Resources
Marketing Intelligence Learning Hub
Clear answers to the essential questions about marketing data, context, AI readiness, measurement and optimization, decisioning, and agentic marketing.
Explore the concepts shaping modern marketing performance — from data foundations to AI-driven decisions.
46 answers
6 topics
Updated August 2026
Explore by topic
01 — Data
Marketing Data & Infrastructure
Marketing data infrastructure is the foundation that turns fragmented information from across the marketing ecosystem into data that can be consistently understood, analyzed, measured, and used. Explore the fundamentals of marketing data, the platforms that manage it, and how modern infrastructure is evolving to support measurement and AI.
What is marketing data infrastructure?
Marketing data infrastructure is the systems, processes, and data architecture used to collect, connect, standardize, govern, and make marketing data usable across an organization.
Marketing data originates across a highly fragmented ecosystem—from paid media and social platforms to CRM and loyalty programs, websites and apps, sales and commerce systems, research, and offline channels. Each source can use different naming conventions, identifiers, hierarchies, metrics, and levels of granularity. Infrastructure creates a consistent foundation across those sources so marketers aren’t continually reconciling incompatible versions of their data.
Traditionally, the goal of marketing data infrastructure has largely been to make data accessible for reporting, analytics, activation, and measurement. AI raises the standard. Data must not only be available; it needs to be clean, structured, connected, governed, and understandable in the context of the business.
Many organizations have not yet made that transition. While 71% of marketing leaders describe their data as AI-ready, only 37% meet the combined foundational conditions required to operate AI at scale. The gap underscores an important distinction: data availability is not the same as data readiness.
Modern marketing data infrastructure therefore increasingly acts as a foundation for both human and machine decision-making—supporting analytics and measurement while giving AI systems the structured data and context they need to reason reliably.
What is marketing data?
Marketing data is the information generated by and about marketing activity, customer interactions, and the business outcomes marketing is intended to influence. It provides the raw material marketers use to understand audiences and journeys, measure performance, optimize investments, and inform decisions.
Marketing data can include several types of information: media and campaign data such as spend, impressions, reach, and creative; customer and engagement data such as website visits, app activity, CRM interactions, and loyalty behavior; commerce and outcome data such as leads, conversions, transactions, and revenue; and brand and market data such as awareness, consideration, survey research, competitive activity, and market conditions.
That data comes from a wide range of sources, including paid media platforms, earned and owned media, websites and apps, CRM and loyalty systems, sales and commerce platforms, marketing automation tools, research and survey providers, data warehouses, and other enterprise systems. External signals—such as economic conditions, seasonality, weather, or industry data—can also provide important context for understanding why marketing performance changes. The Substrate unifies, enriches, and contextualizes data across paid, earned, and owned media; CRM and loyalty; sales and commerce; websites and apps; research; external signals; data warehouses; and enterprise systems.
The challenge for most organizations is not simply accessing more marketing data. It is making data from these different sources consistent, connected, and meaningful across the business. Different systems often use different identifiers, structures, taxonomies, timeframes, and definitions. Unifying and contextualizing that data creates a usable foundation for measurement, intelligence, optimization, and AI.
What is a marketing data platform?
A marketing data platform is technology designed to collect, integrate, standardize, and manage data from across the marketing ecosystem so it can be used for analysis, measurement, reporting, optimization, and increasingly, AI.
Unlike systems designed primarily around a single function—such as customer profiles, media activation, or business intelligence—a marketing data platform focuses on making fragmented marketing data usable across functions.
That can include connecting data from ad platforms, paid and organic media, CRM and loyalty systems, sales and commerce, websites and apps, research, data warehouses, and other enterprise systems. The platform can then perform work such as normalization, taxonomy alignment, data-quality management, identity or journey resolution, and governance.
The category is evolving as AI changes what downstream systems require. Simply centralizing data does not necessarily make it usable by AI. The question is shifting from “Can we get our marketing data into one place?” to “Can our systems understand what that data means?”
That requires more than connectivity. Marketing data needs consistent definitions, relationships, taxonomies, business rules, and historical context that can persist across analytics tools, measurement methodologies, AI models, and agents. Without that shared context, different systems can access the same underlying data and still interpret or use it differently.
As a result, the role of the marketing data platform is expanding from moving and organizing marketing data toward creating a persistent, AI-ready foundation for measurement, intelligence, and decision-making.
What is the difference between a marketing data platform, a CDP, and a data warehouse?
Marketing data platforms, customer data platforms (CDPs), and data warehouses can all contain marketing-related data, but they are designed to solve different primary problems.
A data warehouse provides scalable infrastructure for storing and querying data from across the enterprise. It can hold large volumes of marketing data alongside finance, product, operations, sales, and other information. But a warehouse does not inherently reconcile the different taxonomies, KPIs, hierarchies, and relationships marketers use.
A CDP is primarily organized around customers. It brings together first-party customer data to create persistent profiles that can support segmentation, personalization, and activation.
A marketing data platform is organized around the broader marketing data environment. It can connect customer data with media, campaign, creative, channel, conversion, sales, research, and other performance data and prepare that information for marketing-specific uses such as measurement, analytics, and optimization.
These technologies aren’t necessarily substitutes. An enterprise may use all three. The more useful distinction is what work needs to happen to the data. Storing enterprise data is different from resolving customer identity, and both are different from creating a consistent, marketing-specific data foundation that can support measurement, intelligence, and AI.
What are data generation and journey reconstruction?
Data generation and journey reconstruction are approaches for creating useful marketing intelligence when the complete customer journey cannot be directly observed.
Traditional person-level attribution depends heavily on observed signals: which ads someone encountered, what actions they took, and whether they ultimately converted. But marketers increasingly cannot observe every exposure or interaction. Privacy restrictions, walled gardens, offline media, cookie loss, and other forms of signal fragmentation have made deterministic customer journeys less complete.
Journey reconstruction addresses this gap by using the signals that are available to estimate missing parts of the marketing journey rather than assuming that unobserved activity did not occur. Data generation goes further: statistical techniques can generate modeled information about exposures and relationships that cannot be directly observed, creating additional usable signal from incomplete data.
