Skip to content
Module 06 · Analytics Intelligence · live

Five sources. One queryable rollup.

Cross-channel rollup with cohort retention, per-customer channel preference, send-time optimisation, and predictive LTV. Built on a multi-tenant data plane with per-shop schema isolation. Replace four Looker dashboards with one SQL surface.

Sources
0
unified rollup
Retention
0 mo
cohort curves
LTV models
0
gamma-gamma + BG/NBD
Tenancy
per-shop
schema-level isolation
The problem

Most analytics is a dashboard.

Reports answer 'what.' Operators need to know 'who, when, and what next.'

No unified rollup

Five tabs, one cohort.

GA4 in one tab, Shopify in another, Google Ads in a third, Looker on top. Reconciling channels is manual. The cohort never converges.

No per-customer model

Send everyone the same thing.

No channel preference, no send-time. Every customer gets the email, every customer gets the WhatsApp. Half feel spammed.

No LTV signal

"Is this customer profitable?"

Retention curve says one thing. CAC says another. Without an LTV model that fuses both, you're guessing on each acquisition.

How it works

Ingest. Cohort. Predict. Surface.

Every event lands on the data plane, gets cohorted, and feeds the LTV models.

01 · Ingest

5 sources, one schema

Shopify orders, GA4 events, Google + Meta + TikTok ad metrics, GSC queries — all routed into a per-shop schema with explicit tenant scoping.

02 · Cohort

First-touch attribution

Customers cohort by first-touch month + acquisition channel + geography. Retention computed nightly out to 12 months.

03 · Predict

LTV + channel + send-time

Gamma-gamma + BG/NBD per cohort. Bayesian per-customer channel preference. Send-time optimisation retrained weekly.

04 · Surface

Query · webhook · export

Live UI for queries. Webhook export of LTV-changed events. BigQuery / Snowflake bulk export at Enterprise tier.

The pipeline

Five sources. One cohort engine. Four surfaces.

The exact runtime topology. Hover any node to inspect.

ANALYTICS · PIPELINE ingest · cohort · predict · surface
Refresh · 5min · 14k events/min · live
shopify orders · refunds · customers ga4 events · sessions · channels google ads campaign metrics · CV upload meta ads CAPI · audience · CV upload search console organic queries · CTR ANALYTICS ENGINE idempotent · audited · multi-tenant 01 · INGEST 02 · COHORT 03 · PREDICT Per-shop schema tenant-scoped · dedupe· schema check· shop tag· trace id 14k/min First-touch model channel · geo · month · cohort assign· retention curve· channel split· tier rollup· weekly refresh 12-mo curves LTV + send-time Bayesian + ML · gamma-gamma· BG/NBD· channel pref· send-time 145 RBAC permissions · 365d audit retention · snapshot + undo on every action live dashboard queryable cohort surface ltv webhook ltv-changed events bigquery export scheduled · enterprise comms pipeline send-time + channel pref
5 sources · 3 stages · 2 LTV models · 4 surfaces
In production

Three concrete moves customers made.

$3.2M
annualised LTV uplift

Predictive LTV at the cohort level

A subscription-commerce brand replaced their Looker LTV dashboards with Sumeru's gamma-gamma + BG/NBD model. The cohort view surfaced a high-LTV channel they were under-investing in. Re-allocation drove $3.2M in annualised LTV uplift over 6 months.

11×
send-time response lift

Channel + send-time per customer

WhatsApp message scheduled at the customer's per-channel preferred time saw 11× the response rate vs. blasted-at-9-AM messages. Per-customer Bayesian model retrained weekly. Wired to the comms-pipeline.

5+
data sources unified

One queryable rollup, every cohort

GA4 + Shopify + Google Ads + Meta + GSC streaming into one rollup. Per-shop schema isolation. Cohorts queryable by acquisition channel, first-touch month, geography, customer tier — all in one SQL surface.

By tier

What's available where.

CapabilityStarterProAgencyEnterprise
Shopify · GA4 · Ads · GSC ingest
Cross-channel revenue rollup
Cohort retention curves (12-mo)
Channel preference (Bayesian per-cust.)
Send-time optimisation
Predictive LTV (gamma-gamma + BG/NBD)
BigQuery / Snowflake export
Model retrain cadencewkwkdaily
30-minute call

See your data in one rollup.

We connect to your sources, build a cohort view in under 10 minutes, and walk through one channel-preference + LTV insight live.