it is more than software: it is closer to a "digital employee."
And that is exactly what YUHE.AI is building.
Why Did Someone from Fanruan Ultimately Bet on Agents?
Liao Can's career path is both typical and atypical.
It is typical in that he has fought his way through the ToB enterprise market throughout his career: from Fanruan to RPA+AI at a startup, then enterprise ecosystems at a major company, and finally into agents at YUHE.AI.
It is atypical because he has experienced not just several jobs, but four shifts in enterprise digitalization: from BI and big data to traditional AI tools, then SaaS ecosystems, and finally AI agents.
Anyone familiar with the evolution of China's ToB software industry will recognize how representative this path is. Many people start by "selling systems," but relatively few reach the point of "selling results."
Liao Can says Fanruan's greatest influence on him can be expressed in two Chinese characters: pragmatism.
The company is what many ToB practitioners call a "Whampoa Military Academy"—a training ground for the industry. It does not rely on flashy posturing or conceptual packaging. Internally, performance speaks; externally, product results speak. After customers buy the software, can they actually use it, and can it survive in their business? That is the most important test.
This way of thinking later became the foundation of how he assesses AI.
So when he builds agents today, he spends less time telling customers "how advanced large models are" and more time asking:
"Can a digital employee reduce work that currently takes someone in this role 4 days to 20 minutes?"
He Did Not Turn to Agents out of "Technological Excitement," but Because the Old Solutions Had Hit a Wall
Many people enter AI because the technology inspires them.
Not Liao Can.
He seems to have been pushed here, step by step, by reality.
After leaving Fanruan, he joined a Series B startup to lead commercialization for its SaaS division, focused on RPA+AI. The team progressed from Series B to Series C, built a GTM system from 0 to 1, developed more than 300 standardized application scenarios, served over a hundred clients, and reported 10-fold year-on-year business growth.
Judged by the numbers alone, those were already very respectable results.
But he later discovered a fundamental problem with that direction:
"
RPA had evolved hands, feet, and eyes. It had no brain.
It could execute actions faster and move processes along more efficiently. But what enterprises really lacked was not merely a faster operator: it was a role that could understand context, exercise judgment, and participate in complex business.
When ChatGPT arrived, he immediately realized that the missing piece had finally been supplied.
So his judgment was not "AI is hot; I'd like to try it," but rather:
The underlying capabilities had matured. The enterprise pain points had always existed; previously, no one had been able to make this work for real.
Over the past few years, enterprises have bought too many tools: knowledge bases, process systems, RPA, BI, SaaS platforms, collaboration systems. Almost every part of the business has been turned into a tool.
But what enterprises actually feel is that systems keep multiplying while people keep getting more exhausted.
Why?
Because most of these tools address "individual actions" without taking on "a complete role."
YUHE.AI's judgment is different:
Enterprises do not need one more tool. They need a digital employee that can genuinely start work.
The Hardest Part Is Not Selling a "Digital Employee," but Convincing Customers It Can Deliver Results
Moving from "selling software" to "selling digital employees" may look like a change in sales language. In reality, the hardest thing to change is customer expectations.
Liao Can spoke very practically about this.
In the past, when you sold software, customers bought features. Once you delivered it, provided training, and put it into production, whether it was used well was often the customer's own problem.
Digital employees are different.
Customers are buying results.
If you say it can shorten my quotation cycle, the cycle must actually become shorter. If you say it can let 5 people handle a workload that used to require 10, they must actually be able to handle it. If you say it can improve presales efficiency, efficiency must improve.
So today, their customer conversations are no longer about "what features we have." They go straight to breaking down a role:
• What does this role do every day?
• Which actions can a digital employee take over?
• How much improvement is expected after it takes over?
• Who ultimately reviews the work, makes the final decision, and takes responsibility for the result?
As a result, the most common customer response is not "I'm interested," but one of two very typical forms of skepticism.
One is: "Isn't it just more of the same?"
Customers have used AI to help write emails or translate documents before, so they naturally put you in the same category of "more or less the same thing."
The other is: "What if something goes wrong?"
Especially in quotations, tender documents, and technical proposals, a single incorrect parameter can invalidate a bid. At that point, customers care less about whether AI is dazzling than whether it is reliable.
YUHE.AI's response is equally straightforward: skip abstract concepts and go directly to a demo.
