Zunou has data.
It has no dashboards.
Postgres has the events. CloudWatch has the metrics. The agent has the logs. Nobody has a dashboard. The GTM is gated on the magic number — and we can't read it. Five small moves close that gap in a week, with JP residency intact.
- The GTM is gated on a magic number we can't currently read. The 5/1/3-in-14-days metric needs an event log + a cohort dashboard — neither exists today.
- Pick: PostHog Cloud US for product analytics (free to 1M events/month) with a strict no-PII event taxonomy — hashed user IDs, no message bodies. Metabase for SQL dashboards, Grafana for ops. JP residency stays clean because nothing identifying leaves the country.
- Five dashboards, one week. Activation funnel · Notification SLO · Per-user inference cost · Per-tenant health · Engineering health. Each maps to a specific GTM or Foundation goal.
- Use PostHog Cloud US to start; self-host on Tokyo only if event payloads need PII later. Free tier covers pilot volume. Saves ~$200/mo + operational burden of running our own Postgres/ClickHouse stack.
- Founder owns the activation + reliability dashboards. Walked every Monday. Dashboards no one walks become noise; single owner per dashboard.
Ten metrics. Each tied to a GTM commitment.
These are the readings we'd be embarrassed not to have at the stage-gate decision in February 2027. Build for these first; resist anything fancier.
| Metric | Why it matters | Category | Source | GTM |
|---|---|---|---|---|
| Magic-number activation | Percent of new users hitting 5 events + 1 magic moment + 3 returns within 14 days. The single number the GTM is gated on. Framework adapted from PostHog's activation metrics and Reforge's Setup → Aha → Habit model. | Activation | Postgres event log + cohort table | §05 |
| Time to first event | Median seconds from sign-in complete → first user-initiated event. Detects sign-in / onboarding regressions immediately. | Activation | Sign-in mutation + first GraphQL action | §06 |
| Onboarding funnel by phase | Drop-off per phase (welcome → org → calendar → tabs → layout → agent → complete). Surface where users abandon. | Activation | Nova onboardingStore + completePhase mutation | §06 |
| WAU past magic number (NSM) | Weekly active users who have crossed the magic-number threshold, scoped per partner community. | Retention | Cohort scope per organization_id + activity_log | §05 |
| D7 / D30 retention by cohort | Standard cohort retention curves split by acquisition channel: Indelible warm circle, Northface / Headline, TAI, AI Tinkerers, Spaces event, organic. | Retention | user.created_at + last_seen_at + acquisition_source | §06 |
| Free → paid conversion % | Per pricing tier × per community. Drives the ¥30M+ year-1 ARR ambition vs working hypothesis. | Conversion | Subscription state + organization_id | Decision #14 |
| Per-user inference cost | Token spend / paid-user / month. Required to validate the unit economics before locking pricing. | Conversion | ai-proxy logs (needs per-tenant accounting — see Foundation Stage 3) | §09 |
| Notification delivery SLO | Percent of @mentions delivered within 5s, last 24h. The single number the founder checks each morning. Depends on the Tier-1 outbox + idempotency work shipping first. | Reliability | notification_outbox + processed_events (needs Foundation Tier 1) | Posture |
| Sign-in success rate | Percent of sign-in attempts that complete without re-auth or error. Native iOS vs PWA split — see Journey 1 for the bug catalog. | Reliability | Auth0 logs + signInUser mutation outcomes | Posture |
| Per-tenant feature usage | Adoption of voice agent, calendar, tasks, channels per organization. Tells us what to demo. | Reliability | Activity events tagged by feature_id | §06 |
Three tools, no lock-in, JP residency intact.
PostHog for product analytics. Metabase for ad-hoc SQL. Grafana for operational metrics. All three self-hostable on AWS Tokyo. Mixpanel and Amplitude listed for completeness — they'd cost more and forfeit residency.
PostHog
Primary pick Product analytics + session replay + feature flagsBest for: All product event analytics. Funnels, cohorts, retention, replay. PostHog Cloud US is the simplest start — APPI doesn't bite if events carry no PII (hashed user ID, event names, counts only — not message bodies, names, or emails). PostHog's <a href="https://posthog.com/docs/privacy/data-collection" target="_blank" rel="noopener noreferrer" class="underline decoration-[var(--color-accent)]/40 hover:decoration-[var(--color-accent)]">privacy docs</a> recommend exactly this discipline.
Caveat: Self-host on AWS Tokyo only if event payloads need PII (session-replay of UI showing customer data, support transcripts, etc.). Cloud has no APAC region — self-host is the JP-residency option.
Metabase
Complement BI / SQL dashboards over PostgresBest for: Operational dashboards on top of the existing Postgres. Notification SLO, per-tenant usage, ops runbooks.
