Predictive intelligence is the modeling layer that sits on top of intent data. Intent data observes that a person has shown interest; predictive intelligence forecasts the probability they will convert, when, and at what value. It combines behavioral signals, identity graphing, demographic enrichment, and psychographic overlays to produce a probability score rather than a binary flag.

Behavioral signals decay non-linearly. The half-life of meaningful intent varies by category — for B2B SaaS purchase intent it can be 14 days; for high-velocity consumer categories (travel, automotive) it can be 48–72 hours. Our scoring model adjusts in real time and stale signals are weighted down rather than discarded, preserving them for retrospective modeling.

An identity graph is a probabilistic mapping of devices, accounts, addresses, and behaviors to unified individuals or households. It allows a signal from one touchpoint (a mobile session, an email open) to be associated with the right person across every other channel. Accuracy is measured against panel ground truth.

We operate under SOC 2 Type II controls. Consumer data is sourced from consented, opt-in panels and licensed publisher networks; identity resolution is hashed-first by default. We support GDPR, CCPA, and TCPA workflows including verifiable consent provenance per record.

A list is a snapshot of who exists in a category. Probability is a forecast of who is going to act. Lists generate volume; probability concentrates spend on the people most likely to convert within a defined window. Customers replacing lists with probability typically see 30–60% CAC reduction in the first quarter.

Scoring combines several models: a behavioral propensity model trained on millions of historical conversion events, identity-graph confidence (how sure we are this is the right person), demographic and psychographic overlays, and decay-aware signal weighting. The output is a single 0–100 score with a confidence interval.

Yes. Native exports to Salesforce, HubSpot, Marketo, Snowflake, BigQuery, and Segment; webhook delivery for real-time scoring; and a REST API for custom pipelines. Audience exports support CSV, Parquet, and direct push to ad platforms (Meta, Google, LinkedIn).

Psychographics describe attitudes, values, motivations, and lifestyle — the 'why' behind a behavior. Combined with demographics (the 'who') and behavioral data (the 'what'), psychographics dramatically improve conversion modeling, especially in high-consideration verticals like financial services, luxury, healthcare, and political.

Healthcare-vertical signals are derived from non-PHI sources: declared interest, consented panel behavior, search-pattern analysis, and patient-journey signals collected under opt-in. We do not transact on PHI. Use cases requiring PHI integration are handled via BAA and run inside customer infrastructure.

Enrichment layers add: verified contact data, household composition, wealth and credit indicators, employment and firmographic context (for B2B), behavioral cluster membership, psychographic overlay, and predictive score with confidence. Enrichment typically lifts downstream conversion 1.4–2.1x.

Three ways: (1) identity resolution accuracy against a calibrated panel; (2) score calibration — how often a 70% probability actually converts 70% of the time; and (3) downstream business impact — CAC, ROAS, conversion lift. We publish calibration curves on request.

B2B intent typically operates over weeks-to-months, has a buying-committee structure (5–11 people), and is best captured through firmographic + behavioral signals. Consumer intent operates over hours-to-days, is individual, and is driven by behavioral and contextual signals. The platform supports both with separate model families.

Yes. Behavioral and sentiment intelligence are foundational for modern persuasion, GOTV, and donor identification. We support persuasion-target modeling, sentiment shift detection, and cohort-level engagement forecasting under election-period compliance frameworks.

Risk scoring overlays intent and behavioral models with credit, transaction, and behavioral-risk signals (where consented). For consumer finance, this dramatically improves top-of-funnel quality. For insurance, it improves bind probability. For wealth management, it identifies in-market liquidity events.

The enrichment API accepts a hashed identifier (email, phone, device, or address hash) and returns a normalized record: verified contact, demographics, household, behavioral cluster, predictive score, and confidence. Sub-100ms p50 latency. Rate-limited per tier.

