You are using an older version of Internet Explorer. For the security of your Leifeng.com account and a better product experience, we strongly recommend using a faster and more secure browser

This is a temporary link, for article preview only, and will expire at that time

0 Before the National Day holiday, Shilong, an FDE who had recently entered the field, resigned from his consulting firm.

The thought had first crossed his mind in his first week on the job. After a long period of torment, he finally took the step.

Although he was nearly knocked flat by this wave of upheaval, he had no intention of leaving the FDE profession—he still wanted to hold tightly to the opportunities within it. Soon, he joined his second company.

With AI implementation riding the wind, booming market demand has made FDE a red-hot profession.

But when it comes to actual implementation, the reality is something else: no one can say for sure how much AI implementation is actually worth. Some have landed deals worth tens of millions, while others only get enough money to buy WorkBuddy accounts and do data interface modifications—the FDE delivery itself doesn't sell at a premium.

"The demand is real, and so is the freeloading." Two months ago, Gaofeng started his own business doing FDE work. After two months of struggle, his biggest takeaway is: making money is exhausting, and the market is a mess.

On one side, FDEs keep pouring in; on the other, a market with no established standards yet—the first batch of FDEs is struggling to survive in the chaos.

The Absurd Reality of AI Implementation

The chaos of the market is evident from the pricing alone.

Some FDEs have already turned down 4 requests this year, because each request was quoted at only 3,000 yuan, so they can only protect themselves by "rejecting orders." "It's not that we don't want to take them—it's that taking them wouldn't even cover the token costs."

But some are willing to bite the bullet, "even if it means losing money." In order to quickly build a presentable case and make it easier to land deals later, they have no choice but to accept extremely unreasonable budgets.

  • A Million-Yuan Project, Shrunk to 100,000

Zhang Yang, a frontline salesperson, closed a deal with a 100,000 yuan budget over the past two months.

But he had originally negotiated it as a million-yuan-scale AI algorithm project, and to win it, he had even planned to bring in an expert who had managed large-scale algorithm projects at a major tech company.

The client was a cross-border e-commerce trading company with clear needs and pain points: replenishment forecasting—but the task itself was quite difficult.

Order too little and you run out of stock; order too much and warehousing costs rise. Moreover, from production, export customs, import clearance to local warehouse shelving, the entire chain takes a sea freight cycle of over 100 days.

As communication deepened, Zhang Yang discovered that this problem, which looked well-suited for AI algorithms, actually hadn't yet reached the stage where algorithms could play a role. The million-yuan deal he had envisioned could never be implemented.

First, there was a gap in the client's understanding. Their initial idea was simple: "Just throw the data to the AI, and the AI can handle everything."

What's more, the client hadn't even sorted out its shipping data. Dozens of SKUs (stock-keeping units, such as products of different models and specifications) were shipped mixed together, and operations staff could only manually break down and match them, checking one by one how many of each SKU had been shipped and whether anything was missing.

Only then did Zhang Yang realize that the top priority was to return to the most basic data governance and solve the data matching and statistics problems.

After four or five rounds of negotiation, this AI forecasting project, originally expected to reach the million-yuan level, had to be downgraded. The client settled on a 100,000 yuan deal, and that budget shrank to just enough to buy 10 WorkBuddy accounts, plus data interface modifications.

  • An 8-Million-Yuan Quote, a Markup of Over Ten Times

"Foreigners' money is pretty easy to earn, too." Liu Qiang, responsible for AI product commercialization at a certain vendor, saw the other side of the story.

He had heard that a domestic peer had taken on a supply chain FDE project for a multinational tech giant, quoting 8 million yuan. After learning the details, he found it even more "absurd": the project mainly involved data processing, plus "answering a few queries," so employees could directly look up sales, inventory, and other data.

Based on Liu Qiang's own project experience, this kind of work usually takes about two months, with project values generally between 400,000 and 600,000 yuan.

The 8-million-yuan quote was more than ten times higher than the project prices he knew of.

But he could also understand: the service provider might have had to build capabilities from scratch, or the client's data foundation might have been so poor that a lot of dirty, tedious work was needed. But the key reason was still that "their boss had a good relationship, so they managed to sell it in."

In the AI implementation market, pricing an FDE project is like Schrödinger's box: before you open it, it's hard to judge how much it's actually worth.

After long experience dealing with clients, Zheng Tao, who does FDE delivery, found that FDE projects are very different from the SaaS work he used to do. SaaS is a standardized product with relatively clear pricing, but FDE delivery often involves a solution that needs constant adjustment based on the client's situation.

"There's a lot of subjective judgment involved. Sometimes it's hard to explain the difference between 80% and 90%."

After experiencing the absurd reality of "a million-yuan project shrinking dramatically," Zhang Yang set two "rules" for himself: no projects from unfamiliar clients, and no projects not demanded by the "top boss."

