Original · Unique Research · 2026-04-11
Editor's note: This complete text translation preserves the original author's analysis and all supplied speaker turns. Headcount changes, contract figures, customer counts, conversion improvements and renewal effects are source-reported claims, not independently audited results. The claim that an audio recording captures only 10% of communication is reproduced as the speaker's view; no supporting study is cited. The source refers both to 2017–18 and to “five or six years,” an elapsed-time inconsistency retained here. Its narrative attributes the sales-training salary remark to Wang, while the transcript assigns it to moderator Duan; both passages are preserved. Currency is not explicitly stated for the contract, BI-quote and annual salary figures. English names follow the source and may not be official brand renderings.
Unique Research
AI Arrived—and This Export SaaS Company's Sales Team Grew
AI is not simply cutting out sales. At different stages, it is rewriting the division of labor between salespeople and customers.
After rolling out AI comprehensively, a company providing SaaS for export businesses increased, rather than reduced, its sales headcount.
Chen Erhang, senior vice president of Yixuan Technology, shared this story. He added: “Because we saw more opportunities. We could build a bigger business.”
After all the time spent debating whether AI will eliminate sales, someone with 16 years in export services tells you that he hired more people after AI arrived.
At a Hangzhou AI WEEK roundtable on AI sales, four guests represented four very different fields: CRM, AI for consumer brands, international expansion for exporters, and sales training. The moderator opened with a question:
In the end state, can AI completely replace sales?
A Veteran of Traditional CRM Says “It Will”; an AI Founder Says “It Won't”
The question immediately divided the room into camps.
Wang Zhaoqi, co-founder of Xiaobangbang, has spent more than a decade in CRM. His answer was: “It will be hard to replace sales in the short to medium term. But from the perspective of the end state, it will inevitably eliminate all salespeople.”
His reasoning was simple: He had seen too many small and medium-sized businesses with poor sales execution. AI would replace all those inefficient stages; the rest was only a matter of time.
LynxAI founder Zhang Keyi gave the opposite answer: “It won't replace people. It will necessarily be coexistence and collaboration.” His logic was equally clear: AI can scale intellectual work, but judgment-based work always needs people. Consumer brands ultimately serve human beings. As long as carbon-based life forms make purchasing decisions, human experts' judgment cannot be replaced.
The moderator immediately polled the audience: Those who believed AI could completely replace sales in the end state were asked to raise their hands. Only a minority thought it would.
I Think the Question Itself Misses the Point
Arguing over whether AI “will or won't replace people” is a false framing.
Chen Erhang of Yixuan Technology put it more practically: “If you break a sales task down into its individual stages, AI plays a different role in each.”
They have worked on exporters' independent websites for 16 years and know the entire sales process well. As he described it, the process roughly divides into three stages:
Stage One: Find a few hundred potential leads among millions or tens of millions of users. AI does this. That is already settled; manual work simply cannot compete.
Stage Two: Identify a few dozen prospects with genuine purchasing intent among those hundreds of leads. AI has taken over half of this work, especially by resolving time-zone and language issues. “If a customer came while you were asleep at night, previously you would definitely lose them. That no longer happens.”
Stage Three: Turn interested prospects into actual deals. People still lead this stage, but AI can suggest approaches—for example, offering negotiation advice for customers from different cultural backgrounds.
AI plays a different role in each of these three stages. The issue is not “replacement vs no replacement.” It is “which stage AI handles, and how deeply.”
Li Shouguo, CEO of Beta Data, added a precise observation: “Ultimately, in areas where customers need service, solutions, and a connection of trust, things are relatively difficult for AI.”
AI certainly can handle simple sales. In complex sales, the core is not the product itself, but trust. At present, no model can quantify trust.
So What Can AI Actually Do in Sales?
Wang Zhaoqi shared Xiaobangbang's work over the past year or so. I found this part the most valuable.
He is building an end-to-end system—from customer acquisition and prospecting to lead cleansing, sales role-play training, and BI reports, connecting the entire chain.
The results are real: AI commercialization began last October, contracts totaling 1 million were signed within two months, and more than 100 customers now use it in practice.
One of the most important elements is how he handles the closing stage.
Anyone who has worked in sales knows that the hardest part is not finding a customer. It is knowing what to do after the customer says, “I'll think about it.”
Their approach is to use AI to monitor the entire sales follow-up process through a traffic-light dashboard. Whenever an abnormal signal appears as a customer is being followed up, a red light immediately prompts management intervention. Once management intervenes, the system automatically uses all of that customer's historical records—including audio recordings and videoconference data—to generate a three- to five-minute practical role-play scenario. Salespeople can open the CRM and start practicing.
