product sense · standard

Design a new feature for Discord

Design a new feature for Discord.

Updated Jun 2026 Calibrated to the strong-hire bar

This question is a scope and judgment test before it is a feature design test. Only about 25% of candidates finish the exercise in time. The ones who fail do not run out of ideas; they run out of time because they spent half the session clarifying and brainstorming and never reached metrics or trade-offs. The move is to narrow scope in the first 60 seconds, name a specific persona, and drive to a concrete feature proposal before the halfway point.

What Discord interviewers actually look for

Discord’s product is built on community trust. The company earns revenue through Nitro subscriptions (roughly 54% of $561M in 2025 revenue), server subscriptions where they take 30% and creators keep 70%, and a growing Quests and ads layer. A feature that damages community trust or feels like a monetization grab to users can destroy the relationship that makes all of that possible. Interviewers award points for:

  • Naming a persona who is not a gamer (46% of Discord’s users now identify as non-gamers; productivity and education communities grew 40% year-over-year in 2025)
  • Weaving safety and privacy into the feature design from the start, not as a footnote
  • Connecting the feature to Discord’s actual revenue model rather than proposing something that requires an entirely new business line
  • Acknowledging real platform gaps: Server Insights requires 500+ members, leaving most small and mid-size communities blind to retention data; there is no algorithmic discovery; no native async video like Loom
  • Sketching a live wireframe or flow, even verbally

Structure a strong answer

strong

"Before I dive in: I want to pick a scope that maps to something Discord actually cares about strategically. Discord has 259 million MAU across 28 million servers, but only 19 million of those servers are weekly active. That means a large activation and retention problem is hiding in the dormant server cohort. I'd focus on mid-size servers, 100 to 500 members, because they are too big to feel like a friend group and too small to unlock Server Insights. A server admin in that range has no visibility into whether her community is actually working. Is that a good scope for this exercise?

Persona: a 28-year-old running a 200-member study-abroad community. She is not a gamer. Her group landed on Discord because someone in the group already used it. She spends three to four hours a week on moderation and onboarding. Her biggest problem: she loses 30 to 40% of new members within 48 hours and finds out weeks later, when the server just feels quieter. There is no warm handoff moment. She can see who joined but not who is about to ghost.

The real pain is not that she lacks analytics. It is that she cannot act at the moment of dropout. By the time she notices a member is gone, the window to intervene is closed.

Feature: New Member Momentum. An opt-in digest that fires 48 hours after a new member joins if they have not sent a message. It surfaces to the server admin: which channel types the member browsed (directional signal only, not a per-message read receipt), the one channel whose topic keywords most closely match the member's intro message, and a single-tap action to send a personalized DM using a template the admin set up at server creation. No AI-generated message sent in the admin's name without her reading it. Human-in-the-loop by design, because one bad auto-DM to the wrong person damages a small community fast.

Viability: this feature extends the value of Nitro Server Boosts. Right now, Server Insights requires 500 members. New Member Momentum can be the 'Insights for growing servers' benefit unlocked at a lower boost tier. Admins who retain members grow communities that pay more in subscriptions. Discord's revenue interest and the admin's interest are aligned.

Lovability: the feature works in Discord's existing vocabulary. It is a ping, a DM, a channel reference. It does not add a management console or a dashboard tab. It feels like a helpful nudge, not a SaaS product grafted onto a community tool.

Privacy design: browsing activity is surfaced as directional signal only, which channel type the member visited, not which messages they read. Members see a disclosure on join: 'Your server admin may receive anonymous activity signals to help you get oriented.' They can opt out in privacy settings. This is a disclosure, not a dark pattern. It mirrors Discord's value of community warmth without making members feel watched.

Metrics. Primary: D2 new member message rate, the share of new members who send at least one message within 48 hours of joining. Secondary: 30-day member retention per server. Guardrail: admin DM send rate capped at one per member per week to prevent spam. Counter-metric to monitor: opt-out rate on the join disclosure. If it exceeds 15%, the trust cost is too high and we pause to redesign the disclosure.

Trade-offs I chose not to pursue: I did not propose AI-generated welcome messages because Discord's community trust is under active pressure from recent monetization prompt testing. A feature that sends a message in the admin's name without her reading it risks breaking that trust on the first bad interaction. The value of the human-in-the-loop is that the admin owns every DM she sends."

weak

"I would design a feature that uses AI to automatically moderate servers." No persona, no pain point, no business case, and safety is framed as a technical problem rather than a community trust problem. Interviewers at Discord have heard this answer from hundreds of candidates. It does not demonstrate product judgment; it demonstrates familiarity with the phrase 'AI moderation.'

A close second: "Discord should add a TikTok-style discovery feed so users can find new servers." Discovery is a real gap in Discord's product. There is no algorithmic feed, which is a structural weakness compared to Reddit or TikTok for content-forward communities. But this answer fails for three reasons: it ignores that Discord's identity is built around opted-in communities where intimacy is the product; it does not name who specifically is experiencing a discovery problem; and it does not address how you handle safety at scale on a new content surface. Proposing a feed for Discord without addressing those tensions signals you have not used the product seriously.

The structural failure of weak answers is the same in almost every case: candidates spend 15 to 20 minutes on clarifying questions and brainstorming, rush the feature design, and have nothing left for metrics or trade-offs. Since only 25% of candidates finish the exercise, any answer that never reaches trade-offs is failing the second half of the rubric by default.

The safety rule: design in, not bolt on

Discord interviewers treat safety as a first-class product requirement, not a compliance checkbox. Naming it as a footnote (“and of course we’d add content moderation”) signals that you do not actually understand the product. The tell of a strong answer is that safety constraints shape the feature from the beginning. In the example above, the reason the feature uses a human-in-the-loop DM rather than an AI-generated one is a safety and trust decision, not a technical limitation. That is the kind of reasoning that earns bonus points.

The 2026 context you need to know

Discord’s growth problem in 2026 is not acquisition. It has 656 million registered users and 94 minutes of daily engagement per active user. The problem is activation and retention of the 46% non-gaming communities that joined because their group ended up there, not because they chose Discord deliberately. These admins are not community managers by training. They do not have the instincts a gaming server moderator has built over years. Feasibility is largely free: read receipts, keyword matching, and DM templates are buildable in weeks. The hard problem is lovable: any feature that feels like surveillance, a SaaS dashboard, or a monetization grab will be rejected by Discord’s culture before it earns its place in the product. The winning feature amplifies human warmth at scale without replacing it with automation.

See also how AI changed what viable and lovable mean for PMs and the Discord interview process overview.

Asked at