Chief Product Officer / China
Three defaults we changed and what each one cost
- Annual revenue
- RMB412m
- Calendar 2025; memberships, live classes and course materials
- Paying members
- 1.38m
- June 2026; 62% aged fifty or over
- Product colleagues
- 214
- Product, design, research and safety tooling
Decisions & consequences
01Ending a one-yuan trial that renewed itself at full price
In 2020 our growth team introduced a seven-day trial for one yuan that converted automatically into a RMB28 monthly membership unless cancelled. By early 2022 it produced most of our new paying members, and also about 3,900 refund requests a month. When customer service sampled them, four in ten came from an adult child who had found the charge on a payment account shared with a parent over sixty. Many did not know they had agreed to a renewal.
The growth director argued that the trial disclosed its terms and that removing the automatic conversion would cut new-member revenue sharply. He was right about the revenue. I proposed that the trial end with a choice rather than a charge: on day six the member would see the plan, the price and two equally prominent buttons. I accepted that conversion would fall. Our chief executive approved the change after we agreed to review it after two quarters.
We redesigned the whole renewal path at the same time. Annual members received reminders in the app and by text message seven days before a renewal. Cancellation moved to the same screen as the membership details. Where a relative paid, that relative received the reminder as well as the learner. Customer service was given a single refund rule for disputed first charges.
Trial-to-paid conversion fell from 34% to 22%. Refund requests linked to the trial fell to about 760 a month, and billing contacts fell 58%. Twelve-month retention among new members rose from 39% to 55%. New-member revenue in year one was RMB14 million below plan. When implementing regulations under China's consumer-protection law took effect in July 2024 and called for prominent notice before automatic renewal, our flow needed only wording changes.
Parents of children's course members wanted lessons to continue without interruption and found the end-of-trial choice tiresome. I should have tested older members and parents separately, and offered an explicit, reminded renewal option at sign-up to those who wanted one.
Taking children's practice videos off the public community wall
Our children's arts courses let parents upload a child's calligraphy, drawing or guzheng practice for a teacher's feedback. Until 2023 those videos also appeared on a public practice wall where any member could like, comment and share. The wall was popular: the community team attributed 17% of new children's enrolments to shared videos. It also generated about 41,000 items a week for review, and in mid-2023 our reviewers identified a cluster of adult accounts that followed children's videos systematically.
China's minors' protection law, and the regulations on protecting minors online then being finalised, pointed in one direction, but the product decision was ours. The head of children's courses and the community director both opposed full removal and proposed stronger filtering instead. I recommended that a child's uploads become visible only to a guardian's own account, the assigned teacher and, if a guardian chose, a class group of verified guardians. Public likes and downloads would be removed. Our chief executive agreed, and the board received the paper at its next meeting.
We introduced the change in September 2023 with four weeks' notice to parents and teachers. Guardian accounts completed real-name verification before joining a class group, and groups were limited to thirty families. Teachers received a private showcase they could share with a guardian's permission. Comments on children's work were limited to the teacher.
New children's enrolments fell 11% over the following two quarters, roughly RMB9 million of revenue. Some teachers lost the visibility that had attracted students. Weekly review items involving children fell from 41,000 to 9,500, and reports of inappropriate contact on children's content fell from 230 a quarter to 12. A year later, renewal of children's courses had risen from 61% to 66%, which I attribute partly to parents' greater confidence and partly to course changes made in the same period.
I would brief teachers earlier and have the private showcase ready before the wall closed. For several weeks they had lost their best recruitment tool and had nothing to replace it.
Closing private messages between adults without an established connection
Through 2024 our older members reported a growing number of private messages from accounts they did not know. Most offered health products, paid singing groups on other apps or investment clubs. Members over sixty made 70% of those reports. By early 2025 the review queue for messages and comments had reached about 52,000 items a day, and monthly solicitation reports had reached 8,600. Some of our own contracted teachers were also using private messages to recruit students into paid groups elsewhere.
The community director argued that messaging was how members found practice partners and that most messages were harmless. Two of our best-known teachers said the change would end their relationship with students. I recommended that private messages be allowed only between members in the same class or who had followed each other for thirty days, that comments containing contact details or links be held for review, and that a warning appear whenever an unfamiliar account asked a member to move to another app. Teachers would message students through class channels the platform could see.
We introduced the rules in April 2025 after testing them with our member panel. Members could appeal a blocked contact, and in June we added a class-code invitation so that genuine practice partners could connect without the waiting period.
Private messages fell 64% and public comments fell 18%. The two teachers left, and with them a combined following of 31,000 students, some of whom later returned to other teachers. Solicitation reports fell to about 1,900 a month, and the review queue fell to around 30,000 items a day. We moved 22 reviewers to children's content and appeals. Ninety-day retention among members over sixty dropped one point in the first quarter and then recovered.
I should have built the class-code invitation before the rules took effect. For two months the thirty-day rule blocked genuine partners, and appeals overwhelmed a team that had not been staffed for them.