1. Home
  2. How it works

How it works

From scattered ad data to AI that acts.

Six steps, run on a schedule. We can build all six, or start with the one that hurts most and add the rest later.

Walk through the steps Talk to an engineer

Every scheduled run

  1. 01Pullevery platform, on schedule
  2. 02Storelands in your warehouse
  3. 03Cleantests passing
  4. 04Dashboardsrefreshed by 6:00 am
  5. 05AI insights3 alerts written
  6. 06AI actionsinside your caps

The six steps

What happens to your ad data on every run.

Pick a step to see it run. Each one works on its own, so you can start anywhere.

Your warehouse
Scheduled · every 15 min to daily✦✦

Step 01 of 06 · Pull

Every platform, pulled automatically.

Scheduled API jobs pull Amazon, Walmart, Google, Meta and TikTok, your Shopify store and your attribution tool. No CSV exports, no stale snapshots.

What you get

  • Scheduled pulls, from every 15 minutes to daily
  • Retries and backfills built in
  • Alerts when a platform changes a report

Typical tools

  • Amazon Ads API
  • SP-API
  • Walmart Connect API
  • Google Ads API
  • Meta Marketing API
  • TikTok Ads API
  • Airflow
  • Dagster
  • Airbyte or Fivetran
See every platform we pull
Your warehouseSnowflakeBigQueryRedshift
Reporting-readyfct_ad_spend_daily
  • Spend$48.2k
  • Revenue$127.7k
  • ROAS2.65
Cleanedstg_campaigns
  • ✓ One schema
  • ✓ USD
  • ✓ UTC
  • ✓ Deduped
Raw, kept foreverraw_ads_*
  • amazon_ads.json
  • meta_ads.csv
  • walmart.json
  • shopify_orders
🔒 Your cloud · your account · your keys

Step 02 of 06 · Store

One place, and it's yours.

Everything lands in your own Snowflake, BigQuery or Redshift, in your cloud. Raw data is kept, so history can be rebuilt any time.

What you get

  • A warehouse in your own cloud account
  • Raw data kept, so history can be rebuilt any time
  • Raw, cleaned and reporting-ready layers kept apart

Typical tools

  • Snowflake
  • BigQuery
  • Redshift
  • AWS
  • GCP
  • Azure
  • Terraform
Marketing data warehouse
03/14/26£7401.2knull
→
2026-03-14USD 926.3012000
✓ tests passing

Step 03 of 06 · Clean

Messy in, clean out.

Columns, currencies, time zones and attribution windows lined up into one model, with one ROAS rule you agree on. Tests catch what people miss.

What you get

  • One model across every platform
  • One ROAS rule you agree on
  • Currencies, time zones and attribution windows lined up
  • Automated tests on every run

Typical tools

  • dbt
  • SQL
  • Python
  • dbt tests
See how fields are mapped
PlatformAccountCampaignProductRetailerRegion
Blended ROAS by retailer
ROAS 2.65Refreshed 6:00 am

Step 04 of 06 · Dashboards

Reports that run themselves.

Dashboards and client reports refresh on their own in Looker, Looker Studio, Power BI or Tableau, white-labeled for agencies.

What you get

  • Dashboards that refresh themselves
  • White-label client reports for agencies
  • Every view reads the same numbers

Typical tools

  • Looker
  • Looker Studio
  • Power BI
  • Tableau
  • Metabase
Agency reporting automation
AI
Meta ROAS −8% · frequency rising
Google CPC spike · 12 keywords
Amazon ACOS above target · 3 ASINs

Step 05 of 06 · AI insights

AI that spots it first.

AI flags drops, waste and odd spikes before anyone opens a dashboard, writes the summary, and answers questions in plain English.

What you get

  • Alerts on drops, waste and odd spikes
  • A written summary of what changed
  • Plain-English answers from your own numbers

Typical tools

  • Anomaly detection
  • LLM summaries
  • Semantic layer
The AI layers on our Technology page
Your cap
↓ Bid −12%within your cap ✓
↑ Budget +$240/daywithin your cap ✓
⏸ Pause ad clusterneeds your OK
Every change logged · roll back anytime

Step 06 of 06 · AI actions

AI that acts, within your rules.

Agents move bids and budgets and pause waste inside caps and approval queues you set. Every change is logged and can be rolled back.

What you get

  • Bid and budget changes inside your caps
  • An approval queue for anything outside them
  • Every change logged and reversible

Typical tools

  • Rules engine
  • Platform write APIs
  • Approval queues
AI bid & budget management

Your first weeks

From first call to a live dashboard.

Every build is quoted on its own scope. Multi-platform builds with automation typically run 8–14 weeks, depending on sources, history and reporting needs.

