The marketing data lake
for intelligent marketing.
A marketing-specific data foundation, built in your BigQuery environment, that connects customer, sales, campaign, audience, creative and response data into a persistent marketing memory.
Identity. Measurement.
Campaign memory.Built in your BigQuery environment.
AI needs something
reliable to reason over.
Connecting systems does not, on its own, create shared definitions, history or consistency. Ask the same question against four tools and you can get four totals. The foundation gives every agent the same customers, the same definition of a sale and the same record of what was tried before.
Access helps agents reach systems. A shared foundation gives them consistent context.
Built around how
marketing actually works.
Not a generic warehouse schema. The foundation organizes the entities marketing decisions depend on, and the relationships between them.
- Customer
- People, identities, behavior and eligibility.
- Commerce
- Transactions, products, SKUs, categories and revenue.
- Marketing
- Campaigns, investment, exposure and response.
- Audiences
- Criteria, segments, versions and destinations.
- Creative
- Assets, campaign context and links to products.
- Measurement
- Conversions, experiments, attribution and models.
- Memory
- Decisions, execution, results and what was learned.
A conceptual model. Which entities and modules are deployed depends on each client’s data and scope.
A store records 1,000 MXN in settled gross sales and 200 MXN in refunds. Net sales are 800 MXN, and every agent uses that definition, or makes a different one explicit.
One reporting period and currency · taxes and shipping excluded in this example.
Connect sales
with marketing.
Analyze response by brand, SKU, category, region and customer when the data has enough granularity. Where a link is not supported, it stays unknown instead of being guessed.
- Campaign
- Exposure / response
- Identity, where supported
- Customer
- Transaction
- Product / SKU
- Revenue
Not every exposure can be matched to every sale. Coverage and unmatched records are reported with the result.
One person, where the evidence supports it.
Records are linked when there is a supported, permitted correspondence between them, such as a shared customer ID. A record without that evidence remains a separate record.
Unknown remains unknown.
Different questions.
Different methods.
The foundation keeps the inputs each method needs. Nexus recognizes which question is being asked and says what the answer can and cannot support.
| Discipline | The question | The limit |
|---|---|---|
| Attribution | What appears to contribute? | Credit across observed touchpoints is not causality. |
| Incrementality | What changed because of marketing? | Needs an appropriate design: a comparable control or holdout. |
Marketing Mix Modeling Meridian | How do channels contribute over time? | Depends on the data and the fit of the model. Workflows built with Google’s Meridian framework. |
| Experimentation | What did the test change? | Population, control and the metric are defined before the test. |
| Forecasting, where available | What might happen next? | Expressed as a range with uncertainty, never as a certainty. |
How the methods differ
Attribution shares credit for conversions already observed, using rules or a data-driven model. It is useful for day-to-day optimization and says nothing about what would have happened without the advertising.
Incrementality and experiments compare a group that received the marketing with a comparable group that did not. They answer causal questions, within the population and period tested.
Marketing mix modeling relates sales to marketing activity over time from aggregate data, accounting for other factors and stated assumptions. Random Truffle runs MMM workflows with Meridian, Google’s open-source framework; results depend on the data available and on how well the model fits.
Examples on this site use illustrative, synthetic data. They are not a fitted client model.
Brand Hub.
The context behind the work.
Data Foundation connects what happened. Brand Hub adds the context needed to interpret it and to produce consistent work.
- Audience + objectiveDefined with Aura from the foundation.
- PrismCreates the brief and the variants.
- Brand HubAdds the product, approved claims and brand rules.
- Campaign assetReturns to the system, linked to its audience and product.
Structured product data can live in the foundation. Brand Hub adds context and rules on top of it; it is not a duplicate copy.
Change tools.
Keep your marketing memory.
Agencies, media accounts, automation platforms and analytics tools change. The history of what you did and what it produced stays in your own infrastructure.
- Agency A
- Google Ads · Meta
- Email platform
- Analytics tool
Same history. Same definitions.
- Agency B
- Google Ads · Meta · TikTok
- Marketing automation platform
- Analytics tool
Changing a tool still takes technical work to connect it. What you keep is the history and the definitions, not an instant migration.
Your cloud.
Your data.
Your rules.
Your history.
The marketing data lake lives in your BigQuery environment, and it stays with you if you stop working with Random Truffle.
Ownership of your data is separate from the licence for Random Truffle’s software and services, which follows your contract.
Built in BigQuery,
on Google Cloud.
The foundation is deployed in your BigQuery environment. Agent workflows use Gemini models where your deployment is configured for them. Other clouds and warehouses can be connected as sources; they are not equivalent deployment targets.
See how your tools connect →Questions worth asking.
Does this replace our warehouse or CRM?
No. The foundation is built in BigQuery and connects the context those systems hold; your systems of record keep doing their job. Other warehouses can feed it as sources.
Who owns the data?
You do. The marketing data lake is built in your BigQuery environment and stays there if you stop working with Random Truffle. The licence for our software and services is a separate matter, defined in your contract.
Does identity resolution make attribution causal?
No. Linking eligible records improves the observable journey. It does not establish what would have happened without the advertising. A causal question needs an appropriate comparison or experiment.
Is this a CDP?
Not a generic one. It is a marketing-specific data lake: customers, sales, campaigns, audiences, creative and response, with the business definitions and history that connect them.
What can I explore on this site?
This page describes the architecture. The agent conversations use example data to show identity, audience rules, model assumptions and a decision record.
Start with a better question.
Ask how a business definition, an audience or a model would travel through the foundation.