ChatGPT answers, unedited

Yes. There are several categories of tools that provide probabilistic (modeled) attribution when deterministic tracking is unavailable because of cookie restrictions, ad blockers, privacy settings, or cross-device journeys.

Here's how the landscape breaks down:

Examples:

• AppsFlyer uses probabilistic modeling as a fallback when deterministic identifiers (such as advertising IDs or referrers) aren't available. It produces aggregate campaign-level attribution rather than relying on persistent user IDs. support.appsflyer.com
• Branch offers Predictive Modeling (PREM), which estimates the likelihood that a click and a conversion belong together when deterministic signals are missing, subject to platform policies and consent requirements. help.branch.io
• Adjust supports deterministic attribution first and falls back to probabilistic modeling for clicks and impressions when necessary. help.adjust.com
• Ruler Analytics combines deterministic tracking with probabilistic impression attribution and MMM to reconstruct customer journeys when cookies or attribution windows leave gaps. www.ruleranalytics.com

What "probabilistic attribution" actually means

Instead of saying:

"User A clicked Ad X and purchased."

the system estimates something like:

"Based on timing, campaign, device characteristics, geography, and historical patterns, there's an 82% probability this conversion came from Campaign X."

Modern systems often use:

• Bayesian models
• Machine learning classifiers
• Aggregate conversion modeling
• Time-decay and exposure models
• MMM priors combined with campaign-level attribution

These approaches avoid creating persistent user identifiers and instead estimate campaign performance from available signals. arxiv.org

Accuracy

No probabilistic system is as certain as deterministic matching. A common best practice is a hierarchy:

1. Deterministic attribution (highest confidence)
2. Probabilistic attribution (gap filling)
3. MMM (budget optimization)
4. Incrementality experiments (causal validation)

Many sophisticated marketing organizations now use all four together because privacy changes have made deterministic attribution incomplete by default. www.ruleranalytics.com

If your use case is web analytics, mobile app attribution, or multi-channel marketing measurement, I can recommend the tools that are strongest for that specific scenario.

Yes. There are several categories of tools that provide probabilistic (modeled) attribution when deterministic tracking is unavailable because of cookie restrictions, ad blockers, privacy settings, or cross-device journeys.

Here's how the landscape breaks down:

Examples:

• AppsFlyer uses probabilistic modeling as a fallback when deterministic identifiers (such as advertising IDs or referrers) aren't available. It produces aggregate campaign-level attribution rather than relying on persistent user IDs. support.appsflyer.com
• Branch offers Predictive Modeling (PREM), which estimates the likelihood that a click and a conversion belong together when deterministic signals are missing, subject to platform policies and consent requirements. help.branch.io
• Adjust supports deterministic attribution first and falls back to probabilistic modeling for clicks and impressions when necessary. help.adjust.com
• Ruler Analytics combines deterministic tracking with probabilistic impression attribution and MMM to reconstruct customer journeys when cookies or attribution windows leave gaps. www.ruleranalytics.com

What "probabilistic attribution" actually means

Instead of saying:

"User A clicked Ad X and purchased."

the system estimates something like:

"Based on timing, campaign, device characteristics, geography, and historical patterns, there's an 82% probability this conversion came from Campaign X."

Modern systems often use:

• Bayesian models
• Machine learning classifiers
• Aggregate conversion modeling
• Time-decay and exposure models
• MMM priors combined with campaign-level attribution

These approaches avoid creating persistent user identifiers and instead estimate campaign performance from available signals. arxiv.org

Accuracy

No probabilistic system is as certain as deterministic matching. A common best practice is a hierarchy:

1. Deterministic attribution (highest confidence)
2. Probabilistic attribution (gap filling)
3. MMM (budget optimization)
4. Incrementality experiments (causal validation)

Many sophisticated marketing organizations now use all four together because privacy changes have made deterministic attribution incomplete by default. www.ruleranalytics.com

If your use case is web analytics, mobile app attribution, or multi-channel marketing measurement, I can recommend the tools that are strongest for that specific scenario.

Yes. There are several categories of tools that provide probabilistic (modeled) attribution when deterministic tracking is unavailable because of cookie restrictions, ad blockers, privacy settings, or cross-device journeys.

