Technology
The stack behind the clarity.
Most agencies rent their reporting. We built ours — the ingest pipelines, the warehouse model and the reporting layer, from the Amazon Ads API and SP-API through to the dashboard your board reads.
That means every number has a lineage: where it came from, when it was restated, and which attribution model produced it. It also means it can run on our infrastructure or inside yours.
Tech agnostic
Our stack, or yours.
The pipeline above is ours, and most brands run on it. It is not a condition of working together. If you already have a warehouse, a BI layer or a data team, we build into what you have rather than around it.
Bring your own tech (BYOT)
Already on Snowflake, BigQuery or Databricks, reporting in Looker, Power BI or Tableau? We deploy the same ingest and models into your environment. Your warehouse, your credentials, your cost line.
Or use ours
No infrastructure to stand up and nobody to hire. We run ingest, storage and the reporting layer, and you get the dashboard, the AMC work and the audience activation on top of it.
Usually it is both
Most set-ups are part-built — solid retail data but no ads pipeline, or reporting that stops at last click. We build the missing piece and leave the parts that already work alone.
Nothing is locked in
Models, SQL and documentation are handed over as part of the engagement. Take the work in-house or move agency and none of it stays behind.
Data enhancement
Capture. Refine. Model.
Our pipeline takes raw data, adds context at every step, and enhances it for analytics and AI. Three steps, each with one job — and none of them overwrite the step beneath.
Capture
Raw · Source of truth
Every record exactly as Amazon returned it — Sponsored Products, Display, Brands and DSP from the Ads API, plus retail data from Seller and Vendor Central via SP-API. Landed continuously by pipelines we built and run ourselves. Nothing transformed, nothing dropped, nothing overwritten. This is the source of truth: if a number is ever questioned, it is settled here.
Refine
Deduplicated · Conformed
Dynamic Tables deduplicate on arrival, type and conform each source, and track the state of every record — so late corrections land cleanly and history stays consistent without being rewritten.
Model
Modelled for questions
A star schema built around the questions people actually ask: category, brand, price band, audience and channel, joined to the spend that produced them. This is the step Looker Studio, AMC work and any AI agent queries.
Context & enrichment
Raw data doesn't answer questions.
Ad platforms hand you IDs. The modelled data joins those IDs to the things you actually plan against, so any number can be sliced by category, segment or channel without exporting it to a spreadsheet first.
Amazon Marketing Cloud
- Audience cohorts
- New-to-brand
- Path to conversion
- Frequency & overlap
- Incrementality
Retail
- Category & subcategory
- Brand & parent ASIN
- Price band
- Availability & OOS
- Buy Box state
- Review volume
Ad media
- Campaign & ad group
- Targeting & match type
- Placement
- Creative
- SP · SD · SB · DSP
Every dimension above is available as a filter, a grouping or a segment — on the same row as the spend that produced it.
Why it matters
What good foundations unlock.
None of this is engineering for its own sake. Each layer removes a specific reason a decision would otherwise be made on a guess.
Plan at category level, not account level
ACoS averaged across an account hides the categories subsidising the rest. With category, brand and price band on the same row as spend, budget follows margin instead of averages.
Numbers that don't move after you've sent them
Amazon revises historical figures after the fact. The pipeline reconciles those revisions as they arrive, so last month's number doesn't quietly change underneath a board deck you already sent.
Stop paying for what you can't sell
Availability and Buy Box state sit alongside campaign spend, so out-of-stock and lost-Buy-Box ASINs can be suppressed rather than discovered at month end.
Data an agent can actually query
A conformed star schema with unambiguous column names is what makes AI answers reliable. Duplicate rows and mystery fields are how models get confidently wrong.
Measurement & attribution
Every number, traceable.
Attribution is where most reporting quietly falls apart. These are the four places we refuse to hand-wave.
Amazon Marketing Cloud
Clean-room audiences and new-to-brand. AMC is the single source of truth for NTB — it is deliberately excluded from the Sponsored Ads tables.
AAT & CAPI
Server-side conversions with dedup_id, so an order is counted once whether it arrives from the pixel or the API.
Consent
ISO 3166 country codes carried alongside TCF, GPP and ACS signals, so measurement stays GDPR and PECR compliant.
Attribution models
View-through and click-through kept apart, never blended. Every figure ships with the model and window that produced it.
Shopify
See which Amazon ads sell your own store.
BASE Conversions is a Shopify app. Install it, and every order on your own store is reported back to Amazon Ads — so you can finally see which Amazon campaigns drove sales that happened on your site, not just on Amazon. Setup is two fields and no developer.
Tracks what happens on your site
A small piece of code (the Amazon Ad Tag) records when someone lands on your store, adds to basket and buys — so Amazon Ads can see the sale it created.
Counts every sale once
Orders are also sent straight from Shopify's servers (the Conversions API), which catches sales the browser misses. The two methods are matched so nothing is double-counted.
No customer data is shared
Matching happens inside a secure environment (AWS Clean Rooms). Names, emails and addresses never move between Shopify, Amazon or BASE.
Proves what upper-funnel spend is worth
Streaming TV and DSP get judged on the revenue they drive on your own store, not only on what converted inside Amazon.