How this is accomplished varies by methodology. Marketing Evolution combines aggregate and available user-level signals with statistical modeling to estimate missing exposures and reconstruct statistically representative consumer journeys. This generates usable person-level intelligence from incomplete, privacy-constrained data without relying on cookies or persistent PII.
The distinction is important. Data generation is not simply filling empty cells in a dataset. Applied rigorously, it uses statistical inference to recover information that signal loss has made unavailable, allowing measurement and modeling to work from a more complete representation of the marketing journey.
As privacy and platform fragmentation continue to reduce deterministic journey data, the ability to responsibly reconstruct missing signal becomes increasingly important to measurement.
Why isn’t unified marketing data enough?
Unifying marketing data solves an important problem: it brings information that would otherwise remain scattered across platforms, teams, and systems into a connected environment. But putting data in one place does not automatically make it consistent, meaningful, or ready for measurement and AI.
Two datasets can sit beside each other and still describe campaigns, audiences, channels, products, KPIs, or outcomes differently. A system may know that two fields contain numbers without understanding that one represents media spend, another represents incremental revenue, or how either relates to a particular campaign, audience, objective, or business outcome.
This is why modern marketing infrastructure increasingly needs a semantic layer in addition to a data layer. Shared taxonomies, hierarchies, definitions, relationships, and business rules give the underlying data context.
The consequences of missing that shared foundation become clearer as organizations apply AI across marketing. Only 29% of marketing leaders completely agree that their teams work from the same data. And organizations operating with unified data report substantially stronger AI-driven improvements across audience targeting, creative effectiveness, campaign optimization, and full-funnel measurement than those operating with fragmented data.
The progression, then, isn’t simply fragmented data → unified data. It’s closer to: connected data → clean data → structured data → contextualized data → usable intelligence.
That final step becomes particularly important for AI. Models and agents need access to information, but they also need sufficient context to interpret what that information represents, how different elements relate to one another, and how they connect to the business.
What is a System of Record for Marketing Performance?
A System of Record for Marketing Performance creates a persistent, governed record of how marketing activity connects to business outcomes. Unlike a data warehouse or connector, it does more than store or move data. It unifies, enriches, and contextualizes marketing data so measurement systems, AI models, and agents can interpret it consistently.
What is The Substrate?
The Substrate is Marketing Evolution’s System of Record for Marketing Performance. It unifies, enriches, and contextualizes fragmented marketing data through Data Unification, Domain Intelligence, Business Ontology, and Journey Reconstruction. The Substrate supports measurement, simulation, optimization, monitoring, Darwin, and any model or agent that reasons over marketing data.
Explore further
02 — Context
Marketing Context & Ontology
Marketing data becomes more valuable when systems understand not only what the data contains, but what it means for a specific business, objective, and use case. Marketing context provides that understanding by connecting data to business definitions, relationships, historical patterns, measurement logic, and external factors. Ontologies help structure and preserve that context so it can be consistently used across teams, measurement systems, AI models, and agents.
What is marketing context?
Marketing context is the information that gives marketing data meaning within the specific business and use case in which it is being interpreted. It connects data to an organization’s objectives, definitions, relationships, historical patterns, and external factors so performance can be understood in the right frame.
Marketing data can tell you what happened: how much was spent, which audiences were reached, what creative ran, and which outcomes occurred. Marketing context helps explain what those signals mean for a particular business—how they relate, what matters, and what else may have influenced the outcome.
For example, if paid social performance declines while television investment increases and overall sales remain stable, looking at each channel independently tells only part of the story. Television may be creating demand that later appears through digital channels; the channels may play different roles in that company’s customer journey; or seasonality, pricing, promotions, competitive activity, economic conditions, or other external factors may be influencing demand.
The relevant context also differs by organization. A conversion might mean an ecommerce purchase for one company, a new account for another, and a qualified lead for a third. The same marketing activity can therefore carry very different meaning depending on the business objective, customer journey, market, and decision being made.
Some context can be explicitly encoded through taxonomies, business rules, and ontologies. Other context is learned through measurement and accumulated over time.
This becomes especially important for AI. Data gives AI access to what happened. Context grounds that data in the specific business and use case—helping AI understand what matters, how variables relate, and what may have influenced the outcome.
What is a marketing data ontology?
A marketing data ontology is a structured representation of the concepts within a marketing environment and the relationships between them. It provides a shared framework for understanding how elements such as campaigns, channels, audiences, creative, products, KPIs, conversions, and business outcomes connect.
Where a taxonomy organizes information into categories and hierarchies, an ontology goes further by defining relationships and meaning. It can establish, for example, that a campaign promotes a particular product, targets specific audiences, runs across multiple channels, contains multiple creative executions, is evaluated against defined KPIs, and contributes to particular business outcomes.
This becomes valuable because marketing platforms and teams often represent the same underlying concepts differently. Without a shared semantic structure, “campaign,” “conversion,” “customer,” or even “revenue” can mean different things across systems.
A marketing ontology helps reconcile those differences by creating a consistent representation of the marketing domain. It also makes relationships explicit, allowing systems to understand marketing data as a connected environment rather than a collection of isolated fields and metrics.
That capability becomes particularly important for AI. Models and agents need to understand relationships—not simply retrieve values—to reason reliably about performance. A marketing ontology is therefore one way of structuring and preserving the context AI needs to understand marketing data.
What is a semantic layer, and how is it different from an ontology?
A semantic layer creates a consistent business-friendly representation of data by defining common metrics, dimensions, relationships, and terminology across underlying data sources. It helps people, BI tools, and AI systems interpret and query data consistently without needing to understand how the underlying data is physically structured.
An ontology goes further by representing the meaning and relationships between concepts in a particular domain or business. In marketing, that can include how campaigns relate to channels, audiences, products, KPIs, customer journeys, business outcomes, and other concepts that shape how performance should be interpreted.
The two can work together. A semantic layer helps establish consistent definitions and access to data; an ontology provides richer context about what those concepts mean and how they relate within the business.
For AI, that distinction becomes particularly important. An AI system may be able to query a consistently defined metric such as conversion rate through a semantic layer, while an ontology can help it understand which conversion matters for a particular campaign, how that conversion relates to the customer journey and business objective, and what other factors should be considered when interpreting performance.
What is the difference between a taxonomy, schema, and ontology?