“Integrating business operations and training completely within a single system.”
And training costs have come straight down. He mentioned that sales-training leaders at medium-sized and larger SaaS companies in China generally have annual pay above one million. AI role-play training eliminates a substantial part of that cost.
What Does It Mean to “Get AI Up to Speed”?
Zhang Keyi used this phrase, which I found interesting.
Many companies say they are making their “organization AI-enabled,” but in reality they are only making individuals AI-enabled. Each employee uses tools independently, while nothing fundamentally changes at the organizational level.
His view is that individual AI adoption means AI empowers human execution. Organizational AI adoption means people empower AI—helping it progress from not understanding to understanding their industry, professional domain, and business leader's preferences.
There are three layers to getting AI up to speed: alignment with the industry, the professional domain, and the business leader's preferences.
LynxAI has no dedicated sales roles: “Whoever sells also delivers.” During a client engagement, the delivery team first develops a deep shared understanding with the customer, then embeds that understanding in the system—in structured databases, knowledge bases, and AI-supported business processes. Customers become “spoiled” at the execution level but demand more at the architectural level. That is how the positive flywheel begins to turn.
Put plainly, renewals do not depend on sales follow-up. They depend on customers discovering, through use, that they cannot do without the product.
Two Real Breaks in the Chain
At Beta Data, Li Shouguo builds AI sales role-play tools for large-business markets such as finance, telecommunications, energy, and machinery. His description of the gaps was measured.
The First Gap: Data
A great deal of offline sales data is never collected at all. When people speak face to face, words and sound account for only a very small share of the interaction. Research over the past thirty or forty years has shown that eye contact, facial expressions, and gestures carry much of the content. “Even if you record the entire conversation with an audio device, its weighting is only 10%.” If that is all the information AI receives, how effective can it be?
The Second Gap: Organization
Today's large organizations still divide work along linear, assembly-line structures. Data systems are fragmented across departments. Sometimes, when connecting data between two teams, “they look like two separate companies.” That is before considering compliance issues.
His conclusion, therefore, was not that AI sales cannot work, but that “there is still a very long way to go.” It is a question of choosing the pace.
People often overestimate what changes in one year and underestimate what changes in five.
Use Cases Matter More Than Technology—but They Are the Hardest Things to Find
All four speakers were highly aligned on this question: Start with the use case.
Wang Zhaoqi offered a down-to-earth example: “You hire a salesperson for three thousand yuan a month, then buy traffic on Douyin. Each lead costs more than 200 yuan. Give that person 15 leads and you have already exceeded their monthly salary. Buy 150 leads and it is ten times their salary. Do you use a person or AI?”
This is not a purely technical question. It is a combined decision about ROI and the use case.
Zhang Keyi said his team spent last year heads-down building its product, only to be “torn apart” by customers after launch. They initially thought customers did not understand AI, then realized that they themselves did not understand the market. Only after adding hands-on support from human experts did they truly understand what customers needed AI to do in specific situations.
Chen Erhang put it more directly: “An established company must immediately identify use cases and implementation approaches. A startup may have a technical highlight while its business model and practical use cases remain relatively unclear.”
Finally, Li Shouguo added a small footnote: There is one exception among business customers—IT procurement departments, which sometimes buy into a “technology parade,” largely following the crowd. Apart from that group, use cases generally have the final say.
More from the Conversation
Unique Awards · Hangzhou AI WEEK Trend Roundtable Panel
“Marketing and Sales: AI-Driven Customer Experience Innovation and Conversion Growth”
Guests:
Wang Zhaoqi, Co-founder and COO of Xiaobangbang
Zhang Keyi, Founder and CEO of LynxAI
Chen Erhang, Senior Vice President of Yixuan Technology
Li Shouguo, CEO of Beta Data
Moderator: Duan Hongyu, Partner at Unique Research
Duan Hongyu: First, I would like each of our four guests, in this order, to take about a minute to introduce themselves. Please also tell us whether, in AI's end state, it really can replace the entire sales process, or what balance of collaboration between people and AI will ultimately emerge in sales. Mr. Wang, please begin.
