What Full-Funnel Attribution Requires at the Data Layer

Every growth-stage company we work with has an attribution problem. Few of them have diagnosed it correctly.

The symptoms are familiar. Finance sees one customer acquisition cost. Growth marketing sees another. Brand marketing cannot produce a number finance will accept. Lifecycle/CRM reports incremental lift on campaigns running against audiences that three other teams are also targeting. Leadership asks what the marketing mix should be and gets four different answers backed by four different data sets.

Full-funnel attribution is a data architecture problem. Teams that cannot answer how marketing drives growth have not failed to select the right attribution model. They have failed to build the data assets any attribution model requires to function. Until those assets exist as durable, centrally governed infrastructure, every attribution tool purchased downstream will compound the fragmentation instead of resolving it.

The Four Data Foundations Every Attribution Model Requires

Attribution is not a method. It is an output. Whatever method you use to produce that output, whether deterministic last-touch, geo-based incrementality, media mix modeling, or a composite of all three, the method depends on four underlying data assets. We call these the four foundations of attribution.

• Touchpoints are the complete record of every meaningful interaction a user has with your brand. Impressions served, ads clicked, emails opened, pages viewed, videos started, sessions initiated. Touchpoints must be captured continuously, from first exposure to recurring interaction, and landed in a data environment where they can be queried alongside other assets. Teams often have strong touchpoint capture for one channel and virtually none for another. That gap alone makes full-funnel attribution impossible.
• Identity resolution is the logic that stitches touchpoints, conversions, and spend back to a consistent entity. Device, user, household, account. The choice of entity matters less than the discipline of choosing one, documenting the tradeoffs, and applying it consistently across every downstream model. Without identity resolution, attribution is a collection of disconnected events that cannot be reassembled into a journey.
• Conversions are the goal-completion events the business actually cares about. Revenue is the most common, but conversions also include signups, viewership thresholds, repeat engagement, demographic acquisition targets, and retention milestones. Conversion definitions must be explicit, versioned, and shared. When finance uses one definition of a new customer and growth uses another, no attribution method in the world will reconcile them.
• Spend is the allocation of financial resources across channels, campaigns, and creative. Spend data must be ingested at a granularity that matches the touchpoint and conversion data it will be joined against. Aggregated weekly totals from an agency partner will not reconcile against daily campaign-level conversions. The mismatch is structural, and no transformation layer can fix it after the fact.

These four foundations are the prerequisites. Not the nice-to-haves. If any one of them is missing, fragmented, or team-owned rather than centrally governed, full-funnel attribution is not achievable regardless of what sits on top of them.

Why Team-Owned Attribution Creates Silos and Contradictory Outputs

A common failure mode we see is not the absence of attribution infrastructure. It is the proliferation of it. Growth teams build their own pipelines, their own cohorting logic, their own LTV forecasts. CRM teams operate inside their messaging platform's native reporting, define their own conversion windows, and measure lift in isolation. Brand teams rely on agency-delivered dashboards built on aggregate data they cannot join to anything else. Each team's work is internally coherent. None of it composes.

This is the shape of attribution failure at growth-stage companies. Not a lack of sophistication, but a lack of shared foundations. When every team builds its own version of the four foundations, you get three or four parallel attribution systems that disagree on what a user is, when a conversion counts, and which spend produced which outcome. The disagreements are not analytical. They are architectural. The teams are answering different questions with different data, and no amount of alignment meetings will reconcile outputs that were never built from common inputs.

The fix is not to force every team onto the same methodology. Different marketing functions genuinely need different attribution lenses:

• Brand marketing legitimately requires incrementality testing because click-based attribution cannot capture its impact.
• Growth marketing legitimately requires deterministic campaign-level reporting for rapid channel optimization.
• Lifecycle marketing legitimately requires short-window conversion analysis to evaluate messaging.

The fix is to ensure every team draws from the same foundational data assets, governed centrally, so their different methods produce comparable outputs.

Method diversity is a feature. Data diversity is a bug.

How Medallion Architecture Makes Attribution Data Durable

The four foundations exist only when they are engineered to exist. At Mammoth Growth, we have spent a decade building them using medallion architecture, and we have not found a better pattern for making attribution data durable, trustworthy, and accessible across teams.

The logic is simple enough to explain and hard enough to execute that most organizations never finish it.

• Bronze tables hold raw ingested data with minimal transformation: the unaltered record of what arrived from the source system.
• Silver tables apply business logic, identity resolution, and enrichment: this is where a touchpoint becomes attributable, where a device ID becomes a user, where spend becomes joinable to conversions.
• Gold tables deliver analysis-ready assets to the teams and tools that consume them: the dashboards, the attribution models, the finance reports.

What this architecture produces that nothing else does is a single place where the four foundations live. Touchpoints are modeled once in silver and consumed by every team from gold. Identity resolution is applied once, documented once, and updated in one place. Conversions are defined centrally and versioned. Spend is ingested through governed pipelines that enforce naming conventions upstream so that downstream joins work without custom cleanup.

The alternative, which we see constantly, is pipelines built per-report. Every new attribution question triggers a new extraction, a new transformation, a new definition of a user or a conversion. The work is never promoted back to a central layer. Institutional knowledge lives in the heads of the three analysts who built it. When they leave, the attribution work leaves with them. This is not a staffing problem. It is the predictable outcome of treating attribution data as a series of ad-hoc queries rather than a governed product.

How to Choose Between Deterministic, Incrementality, and MMM Methods

Once the four foundations are governed and durable, the conversation about attribution method finally becomes productive. Not before.

No single attribution method answers every question. Deterministic attribution gives you fast, granular feedback on campaigns where click-based tracking works, but it systematically underweights channels where the conversion happens days later on a different device. Incrementality testing isolates true causal lift, but it has opportunity cost and operates on slow feedback cycles. Media mix modeling captures offline and long-cycle effects that other methods cannot see, but it requires years of consistent spend data and reacts slowly to market changes. Each method illuminates part of the picture. None of them, alone, illuminates the whole.

The point of building the four foundations centrally is that you can run all three methods against the same underlying data and compare their outputs. Deterministic attribution on gold-layer touchpoints and conversions. Geo-based holdout tests using the same identity resolution logic. Media mix modeling using the same spend data, ingested through the same pipelines, governed by the same conventions. When these methods disagree, the disagreement is informative instead of unresolvable. You know the data is consistent. The variance comes from the methods themselves, which is exactly the variance you want to understand.

This is what full-funnel attribution actually means at the data layer: a common foundation that supports multiple analytical lenses, not a single model that claims to answer every question. Organizations that skip to the methodology conversation before building the foundation will keep buying tools that contradict each other, hiring analysts who cannot reconcile their outputs, and asking leadership to trust dashboards that finance will not accept. The work has to happen in the right order.

Why Attribution Infrastructure Must Precede Methodology Selection

The question companies should be asking is not which attribution model to adopt. It is whether the four foundations exist in a form that multiple attribution models can draw from without contradiction. If they do not, no amount of downstream investment will produce coordinated answers. If they do, the choice of method becomes a tactical decision rather than a strategic bottleneck.

Full-funnel attribution is not a dashboard. It is not a vendor. It is an architecture. Build the architecture first, and the attribution takes care of itself.