He is right that, for many customers, one demo is often enough to move from skepticism to "this is actually great."
What matters is not how many technical terms you use, but whether customers have seen for themselves:
Their own business data actually producing a result.
What Makes Manufacturing Customers Pay Is Not "AI Is Powerful," but "You Can Finally Get This Done"
If you look at many AI startups today, a common problem emerges: what they build is often smart, but does not address a sufficiently painful problem.
Manufacturing is the opposite.
There are too many scenarios in this industry where the need is not to "optimize things a little." If the problem is not solved, it directly holds the business back.
Presales, quotations, tendering, and proposal responses are examples.
That is the kind of scenario YUHE.AI worked on at COSCO SHIPPING Group.
Initially, the customer's technical lead did not believe it would work. The business lead, however, was anxious because responding to large numbers of inquiries was a daily headache. At first, he did not even dare set the bar very high. He said:
"
Don't worry about automatically producing a complete response document yet. Just helping me classify and structure the inquiry materials would already be very good.
But once it was actually running, the results went far beyond what they had initially imagined:
📉 The quotation cycle fell from 4 days to 20 minutes.
📈 Average revenue per customer also rose by 20%.
Why could prices rise as well?
Because the digital employee was not just faster. It could match products more accurately, design more suitable solutions, and even recommend a more appropriate maintenance plan alongside a customer's inquiry.
At that point, it was no longer just an efficiency tool. It was beginning to affect revenue growth.
"Base Salary Plus Commission" Is Not a Business-Model Gimmick; It Changes the Language of Pricing
What I find most worth writing about YUHE.AI is not simply that it builds digital employees, but that it has thought through "how to sell AI."
In the SaaS world, pricing has traditionally taken only a few forms: by feature, by seat, or by usage.
Fundamentally, all of them sell tools.
Digital employees are different.
When a customer faces an "AI employee that has committed to KPIs," the logic in their head is not software procurement, but hiring.
Those two ways of thinking are very different.
Tell a customer, "This is how much our AI platform costs per year," and an automatic thought appears: another system.
But tell them, "I'll assign you a digital employee with an annual salary of 120,000, roughly 40% of a human presales employee's pay. It is online 24 hours a day, takes no leave, never changes jobs, and becomes more familiar with the work over time."
Now the customer is calculating workforce productivity, not an IT budget.
This is also why Liao Can says "base salary plus commission" was not a sudden idea dreamed up in an office. It was gradually refined through customer feedback.
At its core, it does more than change the charging method. It changes AI's language from "buying software" to "hiring an employee."
That step is crucial.
Chinese enterprises are often cautious and hesitant about paying for software, and inclined to push prices down. But for "someone who can deliver results for me," they can more readily decide whether the cost is worthwhile.
What Enterprises Really Lack Has Never Been a "Smarter Knowledge Base"
Any discussion of enterprise agents inevitably reaches "knowledge" and "memory."
But Liao Can made a point here that I strongly agree with:
"
Most solutions on the market are, fundamentally, just smarter search.
You feed in documents, retrieve information, recall relevant material, and generate answers. This certainly has value. But if you look at the most valuable knowledge in an enterprise, you find that:
At least half of it is not in documents.
It lives in experienced workers' heads, in presales staff's judgments about customers, in sales directors' handling of pricing strategy, and in accumulated experience about "why we did it that way last time."
So YUHE.AI does not require enterprises to document all their knowledge first. It lets digital employees accumulate it themselves in the course of their work.
The team proposes a three-layer memory system:
Layer one: a knowledge base containing product manuals, industry standards, and specifications.
Layer two: a business state machine recording project follow-up, customer profiles, and decision history.
Layer three: role-specific experience, where the digital employee's daily "work diary" gradually builds a collection of pitfalls and best practices.
Behind this design is an important judgment:
Enterprises do not want AI that knows a lot of textbook material. They want AI that increasingly resembles an experienced employee as it works.
Traditional RAG is more like a library. What they want to build is someone who has spent three years in the role.
Behind a 100% Renewal Rate Is More Than Good Service: "Replacing It Means Losing an Experienced Employee"
YUHE.AI now reports revenue exceeding ten million in one year of commercialization and a customer renewal rate of 100%.
Many people's first reaction to that number may be: the company does customer success well.