Schemas, taxonomies, and ontologies all bring structure to data, but they solve different problems.
A schema defines how data is organized. It specifies fields, formats, data types, tables, and relationships so information can be stored and processed consistently.
A taxonomy creates a common system for naming, classifying, and organizing information. In marketing, a taxonomy might establish standardized naming conventions and hierarchies for brands, markets, campaigns, channels, tactics, audiences, or creative.
An ontology describes concepts and the relationships between them. It can represent not just that paid social and television are marketing channels, for example, but how those channels relate to particular campaigns, audiences, products, creative, KPIs, conversions, and business outcomes.
A useful shorthand is: schema is how data is structured, taxonomy is how information is classified, and ontology is how concepts and relationships are understood.
These structures aren’t mutually exclusive. They work together. A schema creates technical consistency, a taxonomy creates naming and classification consistency, and an ontology provides a richer semantic representation of how the pieces relate.
That distinction becomes more important as marketing data is used by measurement systems and AI. Organizing information helps systems process it; encoding relationships helps systems interpret it.
Why does AI need marketing context?
AI needs marketing context because access to marketing data alone does not tell a model what that data means for a particular organization, objective, or decision.
An AI system may be able to identify spend, impressions, conversions, and revenue in a dataset. But answering a question such as “Where should we reallocate budget next quarter to increase incremental revenue without weakening long-term brand performance?” requires considerably more understanding.
The system needs to know what the organization considers an important outcome, how campaigns map to products and audiences, the roles different channels play, how performance has historically been measured, what constraints apply, and which external factors may be influencing demand. It may also need to understand relationships learned from previous outcomes.
Data provides signals. Context grounds those signals in the specific environment in which a decision needs to be made.
This becomes more important as AI progresses from retrieving and summarizing information toward recommending and taking action. An AI assistant generating a report can leave substantial interpretation to a human. An agent recommending a multimillion-dollar budget reallocation requires much stronger grounding in the organization’s objectives, definitions, historical performance, measurement logic, and business constraints.
It also changes where durable value can reside in an AI architecture. Models and interfaces will continue to change, but the organization’s accumulated data, relationships, definitions, and historical intelligence can persist. That enterprise data and context asset—not the AI model alone—can become an important source of differentiation.
What is a business ontology, and how is it different from a marketing ontology?
A marketing ontology and a business ontology provide different—but complementary—forms of context.
A marketing ontology represents the concepts and relationships common to the marketing domain: campaigns, channels, audiences, media, creative, exposures, conversions, KPIs, outcomes, and how those elements relate.
A business ontology adds the definitions and relationships specific to an individual organization: its products and services, customer segments, markets, business objectives, organizational structure, KPIs, definitions of success, and other company-specific knowledge.
That distinction matters because marketing concepts do not mean exactly the same thing from one organization to another. Two companies may both measure “conversion,” for example, while one defines it as an ecommerce purchase and another defines it as a new account opening. Their product hierarchies, customer journeys, fiscal calendars, growth objectives, KPIs, and constraints may also be entirely different.
A system therefore needs to understand both how marketing works generally and how marketing works within this particular business.
Together, domain intelligence and business-specific ontology provide a richer semantic foundation: general marketing concepts can be interpreted through the objectives, definitions, and relationships unique to the organization. This is especially important for AI, where useful reasoning depends on grounding generalized model intelligence in the reality of a specific business and use case.
The Substrate applies Domain Intelligence and Business Ontology to encode organizational KPIs, definitions, and relationships, creating governed marketing context for Darwin and other models and agents.
Why isn’t clean, structured data enough for AI?
Clean, structured data is essential for AI, but it does not necessarily contain enough information for AI to interpret a business problem correctly.
Data can be technically pristine—complete fields, consistent formats, standardized names, no duplicates—and still lack the relationships and business meaning needed for reliable reasoning. A model may know that a campaign generated a particular number of conversions without knowing whether those conversions represent the organization’s priority outcome, how the campaign relates to other media activity, or whether external factors contributed to the result.
This creates an important distinction between data quality and contextual completeness.
Clean data helps ensure that AI is working from reliable inputs. Context helps AI understand what those inputs represent within a particular organization and use case. That can include business definitions, relationships between marketing activities, customer journeys, KPIs, historical performance, measurement logic, and relevant external factors.
Both matter. Poor-quality data can produce unreliable analysis; high-quality data without sufficient context can produce analysis that is technically coherent but disconnected from the reality of the business.
For AI-ready marketing infrastructure, the goal is therefore not simply to create cleaner datasets. It is to create reliable, structured, contextualized data that can support meaningful reasoning and decision-making.
Explore further
03 — AI-Ready
AI-Ready Marketing
AI readiness is not simply the adoption of AI tools. It depends on whether an organization has the data foundation, context, integration, governance, and operating conditions required for AI to produce reliable results at scale. For marketing organizations, that means preparing highly fragmented, fast-changing data so AI can understand the business, reason across marketing activity, and support real decisions.
What is AI-ready marketing data?
AI-ready marketing data is data that is sufficiently clean, structured, connected, accessible, governed, and contextualized for AI systems to reliably interpret and use within a specific business and use case.
Having large volumes of marketing data does not make that data AI-ready. Marketing information often spans media platforms, CRM and loyalty systems, websites and apps, commerce, sales, research, and external sources, each with different structures, definitions, identifiers, and levels of granularity. Before AI can reason across those signals, they need to form a usable and consistent foundation.
Context is an important part of that readiness. AI needs to understand not only that a field represents a conversion or a campaign, but what those concepts mean to the organization, how they relate to other marketing and business activity, which outcomes matter, and what historical or external factors may influence performance.
AI readiness is therefore better understood as a set of conditions working together rather than a binary state. While 71% of marketing leaders describe their data as AI-ready, only 37% meet the combined foundational conditions required to operate AI at scale.
AI-ready marketing data is ultimately data that AI can do more than access—it can interpret and use reliably in the context of the business.
What is the difference between using AI and being AI-ready?
Using AI means adopting AI-powered tools or capabilities. Being AI-ready means having the underlying conditions required for those tools to produce reliable, repeatable value.
An organization can use generative AI for content creation, deploy copilots, add AI features to existing platforms, or experiment with AI agents without having an AI-ready marketing operation.