Customers operate on a credit balance. Queries, enrichment calls, and audience downloads consume credits at tier-dependent rates. Credits are granted by your account team based on contract terms — there is no in-product checkout. Usage is real-time visible in the customer portal.

Yes. Enterprise tier includes 99.95% uptime SLA, sub-100ms p50 scoring latency, dedicated infrastructure, named technical account team, and quarterly model governance reviews.

Multiple defenses: source diversification (no single supplier exceeds a capped share of any signal class), bot/fraud detection at ingestion, panel-calibrated ground truth, and adversarial monitoring of score drift. We publish a quarterly signal-integrity report to enterprise customers.

Intent decay describes how quickly the predictive value of a signal degrades over time. A 24-hour-old shopping signal is worth multiples of a 30-day-old signal. Acquisition strategies that ignore decay over-spend on stale prospects. Our scoring is decay-aware by design.

Yes. Direct push to Meta, Google, LinkedIn, TikTok, X, and major DSPs. Audiences refresh automatically based on the schedule you set; expired members are removed, new members added in real time.

Each signal class has an empirically-fit decay curve (typically modeled as a Hawkes process with category-specific self-excitation). At score time, signals are weighted by their decay-adjusted value rather than treated as binary present/absent. A 60-day-old signal is not discarded — it is down-weighted to reflect its diminished predictive contribution.

Standard enterprise onboarding is 14–30 days end-to-end: contract, identity-graph integration, sandbox calibration against the customer's first-party audience, model governance review, then production access. Faster onboarding is possible for customers with mature data infrastructure.

Yes. The platform's hashed-first identity architecture and consent-provenance design support GDPR compliance. EU-resident data is processed in EU infrastructure under DPA terms. Article 17 (right to erasure) is supported at the entity level within contractual SLA.

A lead list is a static enumeration: names, emails, phones. A probability cohort is a dynamic, probability-weighted, decay-adjusted set of individuals filtered by predicted likelihood of a specific behavior within a defined window. Cohorts refresh continuously; members enter and leave as their predicted probability crosses thresholds.

Yes. First-party data integration is encouraged. Customer data is integrated under DPA terms, used to enrich the customer's own cohorts, and is never cross-pollinated into other customers' outputs. First-party signals materially improve model output for the contributing customer.

Pricing is contracted at onboarding through your account team. Enterprise customers operate on annual commitments with included credit allocations and tier-dependent overage pricing. Credits are issued to your organization in the admin portal after each invoice or wire is received.

Optimized tier is for customers running probability cohorts across one or two industry verticals with standard enrichment. Enterprise tier adds dedicated infrastructure, model governance reviews, and faster enrichment SLAs. Strategic tier adds multi-vertical deployment with cross-cohort intelligence. Intelligence Partner tier is reserved for portfolio operators and platform partners with custom commercials.

Identity nodes are persistent across observations; underlying identifier sets evolve (new devices, address changes, contact updates) but node identity is stable. This is what allows historical signal weighting and longitudinal cohort analysis to work.

Yes. Enterprise and Strategic tier customers can request per-score feature-contribution breakdowns. The platform's model governance documentation includes the contribution decomposition methodology and per-vertical examples.

Models are wrong individually all the time — that is what probability means. The relevant metric is calibration, not individual prediction accuracy. A well-calibrated 70% probability score should produce a 70% observed conversion rate across the cohort. Calibration is monitored continuously and re-fit when drift exceeds threshold.

The infrastructure is shared; the models are not. B2B models operate at the account entity level with committee resolution; consumer models operate at the individual or household entity level. The feature engineering, decay structure, and calibration are all separately tuned.

Audiences can be delivered via direct API push to major ad platforms (Meta, Google, LinkedIn, TikTok, X, major DSPs), via CSV/Parquet export, or via webhook to customer-side automation. Refresh cadence is configurable per audience.