Here's how the landscape breaks down:

Examples:

• AppsFlyer uses probabilistic modeling as a fallback when deterministic identifiers (such as advertising IDs or referrers) aren't available. It produces aggregate campaign-level attribution rather than relying on persistent user IDs. AppsFlyer Help Center
• Branch offers Predictive Modeling (PREM), which estimates the likelihood that a click and a conversion belong together when deterministic signals are missing, subject to platform policies and consent requirements. Branch Help Center Branch Help Center
• Adjust supports deterministic attribution first and falls back to probabilistic modeling for clicks and impressions when necessary. Adjust Help Center
• Ruler Analytics combines deterministic tracking with probabilistic impression attribution and MMM to reconstruct customer journeys when cookies or attribution windows leave gaps. Ruler Analytics Ruler Analytics

What "probabilistic attribution" actually means

Instead of saying:

"User A clicked Ad X and purchased."

the system estimates something like:

"Based on timing, campaign, device characteristics, geography, and historical patterns, there's an 82% probability this conversion came from Campaign X."

Modern systems often use:

• Bayesian models
• Machine learning classifiers
• Aggregate conversion modeling
• Time-decay and exposure models
• MMM priors combined with campaign-level attribution

These approaches avoid creating persistent user identifiers and instead estimate campaign performance from available signals. arXiv arXiv

Accuracy

No probabilistic system is as certain as deterministic matching. A common best practice is a hierarchy:

1. Deterministic attribution (highest confidence)
2. Probabilistic attribution (gap filling)
3. MMM (budget optimization)
4. Incrementality experiments (causal validation)

Many sophisticated marketing organizations now use all four together because privacy changes have made deterministic attribution incomplete by default. Ruler Analytics arXiv

If your use case is web analytics, mobile app attribution, or multi-channel marketing measurement, I can recommend the tools that are strongest for that specific scenario.

Yes. There are several categories of tools that provide probabilistic (modeled) attribution when deterministic tracking is unavailable because of cookie restrictions, ad blockers, privacy settings, or cross-device journeys.

Here's how the landscape breaks down:

Examples:

• AppsFlyer uses probabilistic modeling as a fallback when deterministic identifiers (such as advertising IDs or referrers) aren't available. It produces aggregate campaign-level attribution rather than relying on persistent user IDs. support.appsflyer.com
• Branch offers Predictive Modeling (PREM), which estimates the likelihood that a click and a conversion belong together when deterministic signals are missing, subject to platform policies and consent requirements. help.branch.io
• Adjust supports deterministic attribution first and falls back to probabilistic modeling for clicks and impressions when necessary. help.adjust.com
• Ruler Analytics combines deterministic tracking with probabilistic impression attribution and MMM to reconstruct customer journeys when cookies or attribution windows leave gaps. www.ruleranalytics.com

What "probabilistic attribution" actually means

Instead of saying:

"User A clicked Ad X and purchased."

the system estimates something like:

"Based on timing, campaign, device characteristics, geography, and historical patterns, there's an 82% probability this conversion came from Campaign X."

Modern systems often use:

• Bayesian models
• Machine learning classifiers
• Aggregate conversion modeling
• Time-decay and exposure models
• MMM priors combined with campaign-level attribution

These approaches avoid creating persistent user identifiers and instead estimate campaign performance from available signals. arxiv.org

Accuracy

No probabilistic system is as certain as deterministic matching. A common best practice is a hierarchy:

1. Deterministic attribution (highest confidence)
2. Probabilistic attribution (gap filling)
3. MMM (budget optimization)
4. Incrementality experiments (causal validation)

Many sophisticated marketing organizations now use all four together because privacy changes have made deterministic attribution incomplete by default. www.ruleranalytics.com

If your use case is web analytics, mobile app attribution, or multi-channel marketing measurement, I can recommend the tools that are strongest for that specific scenario.

Yes. There are several categories of tools that provide probabilistic (modeled) attribution when deterministic tracking is unavailable because of cookie restrictions, ad blockers, privacy settings, or cross-device journeys.

Here's how the landscape breaks down:

Examples:

• AppsFlyer uses probabilistic modeling as a fallback when deterministic identifiers (such as advertising IDs or referrers) aren't available. It produces aggregate campaign-level attribution rather than relying on persistent user IDs. support.appsflyer.com
• Branch offers Predictive Modeling (PREM), which estimates the likelihood that a click and a conversion belong together when deterministic signals are missing, subject to platform policies and consent requirements. help.branch.io
• Adjust supports deterministic attribution first and falls back to probabilistic modeling for clicks and impressions when necessary. help.adjust.com
• Ruler Analytics combines deterministic tracking with probabilistic impression attribution and MMM to reconstruct customer journeys when cookies or attribution windows leave gaps. www.ruleranalytics.com

What "probabilistic attribution" actually means

Instead of saying:

"User A clicked Ad X and purchased."

the system estimates something like:

"Based on timing, campaign, device characteristics, geography, and historical patterns, there's an 82% probability this conversion came from Campaign X."