Wang Zhaoqi: I'm Wang Zhaoqi from Xiaobangbang. Xiaobangbang makes CRM software. Whenever I meet people, they ask whether this business is still viable. The reality is that AI has been quite helpful for traditional SaaS software. Since the beginning of this year, we have clearly felt that the SaaS/CRM business is becoming easier. Returning to the moderator's question: Will AI ultimately replace sales? In the short to medium term, that will be difficult. But I noticed the moderator used the words “end state.” If we insist on looking from that perspective, AI will inevitably eliminate all salespeople. Reaching that end state, however, may take a very long time.
Zhang Keyi: I'm Zhang Keyi, founder of LynxAI. Our product is a product-innovation platform. Specifically, we provide an AI product together with hands-on expert support, because the customers we serve are consumer brands in a business-to-business relationship. On the moderator's question, my view differs somewhat from Mr. Wang's: I do not think people will be replaced. Our thinking is that AI can indeed progressively develop in intellectual work or tasks. But in judgment-based work, AI and people in different circumstances will always make different choices. It is therefore not a replacement relationship. It must be one of coexistence and collaboration, although that relationship is being reconfigured. My position is clear: People will not be replaced. What the connection between people and AI will look like, and what value the connections created by sales will deliver from process to outcome, are still developing. That is why LynxAI's solution includes expert support alongside the AI product-interaction solution.
Duan Hongyu: So Mr. Wang, who works in traditional SaaS, believes sales can be replaced, while Mr. Zhang, who sells outcomes and services, believes it cannot. Let's ask the audience: If you think AI can completely replace sales in its end state, please raise your hand. Now raise your hand if you think it cannot. At the end of this discussion, we will see what everyone thinks again. Let's have a live debate. Mr. Chen from Yixuan Technology, please share your view.
Chen Erhang: I'm very pleased to have this opportunity to talk with everyone. I'm Chen Erhang from Yixuan Technology. We help Chinese export businesses expand internationally. We focus on the ecosystem of exporters' independent websites, providing website construction, operations, and global marketing services to help their brands reach overseas markets. This has always been our focus, and we have worked deeply in it for 16 years. Returning to the topic, in our field it is hard to give a simple answer of replacement or no replacement. There is a development process. If you break a sales task down into stages or subtasks, AI plays different roles in each. In some stages or tasks, we can already see that AI has almost entirely replaced people. In others, such as tasks requiring deep interaction or an emotional connection with customers, people may still lead. I therefore see a complementary relationship. But the overall trend is that AI's role and value will grow. The key is how we use it to realize that value.
Li Shouguo: Hello, everyone. I'm Li Shouguo from Beta Data. We are building a tool for practicing sales skills. Like swimming or cycling, sales skills require repeated practice. That process involves simple, repetitive work and is very tedious. We therefore built a tool in which AI simulates a customer and the salesperson practices against that customer. We currently focus on large-business sales in finance, telecommunications, energy, and machinery. To answer whether AI can replace people: As AI progresses, it will certainly take over some areas of sales, especially simple sales of simple products. In complex sales, unless we see AGI arrive, I think replacement remains relatively difficult. Fundamentally, though, whether AI will replace sales entirely is not determined simply by whether the sale is simple or complex. Why do we perform relatively well in finance, communications, energy, and machinery? Because customers do not merely need a simple product. They need solutions, emotional communication, and the building of trust—things AI can never replace. So my answer is that AI will handle some areas and stages, but ultimately it faces real difficulties where customers need service, solutions, and a connection of trust.
Duan Hongyu: Can I understand that to mean AI can replace people completely in relatively simple delivery scenarios, particularly those with low order values, while more complex, difficult delivery—especially for key-account, or KA, customers—may still require human involvement?
Li Shouguo: Broadly speaking, yes.
Duan Hongyu: I would also like to ask all four of you how you view using OpenClaw for sales today. Many tools on the market address things like finding influencers and customers across borders or domestic business-to-business sales. Many business owners find that AI considerably improves customer-acquisition efficiency but does not perform as well in conversion. One reason may be that it does not provide enough emotional value. What do you think, and do you have your own solutions? Mr. Wang, please begin.
Wang Zhaoqi: First, I have not seen a genuinely implemented sales-management use case or company using OpenClaw. We have seen AI applied in specific settings such as finding distribution partners or working with influencers on distribution. We also see simple sales tasks being replaced—for example, customer service, lead cleansing, responsive interactions, and many roles in e-commerce. But I have not seen this in the form of OpenClaw. More broadly, what I see is that the vast majority of China's small and medium-sized businesses still have extremely poor sales execution. The opportunity I see is for AI to drive execution. There is enormous room for growth there.