The distinction becomes apparent when AI needs to work with the organization’s own data. If information is fragmented across teams and systems, definitions are inconsistent, historical context is missing, or AI cannot access the data within existing workflows, the sophistication of the model cannot compensate for the weakness of its foundation.
This helps explain why AI adoption and AI performance can diverge. Only 29% of marketing leaders completely agree that teams across their organization work from the same data, and just 28% completely agree that AI tools are fully integrated into marketing workflows.
Being AI-ready therefore requires more than access to AI. It requires data readiness, shared context, integration, governance, and workflows capable of turning AI-generated intelligence into action.
The distinction matters because organizations can add AI quickly. Building the foundation that allows AI to operate reliably across a business is a more fundamental transformation.
How do you make marketing data AI-ready?
Making marketing data AI-ready requires more than cleaning a dataset. Organizations need to create a foundation in which data can be reliably accessed, interpreted, connected, and used by AI within the context of the business.
That typically requires several conditions working together:
Quality: Data is accurate, reliable, and consistently maintained.
Structure: Fields, naming conventions, taxonomies, and formats are standardized.
Integration: Relevant marketing, customer, sales, commerce, and other business signals can be connected.
Accessibility: AI systems can securely access the information they need when they need it.
Shared definitions: Teams and systems operate from consistent KPIs, terminology, and sources of truth.
Context: Data is connected to the organization’s objectives, relationships, historical performance, measurement logic, and relevant external factors.
Workflow integration: Intelligence can move into the places where marketers actually make decisions and take action.
Governance: Access, definitions, lineage, privacy, and appropriate controls are maintained as AI use expands.
The important point is that these conditions are interdependent. Excellent data quality does not solve fragmented access. Integration does not solve inconsistent definitions. Clean, unified data does not automatically provide the business context AI needs to interpret it correctly.
That is why AI readiness is better approached as an infrastructure and operating-model problem, rather than a data-cleaning project completed immediately before deploying an AI application.
Why do AI initiatives fail because of data?
AI initiatives can underperform when the data they depend on is incomplete, fragmented, inconsistent, inaccessible, or disconnected from the business context required to interpret it.
AI models are powerful at identifying patterns and reasoning over information, but they cannot independently correct every weakness in the environment they’re given. If two teams define the same KPI differently, critical data sits outside the system, customer journeys are incomplete, or the model lacks the context needed to understand what matters to the business, AI can produce outputs that appear plausible without being sufficiently reliable for decision-making.
The problem becomes more consequential as AI moves closer to action. A weak foundation might result in a poor summary from a generative AI tool. The same weakness in an agent recommending budget allocation or campaign changes can influence real business outcomes.
The performance difference is already visible. Organizations operating with unified marketing data report stronger AI-driven improvement than organizations with fragmented data across every measured use case, including audience targeting, creative effectiveness, campaign optimization, full-funnel measurement, and business outcomes.
This is why adding a more capable model does not necessarily solve an AI performance problem. When the constraint is the underlying data and context, improving the intelligence layer without improving the foundation leaves the fundamental problem intact.
How do you prepare marketing data for AI agents?
Preparing marketing data for AI agents requires a higher standard than making data available for analysis. Agents need to continuously interpret information, reason across relationships, make recommendations, and potentially take action—often without a human reconstructing the context for every request.
The foundation therefore needs to provide reliable data and persistent context. That includes clean and normalized data, shared definitions, relationships among campaigns, audiences, products, channels and outcomes, business-specific KPIs and objectives, historical performance, measurement intelligence, and relevant external signals.
Agents also need information in a form they can actually use. Traditional marketing intelligence is often delivered through dashboards, presentations, reports, and other outputs designed for human consumption. Agentic systems require structured, queryable, continuously accessible intelligence that can be incorporated directly into reasoning and workflows.
Governance becomes increasingly important as autonomy increases. Organizations need to know which data and actions an agent can access, maintain consistent definitions and lineage, and establish appropriate constraints around recommendations and execution.
The goal is not simply to connect an AI agent to more data. It is to give the agent a governed environment in which data, business context, measurement intelligence, and historical learning remain available as it reasons and acts.
That is the shift from AI-ready data to agent-ready infrastructure.
Does AI need structured data?
No. Modern AI can work with both structured and unstructured data—but the ability to process information is not the same as having the structure required to use it reliably for a specific marketing decision.
Large language models can interpret documents, text, images, and other unstructured information. That makes unstructured data increasingly useful within marketing AI applications. But many marketing decisions require systems to reason consistently across spend, campaigns, audiences, channels, creative, conversions, revenue, customer journeys, and other signals.
Structure provides consistency: standardized fields, definitions, taxonomies, relationships, and formats allow systems to connect and compare information reliably. Context then adds the business-specific meaning required to understand what those structured elements represent.
The distinction matters because simply placing reports, presentations, dashboards, or raw platform exports within reach of an AI model does not necessarily create a trustworthy marketing intelligence system. The model may be able to read the information, but the underlying definitions and relationships can remain fragmented or ambiguous.
A strong AI-ready foundation can therefore use both structured and unstructured information. The objective is not to force every piece of marketing knowledge into rows and columns; it is to ensure that the information required for reliable reasoning has sufficient structure, context, and governance for the use case.
Explore further
04 — Measurement
Marketing Measurement & Optimization
Marketing measurement and optimization have evolved through multiple methodologies designed to answer different questions about performance from incremental lift and aggregate marketing contribution to individual customer journeys. Each approach provides a valuable perspective, but differences in data, methodology, granularity, and time horizon can produce fragmented views of performance. As these approaches converge, measurement is increasingly expected not only to explain what happened, but to help marketers evaluate future scenarios, optimize investments, and continuously improve performance.
What is marketing mix modeling (MMM)?
Marketing mix modeling (MMM) is a statistical measurement approach that uses historical, aggregated data to estimate how marketing investments and other factors contribute to business outcomes.
Unlike methods that rely on tracking individual consumers, MMM examines patterns across variables such as media spend, sales, pricing, promotions, seasonality, distribution, competitive activity, and economic conditions. This allows marketers to estimate the contribution of different channels while accounting for factors outside of marketing that may also influence demand.