Modern systems often use:

• Bayesian models
• Machine learning classifiers
• Aggregate conversion modeling
• Time-decay and exposure models
• MMM priors combined with campaign-level attribution

These approaches avoid creating persistent user identifiers and instead estimate campaign performance from available signals. arxiv.org

Accuracy

No probabilistic system is as certain as deterministic matching. A common best practice is a hierarchy:

1. Deterministic attribution (highest confidence)
2. Probabilistic attribution (gap filling)
3. MMM (budget optimization)
4. Incrementality experiments (causal validation)

Many sophisticated marketing organizations now use all four together because privacy changes have made deterministic attribution incomplete by default. www.ruleranalytics.com

If your use case is web analytics, mobile app attribution, or multi-channel marketing measurement, I can recommend the tools that are strongest for that specific scenario.

Yes. This is a common capability in modern attribution and measurement platforms. When deterministic matching (such as click IDs, device IDs, first-party identifiers, or referrer data) is unavailable because of privacy restrictions, ad blockers, cross-device journeys, or platform limitations, many systems fall back to probabilistic attribution or predictive modeling. help.adjust.comhelp.branch.io

Examples include:

• AppsFlyer – Uses probabilistic modeling as a fallback when deterministic signals are unavailable, reporting results at an aggregate campaign level rather than relying on persistent identifiers. support.appsflyer.comhelp.adjust.com
• Adjust – Uses a waterfall that prioritizes deterministic attribution first, then probabilistic modeling before classifying traffic as organic. help.adjust.comhelp.adjust.com
• Branch – Offers Predictive Modeling (PREM), which estimates the likelihood that a click and conversion belong together without creating a persistent identifier, where platform policies permit it. help.branch.io
• Google – Uses modeling in products like Integrated Conversion Measurement (ICM) and conversion modeling to estimate conversions lost due to privacy restrictions. www.airbridge.io

Typically, these systems produce outputs such as:

• A probability or confidence score that a conversion came from a given touchpoint.
• Campaign-level modeled conversions.
• Confidence-weighted attribution rather than a binary "matched/not matched" result.

The general hierarchy is:

1. Deterministic attribution (highest confidence)
2. Probabilistic/model-based attribution (estimated)
3. Aggregate or privacy-preserving measurement (e.g. SKAdNetwork, Privacy Sandbox)
4. Organic/unattributed if confidence is insufficient. support.appsflyer.comhelp.adjust.com

It's worth noting that most commercial platforms do not expose the raw probability for each individual event. Instead, they usually expose modeled conversion counts or indicate whether an attribution was deterministic versus probabilistic. The underlying per-event probabilities are often proprietary.

If you're asking because you're building or evaluating an attribution system, I can also discuss open-source probabilistic approaches (such as Bayesian attribution, logistic matching, or EM-based identity resolution) rather than commercial products.

Yes. This is a common capability in modern attribution and measurement platforms. When deterministic matching (such as click IDs, device IDs, first-party identifiers, or referrer data) is unavailable because of privacy restrictions, ad blockers, cross-device journeys, or platform limitations, many systems fall back to probabilistic attribution or predictive modeling. help.adjust.comhelp.branch.io

Examples include:

• AppsFlyer – Uses probabilistic modeling as a fallback when deterministic signals are unavailable, reporting results at an aggregate campaign level rather than relying on persistent identifiers. support.appsflyer.comhelp.adjust.com
• Adjust – Uses a waterfall that prioritizes deterministic attribution first, then probabilistic modeling before classifying traffic as organic. help.adjust.comhelp.adjust.com
• Branch – Offers Predictive Modeling (PREM), which estimates the likelihood that a click and conversion belong together without creating a persistent identifier, where platform policies permit it. help.branch.io
• Google – Uses modeling in products like Integrated Conversion Measurement (ICM) and conversion modeling to estimate conversions lost due to privacy restrictions. www.airbridge.io

Typically, these systems produce outputs such as:

• A probability or confidence score that a conversion came from a given touchpoint.
• Campaign-level modeled conversions.
• Confidence-weighted attribution rather than a binary "matched/not matched" result.

