In 'Moneyball,' the real crisis for the Oakland Athletics wasn't losing a game, but that their old winning formula suddenly stopped working.
Star players were being bought up by richer teams, the team's budget was a fraction of the big clubs', yet scouts were still in meetings debating who had a prettier swing or looked more like a star.
General manager Billy Beane started asking a different question: if you can't afford the best players, what actually determines winning and losing?
Then came Moneyball. It wasn't about spending less; it was about recalculating what's worth keeping when an old growth model stops working.
Keep did the same recalculation for itself. Should Keepland stay open? Are low-margin large equipment lines still worth pursuing? How many people do events need?
Many things that were once considered "Keep should have" were put back on the table, then gradually disappeared.
This subtraction strategy has paid off. In 2025, Keep posted its first annual adjusted profit since listing; by the first half of 2026, adjusted net profit was 5.88 million yuan, with net loss narrowing to 12.19 million yuan.
But after the interim report came out on August 24, another question emerged. In the first half of 2026, Keep's revenue was 825 million yuan, up 0.4% year-on-year. In 2025, Keep's full-year revenue was 1.637 billion yuan, down 20.7% year-on-year.
Revenue has finally stopped retreating noticeably, but it hasn't really moved forward either.
At the earnings call, Wang Ning summarized: "Revenue scale remains stable, but the core is getting stronger." He also defined the first half of 2026 as a period of adjustment that Keep actively chose, admitting that "the current numbers are not pretty."
The problem has also become clearer. What Keep has found is not a new growth curve, but a way to keep surviving without growth.
When Times Are Best, Don't Overthink the Numbers
If you used Keep early on, you probably remember what it looked like at the start. There wasn't much to it.
Want to lose weight but don't know how to start? Open Keep. Don't know how to do a squat? Put your phone on the floor and follow along.
Keep's most important capability is making something that was originally quite troublesome simple.
Start exercising first, and users quickly followed. In 2022, Keep's average monthly active users reached 36.39 million, with about 2.1 billion workouts completed throughout the year.
With more than 30 million people gathered in one app, for an internet company, many accounts can be settled later.
What Wang Ning wanted went far beyond a fitness app. In 2018, after surpassing 100 million users, he said in an interview that Keep aimed to be a sports brand, with the app 'just a starting point'; if he had to name a benchmark, he cited Nike.
In his vision, Keepland served cities, KeepKit smart hardware entered homes, and apparel, content, social features and user data together formed a complete fitness ecosystem.
That largely explains why Keep later expanded into so many areas. Yoga mats, treadmills, wristbands, apparel, food, Keepland, events, livestreaming and an online mall were all folded into the same brand.
The idea was to move users from their phones to living rooms, gyms and more consumption scenarios, then bring their data back to Keep. The logic was simple: once people were there, Keep should build more business around them.
The problem was that users began to leave. Keep's average monthly active users fell from 36.39 million in 2022 to 29.92 million in 2024, 21.77 million in 2025, and further to 18.58 million in the first half of 2026.
In four years, nearly half of monthly active users disappeared. Keep, which had been aggressively expanding outward eight years ago, began to pull back.
At the 2025 earnings call, Wang Ning described the shift as 'cutting fat and building muscle': proactively eliminating low-margin businesses and refocusing resources on membership subscriptions, fitness equipment and apparel.
As a result, Keep's 2025 revenue fell 20.7% year over year, but the company achieved its first annual adjusted profit since listing.
Wang Ning's operating logic over eight years has nearly come full circle, shifting from 'we should have everything' to re-examining 'what is worth keeping.' The same shift is visible among users.
Wang Ning attributes the continued MAU decline to a user base evolving toward a healthier model of high stickiness and high frequency.
The first-half results do offer some evidence: average monthly revenue per monthly active user rose to 7.4 yuan from 6.1 yuan, and average monthly workout time increased 15.3% year over year. But the paid-user data tell a different story. Average monthly paying subscribers fell to 2.17 million from 2.79 million, a drop of more than 20%, while membership penetration slipped to 11.7% from 12.4%.
The MAU decline could be framed as a deliberate weeding-out of less active users. But with both subscriber numbers and penetration falling, that 'user purification' explanation becomes harder to sustain.
Keep hasn't only lost the casual traffic it could afford to shed. The costs obscured by growth are now surfacing line by line.
Full-time headcount fell to 632 at mid-year from 827 at the end of 2024. First-half employee benefit costs dropped 22.7% year over year, administrative expenses 32.3%, and R&D spending 23.2%.
The cost-cutting has narrowed losses, but there's only so much to cut.
Internet Profit Margins Keep Shrinking
Keep's best-performing segment now is its sports products.
In the first half of this year, revenue from own-brand sports products reached 483 million yuan, accounting for 58.5% of total revenue, with gross margin rising to 40.1% from 34.8%. Growth came from lighter, faster-turning products better suited to content-driven e-commerce, including yoga mats, dumbbells, kettlebells, resistance bands and protein foods.
Sales on Douyin grew more than 50%, and distribution channels rose 34%. Overseas revenue reached 22.1 million yuan, mainly from selling sports equipment through Amazon and TikTok.
Keep is achieving something it had not fully done before: letting the brand sell products independently of the app. A consumer can scroll through Douyin, see a Keep resistance band and place an order directly.
For a consumer brand, that is progress.
It could even be said that this is approaching Wang Ning's 2018 vision in an unexpected way: the app no longer has to be Keep's only center, and the Keep name itself is beginning to transact as a brand.
