July 29, 2016, a Friday. Luo Xu later told Jizi Guangnian that day hurt until numbness.

Fenxiang Xiaoke laid off more than a thousand people that day: over eight hundred in sales, nearly two hundred in R&D—half the company left. Elevator ads still looped outside while desks were cleared inside.

Jizi Guangnian recorded that single-day cut above one thousand. At a Cyzone year-end talk, looking back on the whole amputation, Luo said the peak was about 3,800 people and nearly ¥300M in revenue, with roughly 1,800 cut in a short stretch—more like the full adjustment cycle than a second set of books for the same Friday. Along the same timeline, revenue would fall from nearly ¥300M to under ¥100M.

The numbers sting. The real question is not how many left:

Why is someone who has already succeeded in part the hardest person to convince they were wrong?

Act I: Why he would not admit it

Fenxiang had tried mobile office, light CRM, then briefly rebranded as “Fenxiang Xiaoke (逍客)” and bundled IM, OA, and CRM into a mobile work platform. Luo jumped from media into enterprise software, skilled at trading volume for awareness: street teams, Focus Media ads, consumer-internet playbooks. Users climbed, month-on-month revenue looked good, capital chased. With headcount and noise, an illusion grows easily—we are not selling features; we are seizing the next enterprise entry point.

That is where partial success turns dangerous. It gives you resources—and a nearly self-sealing proof: if we are already running, the judgment must be roughly right.

When DingTalk arrived in 2015, the product was not instantly stronger. Money and “free” were. DingTalk offered terms: Fenxiang drops IM/OA and becomes an ISV. Luo refused on the spot.

Pause for a question that only hurts afterward: what if he had said yes?

Fenxiang might not have burned the later ~¥100M, nor faced that mass layoff. It would also have meant admitting, then and there, that the enterprise entry he wanted to seize was never his. For a founder fattened on growth, that is harder to swallow than “becoming a vassal.” Refusing the ISV deal is not only a commercial fact; it foreshadows character: he chose to trust his own call, even with Alibaba behind the rival.

Niutoushe later recorded his self-mockery: a person’s greatest crisis is underestimating the opponent. The Luo of that moment likely could not hear it.

DingTalk ads hit elevators and celebrity stages. Luo did not retreat; he doubled down—made the competing IM/OA free too. DingTalk answered with free plus subsidies. Inside, the line was probably still: we still have a shot. Entry can be won with will and spend; SMBs adopt first, pay later; we have done CRM, we have a customer base, a latecomer will not choke us.

Wrong beliefs are often not born of ignorance. They are fed by staged victories.

Act II: Reality forces the admission

Nearly a year of blood-and-money war: Fenxiang burned about ¥100M. Activity decayed. Revenue stalled.

Mid-burn, the common psychology is not “I was wrong,” but a more dangerous sentence: one more round. Ads still run, hiring continues, the narrative still says entry—stopping would mean publicly admitting that the night he refused the ISV, he chose wrong. So larger spend protects an earlier judgment.

Data speaks another language. When ads stop, revenue snaps off. Luo later did the math himself: growth rode external force, not the product. The math came after the burn.

In May 2016 free ended; in July the fight stopped; the name returned to Fenxiang Xiaoke. That step feels like: we lost—for now. Then came that Friday layoff—we have to start over. ¥100M is not a line on a tuition receipt. It is him proving with his own hands: a startup cannot win a zero-price war against a platform.

Not long after the cuts, September 4, 2016. A strong distance runner, Luo entered a track meet and his heart stopped for about two and a half minutes. He used four characters for that state: company sudden death, person sudden death.

The books had just burned ¥100M; the office had just cleared a wave of people; outsiders asked if the company still existed; he lay on the track and spent nine months half recovering, half working. Here the question slides darker still—why keep going. Business models have to wait.

When does a founder finally admit they were wrong? This story’s answer is expensive: sometimes only after an obsession fed by staged success is torn apart by ¥100M, ~1,800 people, and two and a half minutes without a heartbeat.

Act III: What admitting error actually means

Admission is not a “strategic pivot” press conference.

Near the end of half-recovery, Luo told the team: he used to love grand stories; now he wanted a “good company,” not merely a “big company.” Inside an organization that had just cut more than a thousand people, that sentence rewrote KPIs and appetite—no longer proving life with headcount and noise, no longer needing “we are still the entry” to keep face.

The real moves hurt more. Product was nearly rebuilt toward connecting CRM; customers shifted from SMB toward mid-to-large; street-team sales became consultative. Luo told Niutoushe that in 2017 sales shrank from about seven hundred to a little over a hundred, while sales revenue rose more than twenty percent. Deal cycles stretched to one-to-three months; complexity became three or four times what it was. Jizi Guangnian wrote that for eighteen months after the layoff, Fenxiang nearly vanished from public view; veterans of street selling often could not handle proposal meetings and implementation; roles split; many left.

He stopped asking how to retake the collaboration entry and started asking: Whom should we serve? What will they pay for?

Admitting error becomes concrete cuts: cut headcount, cut playbooks, cut SMB fantasies, cut the entry dream, cut the self who refused to be a vassal. 3,800 people were not the enemy; they were a monument to the old battlefield. Whether you dare tear it down depends on whether you still need that monument to prove you were once right.

Act IV: The market says this time may be right

Whether the turn works waits for customers to vote with money.

Luo remembers the Digital China deal. Contact began in 2018; he led the pitch himself; it closed in 2019. Public retrospectives put the purchase near a few hundred thousand yuan—small for Digital China, a milestone for Fenxiang.

Hold the contrast: a company that once had ~3,800 people, nearly ¥300M revenue, and elevator ads fighting for entry, now has its founder personally selling a few-hundred-thousand-yuan CRM deal. Once correctness was proved by noise and armies; later, one unglamorous close proved he had finally corrected a little. Digital China kept expanding and renewing. He called it a great encouragement—only when someone pays again is the error rewritten.

In January 2018 Kingdee invested about $50M strategically and became the largest single shareholder; public language said Kingdee customers’ CRM needs would go to Fenxiang. That at least shows that beyond generic collaboration platforms, mid-to-large sales operations still need dedicated software. Fenxiang also sliced sales by industry—to see where deals actually stick.

The rescue was long. On the eve of Covid an investment LOI died while the team had already expanded; later digitization pushed business up. By early 2026 Cyzone interviews, public figures sat near ~¥600M revenue, roughly twenty percent recent growth, renewal above the “100% healthy line”; IDC trackers placed domestic CRM SaaS share and growth near the front for years.

From nearly ¥300M to under ¥100M to about ¥600M is not a straight flip. It is a string of abandonments aimed at the confident self of the past.

Act V: What remains is not a moat blueprint

After DingTalk came WeCom and Feishu; after CRM came AI. Platforms once made collaboration free; models now make many CRM surface features cheap. Luo’s rivals today are not the ones of a decade ago.

He may have learned something he did not know then: when the battlefield changes, do not rush to prove the old choice was right. Ask first—have I picked the wrong battlefield again?

Looking back, the ground did move: from enterprise entry and collaboration, to connecting CRM and mid-to-large customers, then deeper into industry and in-flow intelligence. You can call it changing moats. That is an after-the-fact outline.

What really saved Fenxiang Xiaoke was not finding a new moat first—it was finally admitting he had chosen the wrong battlefield, and daring to cut the methods that once proved he was successful.

Layoff lists, an ECG, and the words “good company” sit closer to the truth than any strategy deck. The river is what you retreat into. Before you retreat, you have to be willing to say: I was wrong.