Caveat: Not built for product event analytics — pair with PostHog.
Mixpanel
Product analyticsBest for: Mature product analytics. Strong cohort + funnel analysis.
Caveat: EU (Netherlands) and India (Mumbai) residency available; no Japan/Singapore/broader APAC region. Vendor lock-in concerns. Cost grows fast at scale.
Amplitude
Product analyticsBest for: Marketing-leaning analytics, strong dashboards.
Caveat: Pricing escalates aggressively post-free-tier. No JP residency.
Grafana
Complement Operational metrics + SLO dashboardsBest for: CloudWatch / Lambda metric dashboards, alerting.
Caveat: Operational, not product. Required for the notification SLO and Lambda health (Foundation Tier 1).
Build these five. Skip everything else.
Each dashboard answers exactly one stakeholder's recurring question. No vanity tiles. No metrics we won't act on.
Activation funnel · the GTM dashboard
PostHog · for Founders + GTM weekly review
Why: This is the dashboard the GTM page describes (§07.7 weekly KPI tiles). Without it, the stage-gate Feb 2027 decision can't be made on data.
Notification delivery · the reliability dashboard
Grafana on top of Postgres + CloudWatch · for Founder daily check + on-call
Why: The single number that tells you whether the Foundation Tier 1 work is holding. Founder checks this each morning before standup.
Per-user inference cost · the unit-economics dashboard
Metabase on Postgres + ai-proxy log table · for Pricing + finance + Marco
Why: Required to validate the working-hypothesis pricing in §09 panel B before locking it.
Customer-team health · the per-tenant dashboard
PostHog + per-organization filter · for Pilot management + sales
Why: Tells us which pilot customers are about to churn and which are becoming reference accounts.
Engineering health · the eng dashboard
Grafana + Sentry + GitHub Actions · for Engineering team
Why: The Quality page's Definition of Done becomes measurable. Catches regressions before customers do.
One week from zero to first weekly review.
Five steps. Each links to the canonical doc. None of this is invention; it's wiring.
Sign up for PostHog Cloud US (free tier)
Free tier covers 1M events/month — comfortably above pilot volume. Adopt a strict no-PII event taxonomy: identify users by hashed Auth0 sub, send event names + numeric/enum properties only. No message bodies, names, or emails. Defer the Tokyo self-host migration until a customer or auditor specifically asks.
Add the JS + iOS + Android SDKs to Nova and Dashboard
Standard `posthog-js` for the web dashboard; `posthog-react-native` for Nova; identify users by Auth0 sub on sign-in. Auto-capture page views; manual capture for the events that map to magic-number actions.
Mirror the GTM weekly KPI tiles as PostHog dashboard
The mock dashboard at GTM §07.7 is the spec. Build it as a PostHog dashboard, share the URL with the team, walk it every Monday.
Stand up Grafana for Lambda + Postgres metrics
Grafana Cloud free tier (Tokyo region) or self-host. Connect CloudWatch and the Postgres replicas. Build the notification delivery SLO panel as the first dashboard — it doubles as Foundation Tier 1 acceptance criteria.
Stand up Metabase for SQL ad-hoc + Notification SLO ops
Self-host on the same Tokyo cluster. Read-only role on the Postgres replica. First dashboard: per-tenant feature usage. Second: per-user inference cost.
Four calls before standing this up.
PostHog Cloud vs self-host
Recommendation: Cloud US — provided the event taxonomy stays PII-free (hashed user IDs, no message bodies). Saves the infra cost and operational burden. Migrate to self-host only when an enterprise customer or auditor explicitly asks.
Why it matters: Without the no-PII discipline, this becomes an APPI exposure. With it, Cloud US is the right call.
Event taxonomy
Recommendation: Lock the magic-number events first (signed_in, first_event, magic_moment, returned), iterate the rest.
Why it matters: Renaming events later breaks the historical funnel — but waiting too long means no data.
Owner of the dashboard cadence
Recommendation: Founder owns the Activation + Reliability dashboards. Eng owns Engineering Health.
Why it matters: Dashboards no one walks become noise. Single owner per dashboard.
Session replay on / off for pilots
Recommendation: On for pilot tenants with explicit consent baked into onboarding ToS. Off for everyone else.
Why it matters: APPI-relevant — needs explicit user notice. Worth it for pilot debugging value.
Where this gets us
A Monday-morning ritual the founder doesn't dread.
One week of work. One PostHog instance. Five dashboards. The GTM stops being a doc and starts being a number on a wall — checked, contested, and acted on every Monday.
Without dashboards, the magic number is a slogan. With them, it's a contract.