That breadth is one of MMM’s primary strengths. It can measure online and offline media, does not require person-level tracking, and provides a portfolio-level view useful for strategic planning and budget allocation.
Traditional MMM also has limitations. It has historically required several years of consistent data, taken weeks or months to complete, and operated at an aggregate level that provides limited visibility into individual customer journeys or tactical performance. Newer modeling approaches can reduce some of these constraints, but those advances should be distinguished from traditional MMM itself.
Privacy and signal loss have renewed interest in MMM because its reliance on aggregate rather than person-level data makes it well suited to a measurement environment in which individual journeys are increasingly difficult to observe.
What is multi-touch attribution (MTA)?
Multi-touch attribution (MTA) is a marketing measurement approach that evaluates how multiple marketing interactions contribute to an outcome across an individual customer journey.
Unlike first- or last-touch attribution, which assigns credit to a single interaction, MTA attempts to distribute contribution across the touchpoints a person or device encountered before converting. This provides granular insight into channels, campaigns, tactics, audiences, and other elements of the journey.
That granularity has historically made MTA valuable for digital marketing and in-flight optimization. But it also creates a fundamental dependency: MTA can only directly measure the journey it can observe.
Privacy regulation, cookie deprecation, walled gardens, cross-device behavior, and platform restrictions have reduced access to the person-level signals traditional MTA depends on. The logic of understanding contribution across the journey remains valuable; the observable data needed to reconstruct that journey has become less complete.
Modern approaches can address this differently. Rather than assuming an unobserved exposure did not occur, data generation and journey reconstruction can use available user-level and aggregate signals to statistically fill gaps in incomplete journeys.
The evolution of MTA is therefore less about abandoning journey-level measurement than about finding more privacy-resilient ways to preserve its granularity.
What is incrementality, and how is it measured?
Incrementality measures the outcomes that occurred because of a marketing activity and would not otherwise have happened.
This is different from simply identifying an association between an exposure and an outcome. If someone sees an advertisement and later makes a purchase, attribution can establish a relationship between those events. Incrementality asks the counterfactual question: Would that purchase have happened without the advertising?
Incrementality can be measured through approaches such as randomized controlled experiments, holdout groups, geo experiments, platform lift studies, and statistical or causal modeling. In a traditional lift study, for example, outcomes from an exposed group are compared with those of a control group to estimate the additional behavior caused by the marketing intervention.
Experiments can provide highly defensible causal evidence, but they also have limitations. Individual studies typically evaluate a specific campaign, channel, or moment in time; they can be costly or operationally difficult to run continuously; and results from multiple isolated tests do not automatically provide a unified view of the entire marketing portfolio.
The broader opportunity is to incorporate incremental contribution into an ongoing measurement framework rather than relying solely on occasional point-in-time tests.
What are the strengths and limitations of MMM, MTA, and lift studies?
MMM, MTA, and lift studies each answer a different part of the marketing measurement problem. None provides a complete view on its own.
MMM offers a broad, top-down view of marketing contribution. It can incorporate offline media, long-term effects, and external variables without relying on person-level tracking, making it useful for strategic budget allocation. Traditional MMM, however, is relatively aggregated and has historically been slower to update.
MTA provides a bottom-up, granular view of individual customer journeys. It can help marketers understand how channels, campaigns, and tactics contribute across observable touchpoints and can support more tactical optimization. Its primary limitation is signal: privacy changes and platform restrictions mean an increasing portion of the journey cannot be directly observed.
Lift studies provide strong causal evidence by comparing exposed and control populations. But they generally produce point-in-time answers about specific interventions rather than a continuous view of the entire portfolio.
These are not competing definitions of measurement so much as different lenses on the same underlying system. The challenge for modern measurement is connecting their strengths without inheriting their blind spots as separate, disconnected views.
Why do different marketing measurement methods produce different answers?
Marketing measurement methods can produce different answers because they use different data, assumptions, levels of granularity, time horizons, and approaches to determining contribution.
MMM may estimate the contribution of television to overall sales using aggregate historical patterns and external variables. MTA may evaluate observable television or digital exposures within individual customer journeys. A lift study may compare outcomes between exposed and control populations. Each is looking at performance through a different methodological lens.
The problem arises when those answers are treated as though they should be interchangeable.
Most enterprise organizations use more than one measurement approach, leaving marketing and analytics teams to reconcile outputs that may not fully agree. Differences in attribution windows, outcome definitions, missing signals, model assumptions, and external variables can widen those gaps further.
This does not necessarily mean one method is right and another is wrong. Different methodologies are designed to answer different questions.
The larger measurement challenge is therefore not simply selecting the “best” methodology. It is creating a framework in which strategic, tactical, journey-level, and incremental views of performance can inform one another rather than requiring humans to continually reconcile separate versions of reality.
What is unified marketing measurement?
Unified marketing measurement brings multiple measurement perspectives together to create a more consistent understanding of how marketing contributes to business outcomes.
Historically, MMM, MTA, and experimental approaches evolved separately. MMM provided a strategic, aggregate view of the marketing portfolio, while MTA provided more granular customer-journey and tactical intelligence. Organizations have often had to reconcile those outputs after the fact.
Unified measurement seeks to connect them.
But putting an MMM model and an attribution model next to one another does not necessarily make measurement unified. The deeper question is whether strategic and granular measurement operate from a common data and modeling foundation.
Marketing Evolution approaches this from the user level up. Available person-level signals are combined with aggregate information, while data generation and journey reconstruction help account for portions of the customer journey that cannot be directly observed. Those granular views can then be calibrated against broader market-level realities rather than operating as a separate measurement system.
This allows marketers to move between questions such as “How much did marketing contribute overall?”, “Where should the next dollar go?”, and “Which audiences, channels, tactics, and exposures drove that contribution?” using a more consistent measurement foundation.
The goal of unified measurement is not simply more methodologies. It is a connected view of performance that can support strategic planning and ongoing optimization.
What is open-source marketing mix modeling?
Open-source marketing mix modeling uses publicly available MMM frameworks that organizations can implement, customize, and operate themselves rather than relying exclusively on a proprietary measurement platform or service. Meta Robyn and Google Meridian are prominent examples.
