# Lookalike Model

### About this export

| Field | Value |
| --- | --- |
| **content_type** | lesson |
| **platform** | contentstack-academy |
| **source_url** | https://www.contentstack.com/academy/courses/lytics-implementation/lookalike-model |
| **course_slug** | lytics-implementation |
| **lesson_slug** | lookalike-model |
| **markdown_file_url** | /academy/md/courses/lytics-implementation/lookalike-model.md |
| **generated_at** | 2026-08-07T05:57:40.312Z |

> Part of **[Lytics Implementation](https://www.contentstack.com/academy/courses/lytics-implementation)** on Contentstack Academy. **Academy MD v3** — structured for retrieval; no quiz or assessment keys.

<!-- ai_metadata: {"lesson_id":"12","type":"text","duration_minutes":14,"topics":["Lookalike","Model"]} -->

#### Lesson text

Lookalike modeling in Lytics extends your rule-based audience strategy by scoring profiles on their similarity to a target audience. When manual segment logic is too coarse and you need the platform to learn high-signal combinations of behavior and profile attributes, a lookalike model gives you a precision tool that writes directly back to profile fields — making its output available everywhere audiences are used. This section covers the full model lifecycle: framing goals, configuring and training a model, interpreting its diagnostics, and activating predictive audiences downstream.

## Overview

### Learning Objectives

By the end of this section, you will be able to:  
\- Explain what lookalike models are and how they identify similar users  
\- Describe when to use lookalike modeling versus other audience strategies  
\- Understand the business value of lookalikes for expanding reach and improving targeting

### Overview

Lytics **Lookalike Models** are propensity models that score users in a **Source Audience** — the audience you want to score and take action on — based on how similar they are to a **Target Audience** — the audience that represents the outcome you want more users to resemble. You use them when rule-based audience logic is too coarse and you want the platform to learn higher-signal combinations of behavior and profile attributes.

In current UI navigation, model management is under **Laboratory > Lookalike Models** (`/laboratory/lookalike-models`).

#### Key Concepts

*   **Lookalike models are profile-native**: scores are written back to profile meta fields, so they are usable anywhere audiences are usable.
*   **Model outputs are explicit fields**:
*   `segment_prediction` (map of model score values, typically 0-1)
*   `segment_prediction_percentile` (map of percentile ranks, integer scale)
*   **ML augments, not replaces, audience strategy**: you still define clear business intent through source/target selection.
*   **Activation controls scoring lifecycle**: building a model and activating scoring are separate operational steps.

#### Step-by-Step

1.  Define the business question in source-to-target form:  
    \- `unknown_users -> known_users_with_email`  
    \- `single_purchasers -> repeat_purchasers`
2.  Confirm both audiences are valid and stable in **Audience Builder**.
3.  Go to **Laboratory > Lookalike Models** and click **Create New**.
4.  Configure source/target and initial model options.
5.  Review model status (`building`, `complete`, `active`, etc.) and summary diagnostics once complete.
6.  Activate the model and build predictive audiences from model score/percentile fields.

#### Examples

```text
Appropriate lookalike scenarios:
- You need better precision than broad lifecycle audiences
- Feature interactions are too complex for manual segment rules
- You want dynamic audience membership as behavior changes
```

```text
Lookalike anti-patterns:
- Source/target definitions are unclear or unstable
- Audience populations are too small to train reliably
- You have no clear activation destination or downstream action
```

#### Diagrams & Screenshots

![Lookalike model workflow from source/target selection to scoring and predictive audience activation.](https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/amcab82ebd9a8794e1/984a3252933be7c0494877cf/lookalike_model_workflow_source_target.png)

### Summary

Lookalike models score every profile in your source audience on its similarity to a target audience, writing those scores back as profile fields that are immediately usable in audience rules. They are most valuable when manual rule logic is too coarse or when feature interactions are complex enough that a machine learning approach outperforms manual segmentation. The key governance discipline is keeping source and target definitions clear and stable before you build — model quality is mostly determined at goal-definition time, not at configuration time.