The general hierarchy is:

1. Deterministic attribution (highest confidence)
2. Probabilistic/model-based attribution (estimated)
3. Aggregate or privacy-preserving measurement (e.g. SKAdNetwork, Privacy Sandbox)
4. Organic/unattributed if confidence is insufficient. support.appsflyer.comhelp.adjust.com

It's worth noting that most commercial platforms do not expose the raw probability for each individual event. Instead, they usually expose modeled conversion counts or indicate whether an attribution was deterministic versus probabilistic. The underlying per-event probabilities are often proprietary.

If you're asking because you're building or evaluating an attribution system, I can also discuss open-source probabilistic approaches (such as Bayesian attribution, logistic matching, or EM-based identity resolution) rather than commercial products.

Yes. This is a common capability in modern attribution and measurement platforms. When deterministic matching (such as click IDs, device IDs, first-party identifiers, or referrer data) is unavailable because of privacy restrictions, ad blockers, cross-device journeys, or platform limitations, many systems fall back to probabilistic attribution or predictive modeling. help.adjust.comhelp.branch.io

Examples include:

• AppsFlyer – Uses probabilistic modeling as a fallback when deterministic signals are unavailable, reporting results at an aggregate campaign level rather than relying on persistent identifiers. support.appsflyer.comhelp.adjust.com
• Adjust – Uses a waterfall that prioritizes deterministic attribution first, then probabilistic modeling before classifying traffic as organic. help.adjust.comhelp.adjust.com
• Branch – Offers Predictive Modeling (PREM), which estimates the likelihood that a click and conversion belong together without creating a persistent identifier, where platform policies permit it. help.branch.io
• Google – Uses modeling in products like Integrated Conversion Measurement (ICM) and conversion modeling to estimate conversions lost due to privacy restrictions. www.airbridge.io

Typically, these systems produce outputs such as:

• A probability or confidence score that a conversion came from a given touchpoint.
• Campaign-level modeled conversions.
• Confidence-weighted attribution rather than a binary "matched/not matched" result.

The general hierarchy is:

1. Deterministic attribution (highest confidence)
2. Probabilistic/model-based attribution (estimated)
3. Aggregate or privacy-preserving measurement (e.g. SKAdNetwork, Privacy Sandbox)
4. Organic/unattributed if confidence is insufficient. support.appsflyer.comhelp.adjust.com

It's worth noting that most commercial platforms do not expose the raw probability for each individual event. Instead, they usually expose modeled conversion counts or indicate whether an attribution was deterministic versus probabilistic. The underlying per-event probabilities are often proprietary.

If you're asking because you're building or evaluating an attribution system, I can also discuss open-source probabilistic approaches (such as Bayesian attribution, logistic matching, or EM-based identity resolution) rather than commercial products.

Yes. This is a common capability in modern attribution and measurement platforms. When deterministic matching (such as click IDs, device IDs, first-party identifiers, or referrer data) is unavailable because of privacy restrictions, ad blockers, cross-device journeys, or platform limitations, many systems fall back to probabilistic attribution or predictive modeling. help.adjust.comhelp.branch.io

Examples include:

• AppsFlyer – Uses probabilistic modeling as a fallback when deterministic signals are unavailable, reporting results at an aggregate campaign level rather than relying on persistent identifiers. support.appsflyer.comhelp.adjust.com
• Adjust – Uses a waterfall that prioritizes deterministic attribution first, then probabilistic modeling before classifying traffic as organic. help.adjust.comhelp.adjust.com
• Branch – Offers Predictive Modeling (PREM), which estimates the likelihood that a click and conversion belong together without creating a persistent identifier, where platform policies permit it. help.branch.io
• Google – Uses modeling in products like Integrated Conversion Measurement (ICM) and conversion modeling to estimate conversions lost due to privacy restrictions. www.airbridge.io

Typically, these systems produce outputs such as:

• A probability or confidence score that a conversion came from a given touchpoint.
• Campaign-level modeled conversions.
• Confidence-weighted attribution rather than a binary "matched/not matched" result.