  1. Week 1

    A free 30-minute call

    We learn your setup: the accounts you run, where the numbers stop matching, and what hurts most.

  2. Weeks 1–2

    Written scope and a fixed quote

    A written scope with a fixed price, agreed before any work starts.

  3. Weeks 2–6

    First pipeline and dashboard live

    We build your highest-spend source end to end first, so one channel is trustworthy quickly. A focused single-source pipeline usually ships in 4–6 weeks.

  4. After launch

    A run plan, or a handover

    A monthly plan where we run and fix it, or a handover with docs, runbooks and a walkthrough. Code and data stay yours either way.

  5. Then

    More sources, then AI on top

    Remaining platforms are added against the proven pattern. Once the data is trusted, AI insights and agents go on top, inside rules you set.

Guardrails

AI that acts only where you let it.

You choose how much control to hand over, and you can take it back any time. Try the switch.

  • Approve every change, or approve the rules once and let the agent act inside them.
  • Bid caps, budget limits and which accounts are in scope are set by you.
  • Every change is logged with the before and after, and can be rolled back.
  • Agents stop if the data feed is late or looks wrong.
  • Start with simple rules; add AI judgment once you trust it.

Recommend only

The agent drafts each change and explains why. Nothing touches your ad accounts until a person makes the change.

  • ↓ Bid −12%Amazon · Sponsored Products keyword
    Suggested · your team decides
  • ↑ Budget +$240/dayMeta · prospecting campaign
    Suggested · your team decides
  • ⏸ Pause ad groupTikTok · below your ROAS floor
    Suggested · your team decides

Approve each change

The agent prepares each change and holds it in a queue. A named approver says yes or no.

  • ↓ Bid −12%Amazon · Sponsored Products keyword
    Waiting for approval
  • ↑ Budget +$240/dayMeta · prospecting campaign
    Waiting for approval
  • ⏸ Pause ad groupTikTok · below your ROAS floor
    Waiting for approval

Act within caps

You approve the rules once. Changes inside your caps go ahead; anything outside them waits for you.

  • ↓ Bid −12%Amazon · Sponsored Products keyword
    Applied · inside your cap
  • ↑ Budget +$240/dayMeta · prospecting campaign
    Applied · inside your cap
  • ⏸ Pause ad groupTikTok · below your ROAS floor
    Outside your cap · needs your OK

Every change logged · roll back any time

Custom work

Built around what your team needs.

The six steps give you clean, trusted data. Most clients then ask for something specific on top of it.

01

Custom metrics

Margin-adjusted ROAS, contribution margin by channel, TACoS by brand, new-to-brand by retailer. Whatever your finance team actually uses.

02

Data joined to your business

Inventory, margins, promotions, CRM and offline sales next to ad spend, for inventory-aware pacing and real profit.

03

Apps and tools on your data

Internal tools, client portals and product features, like the ad platform and creative studio we engineered for Pixis.

Tell us what you need built

For your tech team

What we run in production.

Proven tools, deployed in your cloud and owned by you. Code, docs and runbooks are yours.

  • Sources
    • Amazon Ads API
    • SP-API
    • AMC
    • Walmart Connect API
    • Shopify Admin API
    • Google Ads API
    • Meta Marketing API
    • TikTok Ads API
    • attribution platforms
    • Instacart, Criteo and other retail media APIs
  • Ingestion
    • Custom API pipelines for the hard sources
    • Airbyte or Fivetran where they already work well
  • Orchestration
    • Airflow
    • Dagster
  • Warehouse
    • Snowflake
    • BigQuery
    • Redshift
  • Transformation
    • dbt
    • SQL
    • Python
  • AI
    • Rules engine for agent actions
    • Anomaly detection
    • LLM summaries and plain-English questions over a semantic layer
  • BI
    • Looker
    • Looker Studio
    • Power BI
    • Tableau
    • Metabase
  • Cloud
    • AWS
    • GCP
    • Azure
    • Terraform

See the full architecture, layer by layer, on our Technology page

Solutions

Where teams usually start.

AI bid & budget management

Agents that watch spend and move bids and budgets inside your rules.

Learn more

Agency reporting automation

White-label client reports that build themselves.

Learn more

Marketing data warehouse

dbt + Snowflake or BigQuery, modeled, tested and documented.

Learn more

Amazon Marketing Cloud

AMC setup, audiences and custom queries, landed in your warehouse.

Learn more

Conversions API

Server-side tracking for Meta, Google and TikTok, deduplicated.

Learn more

Marketing mix modeling

Measure each channel's contribution beyond click attribution.

Learn more

Start with the step that hurts most.

Often that's one pipeline and one dashboard. A focused single-source pipeline usually ships in 4–6 weeks.

Talk to an engineer