The reason is simple: "If you can't define the delivery criteria, you can't do acceptance in the end."

The Role Mismatch of Frontline FDEs

For FDEs new to the field, the difficulties lie not only at the client site but also within the company.

"Everyone is crossing the river by feeling the stones." Shilong entered the field full of hope, but only after becoming an FDE did he realize that in this new profession there was no well-worn path, and neither leaders nor senior colleagues had mature experience to pass on.

As his sense of professional worth was worn away bit by bit, he fell into self-doubt: Do I even count as an FDE? What does it take to be an FDE?

  • Marginalized, Reduced to a "Freebie"

Before entering the field, Shilong had gone deep into the workshop floor of a garment factory, using AI to help the factory build an ERP system. That experience convinced him that AI could help traditional manufacturing transform.

Precisely because of this, he always believed that an FDE should at least first go to the frontline of the client's business—this role is closest to AI implementation, and its value lies in eliminating the translation costs of layer upon layer of relayed communication.

But working as an FDE at this consulting firm, he "wasn't qualified to contact clients directly and could only follow his leader," never even setting foot on a client site—he was practically living as an outsourced worker. The client's intent had to be relayed through consultants and leaders, layer by layer, with information constantly degrading. "Sometimes, even the leader himself wasn't sure what the client actually wanted."

The organization also hadn't developed a mature FDE working methodology; the division of labor resembled a traditional software team: one person sorted out business rules, another built the system, and the pieces were stitched together through meetings and code merges—communication costs were extremely high.

Less than a month into the job, Shilong began considering whether to stay. He could never find his place in the organization: neither at the client site nor involved in the company's core business.

His company mainly sold Huawei's methodology: first produce a proposal, and only after the client nodded would they sign and move in. The foundation of the business was the trust between consultants and clients. "They make money from that, not from the FDE's understanding of the business and then coding it up."

The demos he built were simply thrown to clients for free, to prove that "we have AI implementation capabilities." He often mocked himself as a "freebie," even "a freebie nobody wants."

After making up his mind to leave, Shilong quickly received an FDE offer from a Feishu reseller, but he turned it down without much hesitation. In his view, the role was essentially still selling tools, which didn't match his original reason for becoming an FDE. (For FDEs at AI office-productivity vendors, are they selling tools or doing implementation? Feel free to add the author on WeChat at skylar_12_14 to exchange frontline insights)

In the end, he chose to join a startup focused exclusively on FDE work.

  • Hard to Enter the Field, and Hard to Mentor Newcomers Too

But getting AI to work in enterprises is far more complicated than imagined.

Zhang Yang has spent years in the To B industry and has found that newcomers with little To B experience often underestimate the difficulty of AI implementation.

A client once wanted an AI agent for market dynamics analysis, and a young newcomer casually said, "That's easy, we've built one of those before." Zhang Yang reminded him not to mistake a demo he could build for an AI system that could actually be deployed in an enterprise.

"The gap in between is enormous."

Based on his experience, a personally built demo and an enterprise-grade system may look similar in features, but the workload and complexity are on completely different scales. He once built a knowledge base with over 20,000 lines of code, but to make it enterprise-grade, it might take 250,000 lines.

Still, these young people understand AI technology, have ideas, and bring a lot of enthusiasm, and teams are generally willing to use a "mentor-mentee" model to help them get started.

This kind of "old hands guiding newcomers" setup is not uncommon in FDE teams.

Cui Jinyu, CEO of Zhigou Xinqing, is building an AI-native FDE team: an FDE project is typically staffed with two or three people, led by a senior business expert with a deeper understanding of AI serving as pod leader, who brings along one or two juniors responsible for hands-on execution.

But this model puts an organization's way of working to the test. Under a traditional division of labor, FDE newcomers can hardly get enough room to show what they can do.

The same goes for software companies, not just consulting firms. An industry insider told Lei Feng Wang that most software companies claiming to do FDE still start by building an overly complex SaaS middle platform, piling up a huge stack of features, only for users to log in and find the same old thing.

"Their people (FDEs) are trained that way too." Under the weight of legacy baggage and ingrained habits, FDEs can easily degenerate into on-site outsourcing, pushed to the margins.

The Survival Game of Entrepreneurs

Unwilling to be constrained by the old system, some choose to strike out on their own. In FDE entrepreneurship, the top priority is "keeping the company alive first."

Entrepreneurs never run out of things to worry about: at first, it's all about acquiring customers and winning deals; once deals come in, they need to solve conversion; once conversion works, they distill methodologies; as deals pile up, they're busy hiring and expanding the team; and later they discover that in B2B, you still have to build channels and ecosystems.

But to survive, there's an even longer-term matter: how to prove you can stand on your own feet in a line of work that outsiders question. To gain a foothold inevitably involves bargaining, compromise, and leveraging others.