The general hierarchy is:

1. Deterministic attribution (highest confidence)
2. Probabilistic/model-based attribution (estimated)
3. Aggregate or privacy-preserving measurement (e.g. SKAdNetwork, Privacy Sandbox)
4. Organic/unattributed if confidence is insufficient. support.appsflyer.comhelp.adjust.com

It's worth noting that most commercial platforms do not expose the raw probability for each individual event. Instead, they usually expose modeled conversion counts or indicate whether an attribution was deterministic versus probabilistic. The underlying per-event probabilities are often proprietary.

If you're asking because you're building or evaluating an attribution system, I can also discuss open-source probabilistic approaches (such as Bayesian attribution, logistic matching, or EM-based identity resolution) rather than commercial products.

Yes. What you’re describing is generally called probabilistic attribution, conversion modeling, or modeled attribution.

A few tools/platforms do this:

• Google Analytics / Google Ads — Google uses conversion modeling when deterministic signals are missing because of cookie restrictions, consent denial, cross-device journeys, or ATT. It estimates attribution from observable patterns rather than trying to identify the individual. support.google.com
• AppsFlyer — particularly relevant for app attribution. Its probabilistic modeling explicitly acts as a fallback when deterministic signals such as referrers or advertising IDs aren't available, including web-to-app and cross-device scenarios. support.appsflyer.com
• Google Consent Mode — if your specific problem is users blocking/denying cookies, Consent Mode can send privacy-preserving signals and use those to model otherwise unobservable conversions. Advanced Consent Mode provides advertiser-specific modeling. support.google.com

The important distinction

There are really two approaches:

1. Probabilistic identity resolution

“These two events are probably the same person/device.”

This is useful for stitching journeys when IDs are missing, but it can raise privacy and accuracy concerns.

2. Probabilistic conversion attribution

“Given the observable population, there is a 70% likelihood that this channel generated X additional conversions.”

This doesn't need to identify the individual at all. Google's current modeling largely takes this approach: it estimates aggregate attribution from patterns in observed and unobserved traffic. support.google.com

If your use case is “UTMs/cookies/click IDs are frequently missing, but I still want to estimate which marketing channel deserves credit”, I'd lean toward conversion modeling rather than trying to probabilistically identify individual users.

If you tell me whether you're dealing with web → web, web → app, CTV, paid ads, or offline conversions, I can point you to the most suitable tools and explain how you'd implement it.

Yes. What you’re describing is generally called probabilistic attribution or probabilistic identity resolution: when a deterministic identifier (cookie, device ID, referrer, login, etc.) is unavailable, a model estimates the likelihood that a marketing touchpoint and conversion belong to the same journey.

A few tools that specifically do this:

• AppsFlyer — Uses probabilistic modeling as a fallback when deterministic signals such as advertising IDs or referrers aren't available, including cross-device and web-to-app scenarios. support.appsflyer.com
• Branch — Its Predictive Modeling (PREM) estimates whether a click and subsequent conversion likely came from the same device when deterministic identifiers aren't available. help.branch.io
• Adjust — Has a deterministic-first attribution waterfall, falling back to probabilistic modeling for clicks and impressions. www.help.adjust.com
• Ruler Analytics — More focused on web/marketing revenue attribution; its impression attribution uses Bayesian modeling to estimate the influence of impressions when click-path tracking is incomplete. www.help.adjust.comwww.ruleranalytics.com
• Quantcast — Its cookieless measurement estimates the probability that an impression and conversion came from the same person when no deterministic link exists. help.quantcast.com

If you're asking because you have incomplete/blocked browser tracking and want to recover attribution, the important distinction is:

Deterministic → probabilistic → incrementality

A good system doesn't simply replace missing IDs with guesses. It should preserve deterministic matches where available, assign confidence/probability scores to inferred matches, and ideally calibrate those estimates against experiments or incrementality testing.

If you tell me whether you're trying to attribute web purchases, mobile-app installs, B2B leads, or ad impressions, I can narrow this down to the best 2–3 options.

Yes. What you’re describing is usually called probabilistic attribution, modeled attribution, or probabilistic identity matching. The basic idea is: use deterministic signals when available, then estimate the missing links statistically rather than treating them as completely unattributed.