Open-source MMM has helped democratize access to sophisticated measurement. It gives organizations greater transparency into methodology, flexibility to adapt models to their needs, and more control over how measurement is developed and maintained.
But access to a model does not eliminate the work required to make measurement reliable. Marketing data still needs to be collected, cleaned, normalized, mapped across changing taxonomies, connected to business outcomes, and enriched with relevant variables. Models also require validation, calibration, maintenance, and appropriate analytical expertise.
This creates an important distinction between having access to a model and building a durable measurement capability.
Open-source models can therefore coexist with broader measurement infrastructure. Marketing Evolution’s Substrate can provide clean, consistent, ready-to-use data for organizations that choose to run packages such as Robyn or Meridian. Customers can select the modeling approach that best fits a particular use case while maintaining a common marketing performance data foundation underneath it. The Substrate can provide normalized, contextualized, model-ready data for open-source packages such as Robyn and Meridian.
Open-source MMM expands model choice. A modern measurement strategy also needs to consider the data, context, continuity, and decision infrastructure surrounding the model.
What role should managed services play in modern marketing measurement?
Managed services remain important in marketing measurement, but their role is changing as organizations seek to own more of their measurement capability and institutional knowledge.
Measurement has traditionally been highly service-intensive. Specialists may collect and prepare data, configure models, interpret results, and deliver recommendations through periodic reports or presentations. That expertise can be valuable, particularly for complex methodologies and organizations without large internal measurement teams.
The limitation arises when the expertise, methodology, and accumulated learning remain primarily with the service provider. The organization receives measurement outputs but does not necessarily retain the underlying capability or institutional knowledge when the engagement ends.
A modern service model shifts that relationship. Services can support implementation, onboarding, taxonomy configuration, model validation, strategic interpretation, and ongoing model stewardship while the data, context, measurement history, and resulting intelligence accumulate within infrastructure the organization owns. Marketing Evolution describes its services approach in these terms: expertise remains important, but it is used to build and strengthen the client’s measurement capability rather than create permanent dependency on a proprietary services process.
The role of services therefore moves from delivering measurement to enabling, validating, and extending a measurement capability that persists within the organization.
How is AI changing marketing measurement?
AI is changing marketing measurement in two important ways: how measurement is produced and how measurement intelligence is consumed.
AI can automate parts of the measurement workflow, including data diagnostics, model development, analysis, scenario generation, and insight discovery. Forrester already identifies agentic model development and generative AI for measurement insights among the emerging use cases in the measurement and optimization market.
But the larger change may be what happens to measurement after the model runs.
Historically, measurement intelligence has largely been packaged for people through dashboards, reports, presentations, and periodic analyses. AI systems and agents need that intelligence in a different form: structured, consistently defined, queryable, and available within the workflows where decisions are made.
That changes measurement from a periodic analytical output into a persistent source of intelligence. Measured outcomes can inform planning, scenario analysis, optimization, and future recommendations, while new results feed back into the same foundation.
For AI agents, consistency becomes especially important. If MMM, attribution, experimentation, and business KPIs represent performance differently, an agent inherits those contradictions. A shared measurement foundation gives humans and AI a more consistent basis for reasoning about what happened, why it happened, and what to do next.
AI does not eliminate measurement methodology. It increases the value of making measurement connected, persistent, and usable as an input to decision-making.
What is marketing scenario planning?
Marketing scenario planning uses measurement intelligence to estimate how changes in strategy, budget, or market conditions could affect future performance before those changes are made.
Where measurement helps determine what contributed to past performance, scenario planning is forward-looking. It allows marketers to ask questions such as: What happens if we increase the total budget? What if we shift investment from one channel to another? How might a change in market conditions affect the expected return?
Simulation models potential outcomes under different assumptions. Optimization goes a step further by identifying the allocation most likely to achieve a defined objective—such as maximizing revenue, conversions, or return on investment—within a given budget and set of business constraints.
The quality of those scenarios depends on the measurement intelligence underneath them. Models may need to account for factors such as diminishing returns, interactions between channels, historical response patterns, external variables, and the relationship between marketing activity and business outcomes.
This is why connecting measurement and optimization matters. Measurement establishes what has been learned about performance; scenario planning and simulation apply that learning to potential future decisions; optimization helps determine where resources can be deployed most effectively.
When these capabilities operate from the same data and measurement foundation, actual outcomes can also feed back into the system, allowing future scenarios and recommendations to incorporate what the organization continues to learn.
Explore further
05 — Decisioning
Marketing Intelligence & Decisioning
Marketing organizations have access to more data, analytics, and measurement than ever before, but producing an insight and making a better decision are not the same thing. Marketing intelligence becomes more valuable when it preserves what an organization has learned, connects that knowledge to business context, and makes it available where decisions are actually made. Increasingly, AI is compressing the distance between analysis, recommendation, and action—shifting the challenge from generating more insights to building systems that can consistently turn intelligence into better decisions.
What is marketing intelligence?
Marketing intelligence is the organized information and knowledge marketers use to understand performance, customers, markets, and opportunities and make better decisions.
It can include marketing and customer data, competitive and market information, measurement results, historical performance, forecasts, and other signals relevant to the business. But intelligence is more than the underlying data. Data records signals; analysis identifies patterns; intelligence gives those patterns enough context and meaning to inform a decision.
For example, knowing that paid social ROAS declined is a performance signal. Understanding that the decline coincided with audience saturation, increased media costs, a pricing change, and stronger performance in another audience—and knowing how those factors have affected outcomes historically—creates a more useful basis for deciding what to do next.
Traditionally, much of this intelligence has been distributed across dashboards, reports, analysts, agencies, and individual teams. That makes it difficult to preserve and reuse what the organization has learned.
A modern marketing intelligence foundation connects data with measurement, business context, historical learning, and decision logic so intelligence can persist beyond a particular analysis or report. The objective is not simply to generate more insights, but to make what the organization knows continuously available for planning, optimization, and decision-making.
What is the difference between marketing analytics and marketing intelligence?
Marketing analytics is the use of data and analytical methods to understand marketing performance, patterns, and relationships. Marketing intelligence is the broader body of information and knowledge used to interpret those findings and inform decisions.