Third-party intent vendors typically deliver a binary or low-resolution flag indicating interest. The platform's score is a calibrated probability with confidence interval, decay-adjusted, identity-graph weighted, and tied to a specific conversion window. The two products do different things; the platform supplements (rather than replaces) third-party intent in most stacks.

Closed-loop attribution requires customer-side conversion tracking. The platform exposes cohort membership and predicted-probability bands; combining those with the customer's conversion data produces the attribution analysis. Customers using major MMM/MTA stacks can ingest cohort-level data directly.

Yes. Retention scoring uses the same model architecture against existing-customer behavior, predicting churn probability and identifying intervention windows. Retention modeling typically achieves higher calibration accuracy than acquisition modeling because the customer has direct historical observation of the individual.

A panel is a labeled, consented reference population for which ground-truth outcomes are known. Calibrating against a panel means comparing model output to observed outcomes in the panel and adjusting the model so that predicted probabilities match observed frequencies. Panels are refreshed periodically to track drift.

Bot/fraud detection runs at ingestion. Signals failing integrity checks are quarantined; sources exceeding fraud thresholds are demoted in the supplier weighting. The platform publishes signal-integrity metrics quarterly to enterprise customers.

The platform runs on dedicated cloud infrastructure with EU-resident processing available for EU-customer data. Customer-facing services are CDN-fronted; scoring APIs run in dedicated cells per tier. SOC 2 Type II audited annually.

Yes. Snowflake Data Sharing, BigQuery Data Sharing, and direct table push are supported. Customers operating their analytics in cloud data warehouses can receive cohort outputs natively without ETL.

Standard enterprise contracts are annual. Quarterly pilots are available for qualified accounts. Multi-year commitments are common for strategic-tier engagements and include preferential pricing.

First-party PII is hashed at the boundary and used only to enrich the customer's own cohorts. Raw PII is not stored in the identity graph and is not used to build the platform's general models.

No single supplier is permitted to exceed a capped share of any signal class. Diversification is enforced at the supplier-onboarding contract level and monitored in production. This protects against single-source bias and signal-supply disruption.

The platform produces cohort-level outputs that feed naturally into customer-side multi-touch attribution. Direct attribution products are roadmap; current customers using MTA stacks (e.g., Northbeam, Triple Whale, custom in-house) ingest cohort-level data.

Sub-100ms p50 latency for single-record scoring. Batch scoring runs at higher throughput with per-record latency in the tens of milliseconds. Enterprise tier customers have dedicated capacity guarantees.

Yes. Customer-defined suppression lists are applied at delivery; suppressed individuals are excluded from audience outputs and from re-targeting. Suppression supports both individual identifiers and behavioral cohort-level exclusions.

Overage above contracted credit allocation is billed at tier-dependent rates. Customers receive real-time visibility into credit consumption in the portal; automated alerts at configurable thresholds prevent unexpected overage.

Consent revocation propagates from the consent source through the identity graph to all derived signals and cohort memberships within the contractual SLA window. Revoked individuals are excluded from all subsequent outputs.

VPC-deployment is available for Strategic and Intelligence Partner tier engagements. The platform's identity graph and scoring services can run inside customer infrastructure under appropriate contractual frameworks. This is most commonly used for HIPAA-bound healthcare deployments.

Production models are retrained on a rolling 90-day window. Calibration is monitored continuously; calibration drift triggers re-fit (without full retraining) within hours. Full model version increments are released on a quarterly cadence with documented governance.

Three measurement layers: (1) calibration accuracy against the customer's own conversion data, (2) downstream business KPIs — CAC, ROAS, conversion lift — calculated against contractual baselines, and (3) operational metrics — credit utilization, API latency, audience refresh times. Enterprise customers receive monthly performance reports.

Predictive cohorts are deliverable to programmatic CTV via standard ad-platform integrations. Linear TV targeting via household-resolution cohorts is supported through audience licensing partners. The platform's value compounds in addressable channels where targeting precision converts directly to outcome.