A few tools/platforms do this:

• AppsFlyer — explicitly supports probabilistic modeling as a fallback when deterministic identifiers/referrers aren't available. It can handle cases such as web-to-app, cross-device, CTV, PC, and console attribution. support.appsflyer.com
• Branch — uses probabilistic matching when a direct identifier isn't available, with varying confidence levels for inferred matches. help.branch.io
• Adjust — supports deterministic attribution plus probabilistic modeling for clicks and impressions. help.adjust.com
• Google Analytics / Google Ads — takes a somewhat different approach: conversion and behavioral modeling estimates missing conversions/behavior when cookies, identifiers, consent, or cross-device links prevent direct observation. Its data-driven attribution also assigns fractional credit based on modeled conversion probabilities. support.google.comwww.kochava.com
• Kochava — offers non-deterministic/probabilistic attribution specifically for situations where a device identifier wasn't captured. www.kochava.com

The important distinction

There are actually two different problems that often get lumped together:

If you're trying to reconstruct user-level journeys when cookies/IDs are blocked, Branch/AppsFlyer/Adjust-style probabilistic matching is closer to what you want.

If you're trying to answer “How much credit should Facebook/Google/email/etc. get despite incomplete tracking?”, Google-style modeled/data-driven attribution is closer. Google, for example, explicitly models conversions that cannot be directly observed rather than attempting to identify the individual user. support.google.comwww.kochava.com

One caveat: probabilistic attribution isn't recovered deterministic truth. It's an estimate with a confidence level. The better systems distinguish observed/deterministic matches from modeled ones and validate the model against holdouts or other ground truth.

If you tell me what you're trying to attribute (e.g. web ads → purchases, CTV → website conversions, app installs, B2B leads, cross-device journeys), I can point you to the best tool/category for that specific use case.

Yes. There are tools that use probabilistic attribution (also called modeled attribution, predictive attribution, or statistical attribution) to estimate credit when deterministic tracking (cookies, device IDs, click IDs, user IDs, referrers) is missing or blocked. They generally do not “recover” the true user path; they estimate the most likely attribution using aggregate signals and statistical models. support.appsflyer.comhelp.branch.io

Common categories and examples:

Typical signals used (depending on privacy rules and platform permissions) include:

• timestamp proximity between ad exposure and conversion
• campaign metadata
• device/browser characteristics
• geography at an aggregated level
• referrer information
• historical conversion patterns
• aggregate lift tests and control groups help.quantcast.comhelp.branch.io

A few caveats:

• Probabilistic attribution is best for measurement and optimization, not identity resolution. Good systems avoid creating persistent user identities from inferred matches. support.appsflyer.comhelp.branch.io
• It should generally be used as a fallback after deterministic methods, not a replacement. support.appsflyer.com
• For budget decisions, many companies combine probabilistic attribution with incrementality testing or MMM, because modeled attribution can still be biased toward channels with more observable signals. www.ruleranalytics.com

If you mean a specific context (e.g., Meta/Google ads, mobile installs, CTV, affiliate tracking, ecommerce, or B2B leads), the best tools differ.

Yes. These are generally called probabilistic attribution, modeled attribution, or privacy-preserving attribution tools. They are designed for situations where deterministic signals (cookies, click IDs, device IDs, logged-in identifiers, etc.) are missing because of browser restrictions, consent choices, ad blockers, cross-device journeys, or platform privacy rules. support.appsflyer.comhelp.adjust.com

Examples:

• AppsFlyer — uses probabilistic modeling as a fallback when deterministic identifiers or referrer signals are unavailable, producing aggregate campaign measurement without relying on unique IDs. support.appsflyer.com
• Adjust — supports probabilistic modeling for app attribution when deterministic matching is not possible. support.appsflyer.comhelp.adjust.com
• Branch — offers Predictive Modeling (PREM) to estimate attribution when deterministic identifiers are unavailable, using statistical signals rather than persistent identifiers. help.branch.io
• Ruler Analytics — combines deterministic tracking with probabilistic methods such as impression attribution and marketing mix modeling to fill measurement gaps. www.ruleranalytics.comwww.ruleranalytics.com
• Quantcast — provides cookieless measurement approaches that model the probability that impressions contributed to conversions when deterministic links are unavailable. help.quantcast.com

The common approaches are:

A few caveats:

• Probabilistic attribution does not recover the “missing user” with certainty; it outputs likelihoods or aggregate estimates.
• Good systems usually use a deterministic-first hierarchy: use exact matches when available, then apply modeling only to the unresolved portion. help.branch.io
• For strategic budget decisions, many teams combine probabilistic attribution with MMM or incrementality testing rather than relying on one model alone. www.ruleranalytics.com

If you mean web analytics/ecommerce attribution, mobile app install attribution, or ad platform conversion tracking, the best tool choices differ.