Analytics might determine that a channel’s incremental contribution has declined, identify a change in conversion behavior, forecast future performance, or uncover differences between audiences. Those analytical outputs become more useful when interpreted alongside the context of the business: objectives, historical performance, customer behavior, market conditions, constraints, and what the organization has learned from previous decisions.
The distinction is therefore less about competing technologies than about the role the information plays.
Analytics helps answer questions such as “What happened?”, “Why did it happen?” and “What might happen next?” Marketing intelligence connects those answers to a broader understanding of the business so marketers can determine “What does this mean for us?”
Historically, that interpretation has often happened outside the analytical system—in meetings, spreadsheets, presentations, or the minds of experienced practitioners. As marketing intelligence becomes more structured and persistent, that context can become part of the system itself rather than being reconstructed every time a decision needs to be made.
This matters increasingly for AI. An AI system can analyze data, but better decision support requires access to the context and accumulated intelligence needed to interpret that analysis correctly.
How do marketers turn insights into action?
Marketers turn insights into action by connecting analysis to the objectives, constraints, workflows, and decisions the insight is intended to influence.
This sounds straightforward, but the gap between insight and action is a persistent marketing problem. An insight may be accurate without telling a marketer what to change. Teams may also need an analyst to interpret the finding, evaluate alternatives, reconcile it with other information, secure approval, and manually execute the resulting decision.
A more actionable intelligence system shortens that chain. It makes relevant data and measurement continuously accessible, preserves shared definitions and business context, and allows marketers to explore scenarios and evaluate potential decisions within the workflows where they operate.
That operational layer matters. AI readiness research points to the need to embed intelligence directly into workflows so teams can conduct rapid analysis and scenario evaluation without repeatedly exporting data or waiting on analysts.
The objective is not to eliminate human judgment. It is to remove the unnecessary friction between knowing something and being able to do something with it.
That creates a progression from data to analysis to intelligence to decision to action, with the results of those actions feeding back into the system so future decisions can incorporate what the organization has learned.
What is marketing decisioning?
Marketing decisioning is the process of using data, intelligence, objectives, and business rules to determine what marketing action to take next.
Where analytics helps marketers understand performance and marketing intelligence provides the context needed to interpret it, decisioning applies that intelligence to a choice. That might mean determining how to allocate budget, which audience to prioritize, whether to change a campaign, how to respond to a shift in performance, or which scenario best supports a business objective.
Decisioning can be human-led, machine-assisted, or increasingly automated. What matters is that the recommendation is grounded in the specific conditions of the business rather than generated from data in isolation.
That requires more than a predictive model. Effective decisioning may need to account for measurement results, historical response, business objectives, budget constraints, diminishing returns, market conditions, dependencies between channels, and the organization’s own definitions and rules.
The Substrate provides the governed marketing context. Darwin uses that context to monitor performance, measure impact, simulate scenarios, and optimize marketing decisions.
The result is a shift from systems that primarily surface information toward systems designed to support decisions.
What is marketing optimization?
Marketing optimization is the process of identifying and implementing changes to marketing investments or activity that are expected to improve a defined business outcome.
Optimization can occur at different levels. A marketer might optimize the allocation of budget across channels, shift investment between audiences or tactics, adjust campaign parameters, or evaluate a broader portfolio of marketing investments.
Importantly, optimization is not simply identifying what performed best in the past. The highest-performing historical channel is not necessarily where the next dollar should go. Response can change as investment increases, audiences saturate, market conditions shift, and channels interact with one another.
Effective optimization therefore depends on measurement and simulation. Measurement provides evidence about how marketing has contributed to outcomes; forecasting and scenario planning estimate what may happen under different choices; optimization identifies the allocation or action most likely to achieve the objective within relevant constraints.
This is why optimization is increasingly connected to the broader measurement and intelligence system. Forrester identifies in-campaign optimization, marketing budget optimization, and scenario planning as core use cases for modern measurement and optimization platforms.
Connecting these capabilities creates a continuous loop: measure performance, evaluate alternatives, optimize the decision, observe the outcome, and use that result to improve what the system knows.
How is AI changing marketing decision-making?
AI is changing marketing decision-making by reducing the time and technical effort required to move from data to analysis, recommendation, and action.
Early marketing AI applications largely helped people produce or analyze information. More advanced systems can interrogate performance data, identify patterns, generate forecasts, evaluate scenarios, recommend optimizations, and increasingly initiate workflows based on those recommendations. Forrester’s measurement landscape already includes both automated or agentic recommendation generation and automatic or agentic media plan execution as platform capabilities.
But faster recommendations do not automatically produce better decisions.
AI still needs reliable data, consistent definitions, measurement intelligence, and context about the organization it is supporting. It needs to understand what the business is optimizing for, which constraints apply, how performance is measured, and what previous outcomes have taught the organization.
That is why the underlying intelligence layer becomes more—not less—important as AI becomes more capable. The Substrate gives Darwin and other models or agents access to the organization’s KPIs, definitions, relationships, and governed marketing context. Darwin uses that context for measurement, forecasting, scenario planning, optimization, and monitoring.
The evolution is therefore not simply human decisions to AI decisions. It is a progression from AI that helps marketers access information, to AI that supports decisions, to agentic systems capable of reasoning across that intelligence and taking increasingly autonomous action within defined boundaries.
What is Darwin?
Darwin is Marketing Evolution’s marketing intelligence application built on The Substrate. It monitors performance, measures impact, simulates scenarios, and optimizes marketing decisions using the trusted marketing context The Substrate provides.
How do Darwin and The Substrate work together?
The Substrate creates and preserves complete, trusted marketing context. Darwin uses that context to monitor, measure, simulate, and optimize marketing performance. Separating the context layer from the application allows the organization’s marketing intelligence to persist as models, agents, and interfaces change.
Explore further
06 — Agentic
Agentic Marketing
Agentic marketing describes a shift from AI that primarily analyzes or generates information to AI systems that can reason across marketing context, make recommendations, coordinate tasks, and increasingly take action within defined boundaries. The opportunity is not simply more automation. It is to connect agents to reliable data, measurement intelligence, business context, and workflows so they can support marketing decisions continuously and responsibly.
What is agentic marketing?
