The Ultimate Intelligence Engine for Customer Experience Excellence
Enterprise user journey analytics — rebuilt on a ClickHouse medallion, with AI that shows its work.
Version 2.0 is a ground-up rebuild of the platform's foundation — keeping every analytics capability customers rely on, and re-architecting everything underneath it.
Events, sessions and configuration all live in ClickHouse — a bronze/silver/gold medallion plus a versioned config database.
The stitched-together module frontends become a single React workspace — one sidebar, one design system, route-level code splitting, every page a lazy chunk.
The 22 v1 analytics services are kept and proxied — not rewritten — while high-traffic reads are re-implemented natively over silver and gold for speed.
Every event and session lands 1:1 from RudderStack — 48-column events, 101-column sessions — partitioned monthly with an 18-month TTL.
Cleaned, deduped and enriched views of the same truth — the layer discovery algorithms read, with a 13-month TTL.
Streaming materialized views, dbt-built models and job-written discovery marts — the layer the workspace reads, fast.
Materialized views aggregate on insert, so headline KPIs are current the moment events arrive.
Funnels are computed at query time over sessions — change the steps and re-ask, no rebuild required.
Every read carries an explicit data-source header — gold, silver or sample — so a demo never masquerades as production data.
Role-aware, data-driven, and consistent — descriptor-driven configuration screens, shared Zod contracts between client and server, and a design system carried across every page.
The platform finds what you didn't think to ask.
Window-based funnel analysis computed in the database, journey and path analysis, and channel breakdowns.
Friction points, page performance, navigation analysis and error tracking — the "why" behind the drop-off.
Bot scoring, suspicious-session flags and traffic trust tiers keep automated noise out of your conversion numbers.
Overview KPIs, heatmaps, cohorts, realtime view, engagement and user scoring, churn-risk — a full engagement suite.
ML clustering over behaviour builds personas and segments, with per-persona paths and journeys.
Attribution modelling marts connect campaigns and channels to the conversions they actually produced.
Every number traceable to a row.
Streaming conversations grounded in a live business context — discovery summary, top signals, KPI-vs-target and open recommendations.
The copilot answers through read-only analytical tools — KPIs, funnels, signals, patterns, drop-offs and insight search — not free-form database access.
When it does query, it's a single SELECT against an allowlist with enforced limits, read-only execution and short timeouts — tested against injection corpora.
Multi-provider models (OpenAI and Anthropic Claude) behind one seam, per-request usage accounting, and an enforced budget gate.
Any event category can be flagged as a business conversion goal — with one designated primary goal per deployment.
Declared KPI targets ground the AI layer, so "good" and "behind" are your definitions — not generic best practice.
"Converted" always means the primary goal was achieved — in dashboards, funnels, personas and prompts alike. Per-goal conversion rates are computed query-time.
Grounded in your goals, not generic best practice.
Users and roles, compliance frameworks and rules, privacy controls and checks, audit trails, a data dictionary and security assessment — in the product, not a binder.
Connector and AI keys are write-only: sealed with authenticated encryption, surfaced only as masked hints, and never selectable at the SQL layer.
Platform telemetry is first-party and dependency-free — no cookies, no third-party trackers, no query strings, and Do-Not-Track honored by default.
Each customer runs their own frontend, gateway and ClickHouse — full data isolation by architecture. Operators manage the whole fleet from a separate Platform Console that is never in a tenant's serving path.
The console provisions a tenant's entire data lake end-to-end — schema, config corpus, initial users — with a live step timeline.
Deterministic, seeded demo traffic — from 1k to 250k sessions — shaped to the tenant's own event vocabulary and goals, and purgeable to the row.
Instances pull their resolved configuration outbound with a scoped sync token — the control plane never needs inbound access to customer environments.
Ingest → refine → discover → assess → recommend → act → measure. Web Plateau 2.0 closes the loop between what your customers do and what your team does next.