Zhang Keyi: Let me start with our behavior. Our team has fully embraced it, even making it mandatory. However, OpenClaw's permissions and outputs must strictly undergo human inspection and review. As mentioned earlier, our delivered outcomes are used in business-to-business applications, so rigor and stability must be assured. Another point worth sharing is how we build team workflows involving human–AI interaction. Looking at traditional digital enterprise architecture, using a large technology company's framework, we separate the underlying technology architecture from the data architecture, application architecture, and business architecture above it. We treat OpenClaw similarly. Our human experts also work from that structure, which lets us compare their respective strengths and weaknesses. Intellectual work may scale reasonably well through OpenClaw and genuinely become more efficient. But complex decision-making in product and brand innovation necessarily combines intellectual work with judgment-based work. We use two standards when designing how customers interact with our AI product. First, is feedback more efficient? That means greater accuracy in deciding what should change and what should remain. Second, iterations must happen more frequently. OpenClaw makes a major difference here. But it does not change the fundamental architectural component: human experts. One interesting constant in consumer goods is the end consumer—each of us, a carbon-based life form, consuming in daily life over 24 hours every day. That is the source of value exchange and creation, and it does not change. So the underlying architecture, in which human experts make final judgments and work with AI on overall product and brand innovation, never changes.
Duan Hongyu: Let me follow up on that. If OpenClaw or expert-level skills become sufficiently developed, do we still need experts? For example, suppose I am an experienced salesperson selling alcohol and want to move to a new-energy company. I have no experience in that industry, but the agent has the intellectual capabilities. I have sales experience: I know how to acquire and maintain customers and how to judge whether something should be said to a customer. Might there be a future scenario in which experts no longer exist—business-domain experts no longer exist?
Zhang Keyi: Our team has discussed this, but our consensus is clear: No. Human experts do give AI feedback, so AI may continue advancing in judgment-based work. But return to the specific consumption scenario: In the setting where value is consumed, it is a person doing the consuming. It is always people deciding what kind of life they want, which lifestyles or experiences they want to preserve, and which they want to change. That is the trigger and driving force. My next phrase may not be entirely appropriate, but consider the feeling behind it: “Those who are not of our kind must think differently.” If one day AI acquires agency in rights, responsibilities, and morality, it may become one component of the end-consumer population, and things could change then. But while we serve end consumers—individual carbon-based life forms—the judgment, choices, and feedback of human experts are necessarily irreplaceable. AI can only follow along. The structural difference in efficiency and effectiveness between people plus AI and people alone, however, is very significant. That is clear.
Chen Erhang: We can divide the waves of recent years into several stages. The latest wave is, of course, large models. Before that there may have been AIGC, and before that PGC and so on, all of which had a substantial impact. Each wave has also affected our company internally. We pay very close attention; in the internet industry, we must keep watching closely. Whenever something new emerges, we become anxious about how it relates to us and how we can use the technology to advance our business, because standing still means falling behind. We have an internal principle: Start in the back office. The R&D team first connects with the most popular current AI technology. Next comes the midfield—operations—to see whether it can improve output per employee and reduce costs. Only after that comes the front line of sales and marketing, where the proportion of work involving people is greatest. Every wave makes the R&D team busiest: They must quickly determine how the new technology can improve our products and services and what value it can bring customers. That should be a company's central task. The second is reducing costs and improving efficiency internally.
Duan Hongyu: Understood. Do the solutions you currently provide customers include AI-enabled approaches for their business-to-business sales?
Chen Erhang: We are already quite deeply involved. We help export businesses acquire customers globally. Around 2017–18, at the very front of global customer acquisition—obtaining sales leads, or Leads—the work was already almost entirely automated by systems. People had essentially withdrawn from that stage. That has been the case for five or six years. The second stage comes after Leads return, when customers conduct an initial round of screening and interaction. People used to do that as well. Over the last two years, we have used the latest AI technology to help customers address this step. At present, the split may be roughly half and half, with AI handling some lead screening and simple interactions. Those interactions are not very demanding, and AI can already perform them, improving efficiency and lowering costs substantially. Finally, when it comes to deep interaction and discussing business—establishing strong intent or even making a cooperation decision—customers still do the work themselves. But AI keeps moving further along the sales chain from the front toward the back, playing a larger role.
Duan Hongyu: How many salespeople does your company have? Has its organizational structure changed over the past two years?