  • Getting Exploited for Free, Impossible to Fully Guard Against

Every time Gao Feng meets a client, it's almost a battle of wits: right off the bat, the other side may want both a solution and training.

Getting enterprises to pay is not easy. The first lesson of FDE entrepreneurship is facing the reality of being exploited for free: spend an hour in conversation and end up with nothing but a polite remark. Over the past two months, Gao Feng has been exploited for free at least a dozen times.

At first, he didn't know how to defend against it either. It requires drawing an appropriate boundary—what counts as business investment, and what counts as being exploited for free—but that's hard to define.

Some requests seem, to the client, like trivial favors. But their questions come at him like machine-gun fire, and after he answers them all, they leave with just one line: "Very professional, thank you. Let's have dinner together next time."

Gao Feng often spends half a day or even a full day dealing with these "trivial" requests. That kind of time cost is a real luxury for an entrepreneur—when things get busy, he meets clients, job candidates, and partners back to back all day, at peak times at least six groups of people.

Later he worked out some rules of thumb: first, spend twenty minutes quickly discussing the business requirements. If the other side shows interest, have the business manager step in first, "letting him shield me up front."

This is also a business lesson many FDEs learned after getting burned: "Send pure tech people straight at clients, and sooner or later they'll be exploited for free to death." But turning down clients is not something you can leave to technical staff. Once a technical person says no, the relationship may be beyond repair; if it's a salesperson who says no, even if the client gets angry, the technical staff still have a chance to salvage things.

But Gao Feng treats being exploited for free as tuition for learning the ins and outs of this business.

"What I fear is the company being massively exploited for free in the future." In his view, one person being exploited for free isn't worth much and still builds some insight and experience; but once the company scales up and everyone is being exploited for free, it's a different story.

  • New Players, Old Rules

But whether you can survive is sometimes not something an individual or a team can decide.

Zhao Lei, head of another FDE startup, has felt firsthand that the market is big and demand is growing, but competition is fierce, and a considerable portion of client resources is still held in the hands of established software companies.

What's harder is that traditional vendors still use practices like per-person-day pricing in FDE projects, which easily boxes FDE into the positioning of "premium outsourcing." Once clients get used to that standard, it becomes even harder for new players to prove themselves.

FDE entrepreneurs are gradually realizing that industry standards cannot be left in others' hands—they must first seize the power to define the terms.

In Cui Jinyu's view, FDE should not be a labor service. "If it ends up as a software company or an outsourcing company where life is exhausting and grueling, you might as well go back to working at a big tech firm."

What he envisions is an FDEaaS (FDE as a Software) model: letting AI step by step take over the work of traditional consultants and engineers—drafting solutions, writing code—until these tasks are fully taken over by agents, with "humans" retreating behind the scenes.

There is still a long, gravel-strewn road to walk in between. What can be done now is still to rely on "humans" to dig out reusable methodologies for AI deployment.

Some entrepreneurs have also started considering leveraging channel partners: first help the channel partners build up their own FDE teams and shore up their delivery capabilities, then work together to open up client resources.

The price is that channel partners take a cut of revenue, and the share is not small, which erodes gross margin. But for a startup, in the early days there's no choice but to accept it.

"After all, you have to survive first; only then do you get pricing power." (How can FDE make its business model work? What success stories or cautionary tales are there? Feel free to add the author on WeChat at skylar_12_14 to discuss.)

Some Switch Lanes, Some Charge Ahead

"Whoever can first carve a path out of the chaos has a better chance of surviving longer." The FDE trade is both red-hot and chaotic.

Some choose to switch lanes and start over. Doing FDE at his new company, Shi Long returned to the familiar manufacturing industry, busy helping factories build a highly automated agent system.

He shuttles daily between the company and clients' factories, living an "807" schedule—8 a.m. to midnight, seven days a week. But he's finally doing what he understands FDE work should be, and has found his place in the team: at least he's no longer a marginal "freebie."

Some insist on charging ahead and refuse to leave the table. Cui Jinyu chose to steady himself first. In his view, "Although it can't happen overnight, you must plant a seed from the very beginning and wait for it to sprout."

But in a new trade, no one can say for sure what tomorrow holds. As Shi Long often jokes about himself: "Do it well and you're a pioneer; do it poorly and you're a martyr."

Original content by Lei Feng Wang (WeChat official account: Lei Feng Wang). The author covers big tech's AI-to-B business. Feel free to add the author on WeChat at skylar_12_14 to discuss.

Note: Gao Feng, Zhang Yang, Liu Qiang, Zheng Tao, and Zhao Lei in this article are all pseudonyms.

This is an original article by Lei Feng Wang. Unauthorized reproduction is prohibited. See the reposting guidelines for details.

Related Articles

Editor

For your account's security, please verify your email

Please verify your email