Traditionally, marketers have relied on attribution to answer a seemingly simple question: Which marketing touch led to this outcome? But in today’s hyper-personalized, always-on banking environment, that question isn’t just hard to answer, it might be the wrong question altogether.
Attribution is a bit like trying to navigate today’s world with yesterday’s map.
From basic last-touch rules to more advanced multi-touch and algorithmic variants, attribution models were designed for a world in which customers moved linearly through a funnel and marketers could leverage A/B and multivariate testing. Marketers would launch Campaign A and Campaign B, split their audiences 50/50, wait a few weeks, then crown a winner. It’s clean. It’s controlled. It’s easy to explain.
But in Gartner’s Hype Cycle, attribution has slid into the “Trough of Disillusionment.” By contrast, customer data platforms (CDPs), real-time journey orchestration and machine learning optimization are ascending what Gartner calls the “Slope of Enlightenment.”
The Attribution Mirage
Attribution models strive for clarity, causality and control. But, in fact, they demand impossible conditions: a single customer identity across devices, cleanly timestamped and causally linear interactions with no other confounding interactions, and a full view of every experience and influence. None of these exists, especially in banking where journeys span months and can be determined by everything from the interest rate environment to real-time life events. During that time, customers and members might open a new checking account, activate direct deposit, renew a CD, or more. More specifically, attribution doesn’t fit the unique needs of banking for these reasons:
Singular eligibility constraints. A single customer might qualify for a balance-growth campaign, a direct deposit bonus and a card upsell all at once. If they receive only one message, which outcome should get credit if they later take an action that aligns with multiple campaign goals? When customers flow in and out of campaigns and holdouts dynamically—as well they should in any intelligent system—the signals get noisy. Attribution models often can’t handle that level of variability and nuance.
Fixed variables and timelines. A customer could be part of an email savings campaign on Monday, randomly assigned to a holdout group by Thursday and re-enter a personalized push notification campaign for deposit augmentation the following week. This leaves attribution models scrambling to assign credit for growing the account.
That’s because real-world marketing isn’t a lab experiment with a controlled environment for multivariant testing—it’s an ongoing conversation with diverse, evolving customers. Always-on optimization engines are designed to support dynamic treatment paths: an optimized message today, a holdout tomorrow, a non-optimized control the next day. That variability trains robust models and ensures that every customer gets their fair share of optimized exposure. But this sophistication also highlights the cracks in attribution: which one “caused” the outcome?
Single outcome over customer impact. Marketers want clear impact measurement, and attribution models try to provide it by assigning credit. But this drives behavior that maximizes attributed wins, not necessarily actual impact. Optimization engines, on the other hand, focus on long-term customer outcomes, not quick wins.
The Case for Customer-Level Decisioning
Customer-level decisioning engines, on the other hand, flip the script. Instead of asking which touchpoint should get credit, they ask what the best decision for this customer is right now given everything that is known about them.
This is a fundamentally different paradigm. Dynamic Program Eligibility means customers move in and out of program eligibility naturally, without the artificial constraints imposed by static campaign cohorts. As a customer’s behavior shifts, the engine continuously learns and re-optimizes—without the need to restart the clock as with traditional tests. Rather than anchoring on campaign metadata alone, the system learns from actual customer behavior patterns, allowing it to make better decisions over time. The engine aligns every decision to a desired customer behavior and outcome KPI—whether retention, growth or engagement—and continuously optimizes for them across all campaigns and channels.
Curinos’ AI-fueled decision engine, Amplero Personalization Optimizer, measures what really matters: the cumulative effect of smarter decisions that are made in the best interest of each customer. It closely measures performance impact (against universal holdouts) and reveals customer behavior insights across marketing attributes to uncover actionable opportunities.
Rethinking Impact Measurement
Impact measurement shouldn’t be about attributing a single campaign win. It should be about demonstrating that a marketing engine is improving customer outcomes overall—sustainably, measurably and fairly.
With Amplero’s always-on decisioning, marketers can measure impact at the customer level, comparing trajectories against a persistent, randomized, representative control group receiving default customer experiences. It can observe lift across windows following a decision and optimize for future-state behaviors like lifetime value.
“Attribution will not happen because there is no common timeline,” Scott Brinker and Sergio Maldonado have highlighted. Attribution tries to draw a straight line from message to action. But customer behavior isn’t linear, so marketing shouldn’t be either. The time has come to stop trying to force-fit old models onto the new world. Instead, it’s time to measure what really matters: the cumulative effect of smarter decisions made in the best interest of each customer.