Chen Erhang: Previously, we had fewer salespeople than back-end R&D staff. But after AI arrived, our sales headcount actually increased. Why? That seems contrary to the usual expectation. We saw more opportunities and the possibility of building a bigger business, so we expanded the teams reaching markets and customers. This is a new opportunity AI has brought us, and it has changed our internal staffing structure.
Li Shouguo: There are already some quite good examples in content marketing and SDR. If implementation still falls short, one reason is that it has only been three or four months—a very short time—and much of the engineering is not yet smooth or solid. Another is that many gaps remain, and bridging them takes time. The first is a data gap. Much offline sales data simply is not collected. In face-to-face communication, even if you add an audio recorder, language and sound account for a very small share of the interaction. Theory over the past thirty or forty years has shown this: Eye contact, expressions, and gestures are much more important parts of communication. Even if you record the entire conversation between a salesperson and a customer, its weighting is only 10%. A perfect recording is still only 10%. The second major gap, I believe, is organizational. Today's organizations are not AI organizations. Most divide work along linear, assembly-line structures. Their data systems—especially among the state-owned enterprises we serve—can make two departments, or even two teams within one department, look like two separate companies. Sometimes we even have to mediate their disagreements to connect data structures and APIs. Imagine how difficult that is, before even considering compliance and related issues. Consequently, AI receives very, very little information. It is not that it cannot work, but that there is still a very long way to go.
Duan Hongyu: Mr. Li has raised the question of the entire process. Among AI sales tools and AI companies offering sales solutions, we see two types: Some cover the entire AI sales process, while others focus on a single stage, such as screening and cleansing prospects at the SDR stage. Do you think it is better now to cover the entire process, or would AI for a single stage be better and easier to implement? Mr. Wang?
Wang Zhaoqi: I think I can speak with some experience, because I have been doing this for more than a year. I cover the entire process, and the gains have been substantial. Originally, I only handled the small, well-defined CRM segment of L2C. After AI arrived, I spent all of last year collecting and cleansing data from most of the domestic data-service providers. I then found that AI plus data had matured enough to cover acquisition, prospecting, and lead cleansing. Adding my L2C lead-handling process completed that part. Then I connected a Doubao role-play training module, which had a low integration barrier. Because we had complete process data, we could quickly generate a training scenario for a particular customer. That extended the process into role-play training. In other words, I covered the entire chain from customer acquisition to sales conversion to practice. A customer I met a couple of days ago wanted us to handle BI as well. I looked at the business team's original BI quote: 360,000. After reading it, I said I did not even want to do the job. Even if the customer accepted 360,000, I would not want it, because it would take more than 300 person-days. I said those people would be better deployed in AI R&D. So we explored whether AI could generate the BI analysis. Once connected to our data, the AI-generated BI reports were beautiful—even better-looking than the reports we originally created through coding. This was an entirely new presentation method. I am therefore exploring whether adding a BI component could make the system more powerful. That is the end-to-end perspective. The results have been particularly good. I began commercializing AI last October, signed contracts totaling 1 million within two months, and now have more than 100 customers using it in practice. Those are the results of my own work, offered for reference. There is no absolute distinction between good and bad approaches.
Duan Hongyu: Let me ask another question, Mr. Wang. Anyone who has worked in sales knows that closing is the hardest part. You have discussed the solution, the customer says they are satisfied, and when you ask whether they will buy, they say, “I'll think about it.” In your end-to-end system, how do you address this crucial closing stage? Does AI contribute here?
Wang Zhaoqi: Of course it does. Good and poor salespeople can differ by more than 10 times in overall conversion rates. Mr. Li from Beta Data may have more to say on that. We use AI judgments both in early customer analysis and in warnings during the sales process, with a traffic-light dashboard. If anything abnormal happens at any point during a follow-up process lasting as long as half a year, a red alert calls for management intervention. Second, once management intervenes, it needs a practical means of acting. This is where role-play training comes in to improve sales skills. In Xiaobangbang's CRM, open the customer details and click the simulation-training option in the top-right corner. Using all of that customer's historical follow-up records, audio recordings, and videoconference data, it generates a practical role-play session of roughly three to five minutes. I see this as a complete integration of operations and training, with early-warning work handled in the same system. From the salesperson's perspective, the experience changes from actively searching to passively receiving. After finishing a day's work, they receive a report the next day. It appears in several places: on our APP's homepage and in DingTalk groups, delivered in a way similar to OpenClaw.
Duan Hongyu: So it sounds as though the system uses AI sales capabilities to address this problem.
