Last-click
01Credits the final touchpoint before conversion. Default, predictable, simple to audit.
When to use · Direct response, paid-search heavy mix
Module 02 · Revenue Attribution · live
Multi-touch attribution. Closed in under sixty seconds.
Seven attribution models running in parallel. Eight channels rolled up into one view. Cohort retention curves, predictive LTV, real-time webhook export. Built on the same event bus as Sumeru's automation and orchestration engines — so attribution doesn't just tell you what happened, it triggers what happens next.
P95 latency
0 s
event → revenue
Models
0
parallel + blended
Channels
0
unified rollup
Retention
0 mo
cohort curves
Numbers from production — trailing-90-day average across active deployments.
Most attribution platforms were built for a slower era.
14-day batch jobs. Static last-click logic. Lossy hand-offs between tools. By the time you read the report, the campaign is already over.
Latency
The data is always 4 hours late.
Triple Whale: 24-48h. Northbeam: 4-12h. Klaviyo native: 1-7d. By the time you optimize, the budget is spent and the auction has moved on.
Single model
One model, picked once, never compared.
Most platforms force you to pick last-click at setup. The model never gets revisited. The truth is between models — but you only see one.
No closed loops
Reports, not actions.
Attribution tools tell you what happened. They don't act on it. You read the dashboard, copy the number into another tool, then act manually. Sumeru closes that loop.
Event in. Attribution closed. Action ready.
Every order, refund, and cart event flows through the same four-step pipeline.
01 · Ingest
Event captured
Pixel + webhook + GSC + ad APIs. Tagged with shop, channel, traceId.
02 · Resolve
Identity stitched
Cookie + CRM + email match. Cross-device through Shopify customer ID.
03 · Attribute
Models compute
Seven models run in parallel. Spread surfaced for any conversion in real time.
04 · Close
Loop ready
Webhook fires to Automation Engine. Bid adjustment queued. Audit row written.
Six sources. One engine. Four outputs.
The exact runtime topology — every event flows through these stages, in this order, every time. Hover to inspect.
Attribution · Pipeline
ingest · resolve · model · rollup
P95 · 42s · 12,847 events/min · live
6 sources · 3 stages · 7 models in parallel · 4 outputs all event-driven · all idempotent · all audited
Diagram is illustrative — production deployments add per-shop sharding, dead-letter handling, and retry policy per source.
Seven approaches. Every one runs in parallel.
Each model has its right context. Sumeru doesn't force you to pick — it shows you the spread and lets you blend.
First-click
02Credits the first touchpoint that introduced the customer to your brand.
When to use · Brand campaigns, top-of-funnel evaluation
Linear
03Equal credit across every touchpoint in the customer's path to conversion.
When to use · Long consideration cycles, multi-channel mix
Time-decay
04Recent touchpoints get more credit. Earlier ones decay exponentially.
When to use · Promo cycles, time-sensitive campaigns
Position-based
0540% to first touch + 40% to last + 20% spread across the middle (U-shape).
When to use · Balanced first-impression + closing weight
Custom-weighted
06Define your own per-channel weights. Persisted per campaign.
When to use · Channel-specific economics, agency POV
Model-blended
07All seven models run in parallel. The spread is surfaced — and a recommendation when a single number is needed.
When to use · When you want truth, not a single number
Three concrete moves customers made.
Anonymised but representative. Pulled from active deployments, normalised across shops.
62%
tooling spend cut
Replaced Triple Whale + Klaviyo dashboards + Looker
An $8M GMV apparel brand consolidated their attribution stack onto Sumeru. Weekly attribution review went from 2 hours across 4 dashboards to 8 minutes in one place. Saved $1,800/mo in tool licenses; the operator hours saved were worth more.
$847k
annualised re-allocation
Surfaced the paid-vs-organic discount
An agency running 14 Shopify Plus accounts found that 23% of conversions Triple Whale credited to paid social had non-trivial organic-search assists. The model-blended view re-allocated $847k of paid budget toward higher-leverage channels.
<60s
decision latency
Closed-loop bid response
When a Google Ads campaign starts hemorrhaging ROAS, the attribution engine flags it within 60 seconds and routes a proposed bid reduction into the Automation Engine — gated by the 7-day mandatory dry-run before any paid-spend mutation goes live.
What's available where.
The full stack is included in every paid tier — only the depth, latency, and export surfaces differ.
| Capability | Starter | Pro | Agency | Enterprise |
|---|---|---|---|---|
| Last-click attribution | ✓ | ✓ | ✓ | ✓ |
| First-click attribution | — | ✓ | ✓ | ✓ |
| Linear · Time-decay · Position-based | — | ✓ | ✓ | ✓ |
| Custom-weighted | — | — | ✓ | ✓ |
| Model-blended view | — | — | ✓ | ✓ |
| Cohort retention curves (12-month) | — | ✓ | ✓ | ✓ |
| Predictive LTV (gamma-gamma + BG/NBD) | — | — | ✓ | ✓ |
| Cross-channel rollup (8 channels) | ✓ | ✓ | ✓ | ✓ |
| Webhook export of attribution events | — | ✓ | ✓ | ✓ |
| BigQuery / Snowflake bulk export | — | — | — | ✓ |
| P95 latency target (event → revenue) | 120s | 60s | 60s | 30s |
Attribution depends on real data.
The runtime ingests from these sources and exports to these surfaces. All connections are typed, throttled, and idempotent.
Shopify
orders · customers · refunds
Google Ads
campaign metrics + conversions upload
Meta Ads
CAPI conversions · audience sync
TikTok Ads
Events API conversions
GA4
event-stream + channel rollup
Search Console
organic query attribution
WhatsApp
outbound campaign tracking
Public API
webhook export · BigQuery sync
30-minute call
See attribution in your own data.
We sketch your funnel, your channel mix, and where current attribution is failing you. You leave with a deployment plan and a price range — in writing.