Behavioral Models and Scores

Individual signals can reveal a great deal about an identity

At enterprise scale, the challenge is knowing how those signals work together.

AtData combines longitudinal identity history, behavioral activity, network intelligence, and machine learning to transform complex evidence into scores organizations can use for growth, risk, analytics, and decisioning.

A score should make complexity easier to use — not hide what created it.

That is why AtData provides both model-driven scores and many of the underlying signals that inform them.

---

Models turn observations into usable decisions.

An email address generates signals over time.
How long has it existed? How recently has it been active? How broadly is it observed? How consistently is it used? Does its domain demonstrate risky behavior? Does its activity resemble legitimate or fraudulent identities?

Individually, those signals provide evidence. Behavioral models evaluate relationships across that evidence and translate them into outputs that are easier to incorporate into business logic, segmentation, analytics, and real-time decisioning.

The result is a progression:
Observe → Measure → Model → Decide

AtData’s approach allows organizations to use the sophistication and explainability their decision requires.

Explore Our Methodology

---

Scores built for different questions.

There is no universal definition of a “good” identity. A marketer evaluating engagement is solving a different problem from a fraud team evaluating risk. The same identity may therefore produce different intelligence depending on the decision being made.

AtData develops purpose-built models rather than reducing every use case to one generic identity score.

Quality Score
How valuable is this email identity from an engagement perspective?

AtData Quality Score is a relative engagement score designed to help organizations understand the quality of a valid email beyond simple deliverability.

The machine-learning model evaluates billions of behavioral signals from hundreds of sources, including open, click, web activity, ecommerce purchase activity, source diversity, frequency of use, and overall digital footprint.

Use Quality Score to:

• Prioritize leads and audiences
• Compare acquisition quality
• Improve segmentation
• Identify stronger seed populations
• Prioritize reactivation
• Suppress lower-value records

Explore Quality Score

Open Score
Is this email identity demonstrating recent engagement?

Open Score focuses specifically on recent email engagement.

The 0–10 score reflects the recency and frequency of observed opens across AtData’s network and can be paired with Last Open Date to distinguish more active identities from those showing limited recent engagement.

Use Open Score to:

• Prioritize active audiences
• Re-engage dormant customers more intelligently
• Improve suppression strategies
• Enrich lead or churn models
• Evaluate engagement outside your own first-party view

Deliverability tells you whether an address can receive a message.

Open Score adds evidence about whether that identity appears to be paying attention.

Explore Open Score Intelligence

Identity Risk Score
How much risk does this identity represent?

AtData’s Risk Score converts historical, behavioral, identity, domain, and other risk intelligence into a 0–100 machine-learning score, with higher values representing greater potential risk.

Inputs can include signals such as:

• Email age and longevity
• Velocity and popularity
• Identity correlations
• Email tumbling and sequencing
• Domain behavior
• Validation intelligence
• Fraud-network intelligence
• IP and other contextual information when supplied

Organizations can use the score as an immediate risk indicator or combine the underlying signals with proprietary fraud models and business rules.

Explore Risk Score

Domain Risk Score
Does the infrastructure behind the email show signs of risk?

Fraudulent identities increasingly rely on domains that appear quickly, operate briefly, and disappear before traditional classification systems catch up.

AtData’s 0–10 Domain Risk Score evaluates behavioral and risk signals observed around an email domain across the AtData network, helping organizations identify potentially high-risk domains as they emerge.

Use Domain Risk Score to add context around:

• Newly emerging domains
• Disposable and temporary providers
• Unusual domain behavior
• Coordinated account creation
• Application and transaction risk

Explore Domain Intelligence

---

Use the score… or go beneath it.

A composite score is useful when a system needs a fast, consistent output. It is not always enough.

Data scientists, fraud teams, and sophisticated enterprise decisioning environments may want the individual signals beneath that score so they can determine how much each matters for their own population. AtData supports both approaches.

Use a score when:

• A workflow needs a simple decision input
• Speed matters
• Internal modeling resources are limited
• Consistent thresholds are valuable
• Rapid implementation is the priority

Use individual signals when:

• You have proprietary models
• Explainability is important
• Different segments require different treatment
• You want to test individual predictors
• Your risk or growth strategy is highly specialized

Use both when:

You want AtData’s model output as another predictor alongside the raw evidence supporting your own decision framework.

Fuel Custom Models

---

Build intelligence around behavior, not assumptions.

Static attributes tell you what an identity looks like. Behavior can reveal how that identity actually exists in the world.

AtData’s models draw from a foundation that includes more than 150 billion deterministic activity signals processed each month, decades of email identity history, and broad identity relationships. That scale allows models to incorporate patterns unavailable from a single organization’s first-party interactions alone.

Those signals can help distinguish:

• Active from merely valid.
• Established from newly created.
• Consistent from anomalous.
• Higher-quality from lower-value.
• Lower-risk from suspicious.

The model changes with the question but the underlying advantage remains the same: behavioral context accumulated around persistent identity.

---

Put intelligence inside the systems already making decisions.

AtData scores and signals are designed for use within existing enterprise environments rather than requiring another destination platform.

Depending on the use case, intelligence can be delivered through:

• Real-time APIs
• Batch processing
• Licensed datasets
• Supported integrations
• Custom data workflows

Feed AtData intelligence into:

• Fraud and risk engines
• Lead-scoring models
• CRM and CDP environments
• Customer segmentation
• Identity graphs
• Data warehouses
• Analytics platforms
• AI and machine-learning models
• Proprietary enterprise products

---

Why AtData

The model is only as differentiated as the data beneath it.

Machine learning is increasingly accessible. Proprietary history is not.

AtData’s advantage comes from combining decades of identity history with continuously observed behavioral activity and network-level context. That foundation allows us to build models around how identities behave, and not simply what attributes they contain.

And because different organizations make different decisions, AtData does not require customers to surrender control to a black box. Use the model. Use the signals. Or use both.

---

Make complex identity evidence easier to act on.

Whether the objective is improving audience quality, prioritizing engagement, detecting fraud, evaluating domains, or strengthening a proprietary model, AtData turns behavioral identity evidence into intelligence designed for action.