Wang Zhaoqi: I have an almost faith-like belief that getting management right produces threefold growth. I have helped countless business customers achieve that, so AI should first address the management component.
Duan Hongyu: Yes, that is a valuable capability. As I understand it, sales-training leaders at medium-sized and larger SaaS companies in China almost never have annual pay below one million. If Xiaobangbang can solve the sales-training problem, that capability is very valuable. Mr. Zhang, please also respond to the earlier question.
Zhang Keyi: I strongly agree with Mr. Wang. End-to-end work certainly offers greater value, which is consistent with my earlier point. We look more from the perspective of enterprise architecture: People set rules and design the architecture. A vertical segment and the entire process are relative concepts. At the extreme of vertical specialization, people define the standards or rules, and execution is triggered accordingly. We believe AI is better suited the more specialized the domain becomes. But the entire process faces systemic internal and external changes. That uncertainty requires taking a holistic view to build architecture, dynamically update rules, and assess whether those rules remain effective. It is in this process that people begin to enjoy a harmonious, positive-feedback relationship with AI. Through this combination, we gradually get up to speed. Our team uses the phrase “getting AI up to speed.” Customers often say they are making their “organization AI-enabled,” but they are not: They are making individuals AI-enabled. Anyone can try tools at the individual level; those tools empower human execution. At the organizational level, however, people must empower AI. They must help AI move from not understanding to understanding its industry, the professional domain required, and the preferences of the specific business leader. People must enable those three layers of getting AI up to speed. In professional terms, this means treating AI as one of the employees and restructuring the organization to connect it to the business architecture. People ensure that dynamic iterations of that architecture remain effective. That is what we mean by end-to-end work. LynxAI's entire process covers the product- and brand-innovation lifecycle. At a smaller scale, it is AI adoption across a process; at a larger scale, it is AI adoption across the enterprise. Both our delivery team and the customers we serve are therefore prototypes of future AI-native business-to-business companies.
Duan Hongyu: We know that renewals and repeat purchases are generally where a business earns its highest profits. Since you mentioned a complete end-to-end process, could you explain how you currently create value in renewals or repeat purchases, and how AI does that?
Zhang Keyi: I think this is where the strongest compounding effect in LynxAI's current growth model appears. We genuinely have no dedicated sales roles. Our principle in working with customers is that whoever sells also delivers. During the engagement, our delivery team first aligns with the customer on the three dimensions I mentioned: the industry, the professional domain, and the business leader's trade-offs. This takes the form of a structured database combined with a knowledge base and AI-supported business processes. At that point, customers are effectively “spoiled” at the execution level but place more demands on the architecture and on dynamic rule updates. As long as collaboration continues, it generates positive feedback. They find themselves in a positive growth flywheel, learning how to work effectively with AI. We make the underlying systemic work stable and ensure that our AI solution best understands the industry, the company, and its founder. Renewal then follows naturally.
Chen Erhang: Let me discuss the entire process in the context of our business. We help Chinese export businesses acquire customers globally. The starting point is a worldwide population of potentially millions, tens of millions, or even hundreds of millions of users. The key is finding, among those hundreds of millions, the few buyers who will sign contracts and do business with us each year. That is a long process, which we divide into several stages. First, identify a few hundred sales leads from the behavior of millions or tens of millions of users. AI must do this work; that is already settled, because people cannot match its efficiency. Second, identify a few dozen buyers who actually have purchasing plans now among those hundreds. AI can already do some of this work. Third, identify the final few people with whom a deal can really be discussed among ten or twenty interested prospects. At present, that third stage remains primarily human-led. The second stage has changed the most over the past two years. For example, global promotion creates time-zone and language problems, which used to be major pain points. If a customer contacted me while I was asleep at night, I would definitely lose them. That no longer happens: There is no language problem, and AI can interact with them immediately on your behalf. So the second stage is beginning to replace people and reduce lost customers. Does AI have no role in the third stage? I think it does: It can provide many possible approaches. Many salespeople lack capabilities in certain customer conversations. Someone may be good at European and American markets but not know how to speak with a South American customer because of different cultural traditions. AI can suggest approaches, potentially performing better than people in that capacity. So AI has value and a role even in the third stage.
Duan Hongyu: Because you already have many use cases, you then examine the technology within each particular case.
Li Shouguo: I think we are probably aligned on this: Use cases definitely come first, because customers are buying value and ROI. Among our state-owned-enterprise customers, however, there is one exception: IT procurement, or the “technology parade”—following the crowd. Apart from that, use cases generally account for the greater share.