Ten pillars. Ten leading indicators.

Channels that stay alive.

Each watercooler is bound to a Slack channel — General, Project, Guild, or Random — and the AI matches the tone. A project-thread gets energising questions about craft; a random channel gets playful; a guild channel gets thoughtful. The bot posts, then drops its own first reply 60 seconds later to break the ice.

AI proposes fresh questions from what the channel was actually talking about that week — every suggestion shows permalinks to the messages that inspired it, so admins approve in context. Humor guidelines are hard-coded: no sex, stereotypes, politics, mental health, health, family, finances, performance, or trending crises. When in doubt, kindness wins.

Once a thread gets going, fynn stays engaged — a teammate can @fynn the bot for a brand-voice reply, and an autonomous follow-up drops in after 2+ unique repliers if conversation has momentum (admin-controlled, can be silenced workspace-wide). Threads that stay alive are the whole point.

Know how your team is actually doing.

Create a check-in with a title and a short context paragraph that says what the round is for. fynn suggests questions grounded in that context — you approve, edit, or write your own. The whole exchange happens in one DM thread per teammate: opener, questions, follow-ups, and a wrap-up — so it reads as a conversation, not three system messages.

fynn plans the question order per recipient using their engagement profile and last round's reply, opens with continuity instead of a cold start, and asks curiosity-first follow-ups that anchor on a noun the teammate actually used. The next question only fires after the user replies — no 90-minute stopwatches. When the round closes, a Canvas digest publishes to the channel you chose, weaving cross-run patterns the team should see.

Birthdays and anniversaries that actually move people.

fynn's bar is explicit: a birthday post should make the recipient feel genuinely seen — not "happy birthday to an amazing teammate 🎂". Each celebration pulls from the last three weeks of Channels context, references specific things the person shipped or said, and posts into the public channel you pick (e.g. #team-updates).

Dates come straight from your HRIS via a direct OAuth integration — Zoho People today, with BambooHR / Workday / Rippling as drop-in additions. No CSVs to maintain, no list that drifts out of sync with HR reality. fynn pulls read-only; nothing is written back.

Optional add-ons: Suno generates a custom birthday song from the same context; Leapsome receives the anniversary recognition on the person's public praise feed. Without either integration, the core AI message still ships.

Anyone can DM fynn celebrations off to skip public birthday/anniversary posts for themselves — no admin dependency, no explanation required.

Point-in-time deep dives. Recurring pulses. Follow-ups that read like a colleague.

The multi-question variant is a Slack modal: one DM invite with a "Take survey" button opens all questions at once, submit once, done. Question types cover the real ones — single / multi-choice, rating scale, NPS, free text, yes/no. The single-question variant is a Poll that posts inline to a channel for a quick pulse.

Run them as one-shots, as recurring templates (weekly pulse, fortnightly retro, monthly engagement read — each cycle spawns a fresh instance with its own responses + analytics + close-out post; the parent template's "Past runs" surface shows the trend without polluting the list), or as per-user start-date triggers that fire N days after each recipient's hire anniversary (onboarding day-7 / day-30 check-ins land on each new hire's individual timeline, not a fixed calendar date).

A 17-pack best-practice library ships with every workspace and auto-loads on first use. The EX baseline covers Onboarding, Offboarding, Manager Effectiveness, Engagement, Comp & Benefits, D&I, Mental Health, L&D, Teamwork, Remote Work, Autonomy & Enablement, Change Management, and Hyper Growth. The cultural-transformation cluster is for teams shifting how they work — AI-Native Culture, Psychological Safety, Feedback Culture, Innovation & Experimentation, Values in Action, Decision-Making Cadence — plus short quick-pulse picks (Weekly Pulse, Recharge Check, Employee NPS). The picker auto-substitutes your workspace name in every question; admins clone-then-edit, the canonical pack stays clean. Manage the full library at Surveys & Polls → Packs.

Choice questions support an "Other" with tell-us-more elaboration that captures free-text only when a qualifying option is picked. Any question can carry a show only if rule — gating it on a prior single-choice or yes/no answer — that filters insights and CSV exports to the qualifying respondent set. Re-clicking Suggest with AI or Draft invite with AI always produces materially different output; fynn carries a 24h per-field avoid-list so admins iterating on copy never get the same line twice.

fynn drafts the questions from recent team activity (channel-aware — the thing Polly and Viva fynn don't do), drafts the invite copy, clusters free-text answers into themes on close, and scores sentiment per free-text response. Every draft goes through admin approval — nothing ships without a human nod.

Dynamic-cadence in-thread nudges handle the hard part — getting late respondents back. Cadence scales with open→close window: 1 nudge for a 3-day flash poll, 2 for a 10-day pulse, 3 for a 3-week deep dive. Each lands in the original DM thread (not a fresh cold ping) and progresses one slot per cron tick, so a late-joining user always sees the empathy beat first ("totally fair if your week's been packed — your honest take really shapes what we change next") before the impact and last-shot variants. AI-written copy, brand-voiced; a per-slot emoji react on the parent DM bumps the IM to the top of the recipient's channel list when the thread is collapsed. Anonymous channel polls are vote-once — fynn hashes tenant : survey : slack_user_id with a 24h TTL so a single voter can't stuff the count, and the raw Slack id is never persisted anywhere.

Anonymous mode is a three-layer contract, not a checkbox: nullable user_id on the response, ±30s jitter on submitted timestamps, and a reflection-asserted test that prevents anyone from ever joining SurveyInvitation to SurveyResponse in code. The anonymous-loser path on a double-submit returns an unsaved stub — never another respondent's row — and free-text samples render in randomized order so even created_at can't leak submit sequence. CSV export omits respondent columns entirely — not blanked, removed — so "anonymous" holds up when your DPO reads the schema.