Agentic marketing is the use of AI agents to perform or coordinate marketing work with a degree of autonomy, such as analyzing performance, evaluating scenarios, recommending actions, monitoring changes, or executing defined tasks.
Unlike traditional automation, which generally follows predetermined rules, an AI agent can interpret information, reason toward an objective, select among possible actions, and adapt based on new inputs or outcomes.
In marketing, that might mean an agent monitoring performance across channels, identifying a meaningful change, investigating likely causes, evaluating alternative budget scenarios, and recommending or initiating a response.
The distinction is important because agentic marketing is not simply “AI that does more things.” Agents become useful when they can operate within the specific context of the business: its KPIs, measurement logic, customer journeys, constraints, historical performance, and definitions of success.
That is why agentic marketing depends heavily on the foundation underneath it. As agents become more capable, the quality, structure, and domain specificity of the data and context they reason on become increasingly important.
What is an AI marketing agent?
An AI marketing agent is a software system that can use data, context, and tools to reason toward a marketing objective and take or recommend actions to help achieve it.
Unlike a traditional AI assistant that primarily responds to individual prompts, an agent can participate in an ongoing workflow. It may determine what information it needs, retrieve and analyze data, evaluate possible actions, use specialized tools or models, and respond as conditions change.
In marketing, agents can be designed for specific jobs. An agent might monitor campaign performance and surface meaningful changes, reconstruct customer journeys, evaluate attribution and incrementality, forecast outcomes, simulate different investment scenarios, or recommend how budget should be allocated. Darwin includes specialized agents across audiences and journeys, measurement and optimization, MMM, reporting, data management, and activation.
Agents can also operate at different levels of autonomy. Some may analyze information and make recommendations for a marketer to approve, while others can execute defined actions within established permissions and constraints. Forrester distinguishes between agentic recommendation generation and agentic media-plan execution as separate capabilities.
What makes an agent effective is not simply the AI model powering it. Its usefulness depends on what it can reason on: the quality of its data, the context it has about the business, the measurement intelligence available to it, and the tools and permissions it can access. For marketing, that means grounding agents in the organization’s own KPIs, definitions, customer journeys, performance history, and business objectives—not relying on generic marketing knowledge alone.
What is the difference between generative AI and agentic AI?
Generative AI primarily creates or transforms content and information in response to a request. Agentic AI can use that intelligence as part of a broader process of reasoning, deciding, and acting toward an objective.
A generative AI system might summarize a campaign report, explain a performance change, or draft a recommendation. An agentic system can potentially determine what information it needs, retrieve and analyze that information, evaluate alternatives, recommend a course of action, monitor the result, and continue the workflow as new data becomes available.
The two are not mutually exclusive. Agents frequently use generative models as part of their reasoning and communication. The distinction is about what the system does with the model’s output.
In marketing, this can shift AI from a conversational interface that answers questions to a system capable of participating in ongoing workflows such as measurement, forecasting, optimization, and performance monitoring.
That evolution also raises the requirements for data and governance. A plausible but imperfect answer from a chatbot is very different from a recommendation or action that changes media investment. As AI moves closer to action, systems need stronger grounding, clearer permissions, and more defensible measurement intelligence.
What data and context do AI agents need for marketing?
Marketing AI agents need more than access to raw data. They need reliable, structured, and continuously accessible information grounded in the specific business and use case they are supporting.
That can include media and campaign data, customer and commerce signals, business outcomes, historical performance, measurement results, relevant external factors, and the relationships connecting those elements.
They also need context: how the organization defines its KPIs, how campaigns map to products and audiences, which outcomes matter, what constraints apply, how different channels interact, and what previous measurement has revealed about performance.
This requirement becomes more important because agents may operate without a human reconstructing that context for every decision. The underlying system must preserve it.
As foundation models and agent frameworks become more broadly available, the differentiating factor increasingly shifts toward what those agents reason on: the quality, structure, completeness, and domain specificity of the context layer underneath them.
The Substrate creates complete marketing context through Data Unification, Domain Intelligence, Business Ontology, and Journey Reconstruction. Darwin’s agents use that context, and it’s accessible via the Substrate to any model or agent that uses marketing data.
Can The Substrate support agents other than Darwin?
Yes. The Substrate makes the same trusted marketing context available to Darwin and to any model, agent, or application that reasons over marketing data. Every system can work from consistent KPIs, definitions, relationships, and performance history without rebuilding the intelligence underneath it.
How do AI agents use marketing measurement?
AI agents can use marketing measurement as the evidence layer for understanding what is driving performance, evaluating potential actions, and learning from outcomes.
Historically, measurement was often delivered as a report or recommendation for a person to interpret. In an agentic environment, measurement can become information that software reasons on directly. That changes the requirements: measurement outputs need to be structured, queryable, trustworthy, and available continuously rather than only through periodic reports or dashboards.
An agent might use measurement intelligence to determine whether a performance change is incremental, evaluate which channels or audiences contributed to an outcome, simulate alternative allocations, or assess whether a previous recommendation actually worked.
This creates a feedback loop. Measurement informs the decision; the resulting outcome becomes new evidence; and that evidence can improve future analysis and recommendations.
The value is not simply that AI can read a measurement report faster. Measurement becomes part of the operating context the agent uses to reason and act.
What is an agentic marketing workflow?
An agentic marketing workflow is a process in which AI agents participate across multiple stages of marketing work rather than performing a single isolated task.
For example, a workflow could begin with continuous monitoring. An agent identifies an unexpected performance shift, retrieves the relevant data, investigates likely causes, uses measurement to evaluate contribution, simulates possible responses, recommends an optimization, and—if permitted—initiates the approved action.
The important distinction is continuity. Traditional marketing technology often separates data ingestion, analysis, measurement, optimization, and activation into different tools and handoffs. Agentic workflows can connect those activities into a more adaptive loop.
This is already visible in the direction of the measurement category: modern AI systems can synthesize data, reconstruct missing signals, model performance continuously, and feed decisions back into activation rather than operating through a strictly linear sequence of disconnected tools.
The effectiveness of that workflow still depends on a shared foundation. If each agent or system uses different data, definitions, or measurement logic, automation can accelerate inconsistency rather than resolve it.
Explore further
Stop reporting on last quarter.
Start deciding on the next one.
Join the teams already turning data into context and context into performance.