But that is precisely the problem.
In 2018, Wang Ning said the app was just a starting point, but he did not want to earn hardware margins by selling treadmills. In his plan, the real value of hardware and offline spaces was to distribute Keep's content, services and social features, and bring user data back.
Eight years later, what is propping up Keep's revenue is, first of all, yoga mats, dumbbells, kettlebells, resistance bands and sports food.
That creates an intriguing contrast: Wang Ning once sought to redefine a sports brand through the internet. Today, Keep is indeed becoming more like a sports brand — just not necessarily the one he had in mind.
The most pointed skepticism stems from this. In the first half, gross profit from sports products rose by about 55.69 million yuan, while the higher-margin online membership and paid content business saw gross profit fall by about 60.58 million yuan. The former had a gross margin of 40.1%; the latter, 73.0%.
Unlike physical goods, online membership revenue doesn't carry the same production, storage and logistics costs. Selling a device entails production, warehousing, logistics, platform commissions and after-sales support. In the first half, Keep's fulfillment expenses rose 19.6%, while sales and marketing spending also climbed alongside product revenue.
What's happening at Keep isn't simply "products rising, memberships falling." Rather, high-margin internet revenue is contracting, and consumer-products revenue — lower-margin but easier to scale — is taking its place.
Revenue scale is being supported for now, but the quality of revenue has shifted.
The capital market reacted even before the earnings. In August 2023, Keep's shares briefly reached HK$42.40, giving it a market value of about HK$22.2 billion. On Aug. 24, 2026, the day it released interim results, the stock closed at HK$1.76, leaving a market value of less than HK$900 million.
AI Rebuilds More Than Just a Membership Card
Wang Ning is betting the next opportunity on AI.
In February 2025, he declared "All in AI" in a letter to all employees. Months later, Wang Ning offered an aggressive scorecard for the pivot: In July 2025, Keep's AI daily active users were about 150,000 to 200,000, with AI revenue exceeding 1 million yuan. He projected AI DAU would surpass 1 million by year-end, and the company forecast AI annual recurring revenue could exceed 200 million yuan in 2026.
A year later, Keep has launched Keepace.ai, App 9.0, AI courses, AI coaching, and a Super AI membership. AI now powers course generation, exercise Q&A, voice-assisted running, food recognition, and analysis of eating, exercise, and sleep data.
AI's first proven benefit is helping Keep continue to streamline. In the first half of this year, Keep launched more than 8,000 AI-generated courses, and daily token consumption doubled compared with before the app revamp.
Meanwhile, the company's R&D spending continued to decline. Keep said externally that AI investment is more of a substitution than a simple cost increase. In other words, AI has begun to change how Keep produces content and organizes resources.
But cost reduction is not the answer the market is waiting for.
Keep has proven AI can make course production cheaper, but has not yet proven AI can bring more users back, nor that users are willing to pay more for it.
Keep currently charges 19 yuan per month for its regular membership on a continuous subscription, and 68 yuan for Super AI membership.
That means Keep has effectively posed a very specific question to users: Is AI worth paying an extra 49 yuan per month?
Since the Super AI membership launched, public social platforms have seen mixed reviews on AI recognition accuracy, the degree of personalization in training plans, and user experience.
These responses may not represent Keep's entire user base, but they highlight the toughest challenge in AI monetization: adding an AI feature doesn't automatically boost users' willingness to pay.
Wang Ning's own timetable is also closing in.
In March 2026, after Keep posted its first annual adjusted profit, Wang Ning set a goal for 2026 that went beyond 'continued adjustments.' The company guided for double-digit revenue growth in 2026, returning to more than 2 billion yuan. Six months later, revenue came in at 825 million yuan, up just 0.4% from a year earlier.
At the interim results meeting in August, Wang Ning pushed the prospect of bigger growth further out. He acknowledged the current numbers were weak and said the priority now is to strengthen the models and product capabilities.
Taken together, the two statements are more telling than either one on its own.
In March, the market was told revenue would return to above 2 billion yuan in 2026. By August, the message was that product and AI capabilities would deliver further gains in the coming years.
Also worth watching is another target Wang Ning set earlier: AI annual recurring revenue could exceed 200 million yuan in 2026. In the interim report, Keep disclosed extensive AI usage, course counts and capability metrics, but did not break out AI revenue, AI subscribers or the current ARR progress.
The open question is how many users will keep paying for Keep's AI.
The Moneyball approach ultimately worked not because the Oakland Athletics made their roster cheaper, but because Billy Beane found a variable that the market undervalued and that truly produced wins.
Wang Ning found a similar set of variables for Keep: cut inefficient businesses, shrink the organization, refocus categories, raise revenue per user, and use AI to lower content and organizational costs. In 2025, the company posted its first annual adjusted profit.
But those solved the previous problem: how to lose less money and survive. The next problem is growth.
That is the biggest difference between today's Wang Ning and the Wang Ning of 2018.
Back then, Keep had more than 100 million users, and he was focused on turning an app into a sports brand like Nike and bringing Keep into more living rooms, gyms, and lives. Today, he faces a much smaller but far more efficient Keep and is trying to use AI to recapture growth.
In the eight years in between, Keepland was wound down, large equipment lines were cut, headcount fell, and monthly active users declined steadily.
A company's Moneyball can't ultimately be just about making losing cheaper.
What Wang Ning needs to prove next is whether AI merely makes Keep run more efficiently or can actually help it win again.