Liao Can's answer was clear-eyed.
In the short term, of course it relies on customer success. Dedicated follow-up, regular reviews, and rapid responses all matter.
But in the long term, what really supports renewals is product stickiness.
The greatest difference between a digital employee and ordinary software is that the longer it works at a customer's company, the more role-specific experience it accumulates, and the more it becomes part of that organization.
Replacing ordinary software is, at most, a system migration. Replacing a digital employee that has worked for three years means losing three years of accumulated business memory.
That is why he says the cost of replacing a digital employee is not "changing a system," but "losing an experienced employee who has spent three years on the job."
The Real Divide Between Enterprise and Personal Agents Is Not Intelligence, but "Organizational Awareness"
Many people are building agents now.
But the vast majority still operate from an individual's perspective: I give you a task, and you help me complete it.
This logic works in personal scenarios and can even feel wonderful. In an enterprise, however, it quickly hits a wall.
An enterprise is not a "task-completion system." It is a "system of competing objectives."
Liao Can offered a formulation that I think serves particularly well as a dividing line for enterprise agents:
Enterprise agents must have three built-in calibrators.
First, ROI calibration: ask whether something is worth doing before asking whether it can be done.
Second, permission calibration: not just "who can see this," but "should this be done?"
Third, strategic calibration: short-term KPIs must not destroy the long-term business.
I Increasingly Believe the Next Generation of Enterprise Software Will Sell "Role Outcomes," Not Features
After this interview, one line kept running through my mind:
"AI will not swallow all software first. It will first swallow the work that can be defined in terms of roles."
What does that mean?
It means the most important next change in enterprise services may not be "building an even smarter platform," but gradually shifting from selling modules, systems, and features to selling roles, results, and productivity.
In the past, enterprises bought software as a framework of capabilities. In the future, when they buy AI, they may be more likely to buy a specific role.
Not "give me a collaboration platform," but "give me a digital employee that can do presales."
Not "give me a knowledge-management system," but "give me role-based AI that can remember experience, participate in decisions, and keep evolving."
Not "give me a toolbox," but "give me someone who can take responsibility for KPIs."
If that judgment holds, many of the enterprise agents we see today are only at the beginning.
The real dividing line will become very clear:
Some companies are still selling "tools that know a little AI."
Others have already begun selling "digital employees that can work, grow, and take responsibility for results."
The latter is plainly harder to build. But it also looks more like the next generation of genuinely valuable enterprise services.
Selected Q&A
Q1: Is YUHE.AI actually building a tool or an employee?
In terms of product form, it is of course still a software platform. But in terms of how customers buy it and how it delivers value, it is more like a digital employee. Enterprises buy it not to gain another system entry point, but to have it genuinely take over part of a role's workload and take responsibility for the results.
Q2: Why does "base salary plus commission" resonate more readily with customers?
Because it changes the pricing language of AI from "buying software" to "hiring an employee." Enterprises naturally calculate workforce productivity, salary, and output, rather than treating it as yet another IT procurement project.
Q3: What is the biggest difference between an enterprise agent and a personal agent?
A personal agent is responsible for "task completion"; an enterprise agent must be responsible for "organizational value." It cannot merely execute. It must also understand permission boundaries, the division of roles, ROI constraints, and long-term strategy. This "organizational awareness" is the real threshold for enterprise agents.
Q4: Why can YUHE.AI achieve a 100% renewal rate?
In the short term, through customer success; in the long term, through product stickiness. The longer a digital employee works at a customer's company, the better it understands that enterprise's business and the harder it becomes to replace. For the customer, removing it means more than changing software: it means losing an "experienced employee" that already understands the business and has accumulated experience.
Q5: Will AI digital employees ultimately replace people or augment them?
The question itself is a little dated. A more accurate formulation might be that AI is activating new organizational forms. It is neither simple replacement nor merely light assistance, but a force driving the accumulation of experience, the reconstruction of processes, and the upgrading of organizational capabilities. How work gets done has changed; the result is not that one side becomes stronger. When an organization becomes accustomed to this "symbiotic" model of human and digital employees, it gains the possibility of continuing to evolve in the AI era.
What enterprises really lack has never been "smarter software," but a digital employee that can commit to KPIs, deliver results, and understand the business better the longer it works.