### Documentation Links

*   [Lookalike Models Overview](https://docs.lytics.com/docs/lookalike-models)
*   [Predictive Audiences](https://docs.lytics.com/docs/predictive-audiences)

## How Models Work

### Learning Objectives

By the end of this section, you will be able to:  
\- Understand the machine learning approach used by lookalike models at a high level  
\- Identify what data feeds the model, including behavioral and profile attributes  
\- Explain model confidence scores and how they rank lookalike candidates

### How Models Work

Lytics trains supervised models to distinguish target-like behavior from non-target behavior using profile features. You control inputs at configuration time; Lytics handles feature processing, training, evaluation, and score persistence.

#### Key Concepts

*   **Feature inputs** can include:
*   Behavioral scores (`use_scores`)
*   Content affinity (`use_content`)
*   Additional schema fields (`additional`)
*   Field exclusions (`blocked`)
*   **Model workflow separation**:
*   `Model Training Only` builds model artifacts.
*   Active scoring evaluates profiles and updates prediction fields.
*   **Size guardrails matter**:
*   Minimum source/target threshold is enforced (25 users).
*   Large audiences can trigger alternate modeling behavior and performance tradeoffs.
*   **Health and diagnostics are first-class outputs**:
*   Summary view surfaces model health, accuracy, reach, and feature composition.
*   Diagnostics tab includes `R Squared`, `MSE`, `AUC`, error rates, and decision threshold.

#### Step-by-Step

1.  Open a completed model and review the **Summary** tab first.
2.  Confirm the model is healthy enough for activation:  
    \- confidence/health state acceptable,  
    \- non-trivial accuracy and reach balance,  
    \- no major warning messages.
3.  Open **Configuration** to confirm included/excluded features match your intent.
4.  Open **Diagnostics** to inspect:  
    \- `R Squared`  
    \- `Mean Squared Error`  
    \- `False Positive/Negative Rate`  
    \- `Area Under Curve`  
    \- `Decision Threshold`
5.  Use **Feature Importance and Correlation** charts to identify which fields are driving predictions.

#### Examples

```text
Example interpretation:
- High accuracy + low reach: best for narrow, high-intent targeting
- Lower accuracy + high reach: best for upper-funnel expansion
- Unhealthy model messages: revisit source/target adjacency or feature set
```

```lql
FILTER AND (
  INCLUDE unknown_users,
  segment_prediction.`unknown_to_known_model` >= 0.50
) FROM user
```

#### Diagrams & Screenshots

![Lookalike model summary view with audience cards, threshold explorer, and feature importance charts.](https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/amde3193a79426d72a/bbba365fe3b7975fb7a0aa4e/lookalike_model_summary_view_audience.png)

### Summary

Lytics lookalike models are supervised classifiers trained on the profile features you specify. Behavioral scores, content affinity, and selected schema fields are the primary inputs; feature processing, training, and score persistence happen automatically. After build, the Summary and Diagnostics tabs give you accuracy, reach, R Squared, AUC, and error rate signals you use to decide whether the model is ready for activation. Feature importance charts let you verify that the predictors driving the model are meaningful and acceptable from a governance perspective.

### Documentation Links

*   [Model Diagnostics](https://docs.lytics.com/docs/lookalike-model-diagnostics)
*   [Feature Importance](https://docs.lytics.com/docs/lookalike-feature-importance)
*   [Behavioral Scoring](https://docs.lytics.com/docs/behavioral-scoring)

## Define Goals

### Learning Objectives

By the end of this section, you will be able to:  
\- Define clear model objectives tied to business outcomes  
\- Choose the right seed audience based on size, quality, and relevance  
\- Set success criteria for model performance including target reach and precision

### Define Goals

Lookalike performance is mostly determined before you click build. Clear goal definition and correct source/target design are the most important controls you have.

#### Key Concepts

*   **Frame goals as movement between audience states**: from current state (source) to desired state (target).
*   **Choose adjacent lifecycle stages**: adjacency improves learnability and actionable overlap.
*   **Define success before build**:
*   expected audience size at initial threshold,
*   expected lift/conversion proxy,
*   activation channel and cadence.
*   **Keep targets behaviorally meaningful**: avoid purely administrative fields that do not represent real outcomes.

#### Step-by-Step

1.  Write a one-line objective:  
    \- "Increase users likely to become newsletter subscribers."
2.  Choose source and target audiences that reflect this objective.
3.  Validate each audience:  
    \- clear definition,  
    \- enough users,  
    \- current and regularly refreshed.
4.  Define threshold strategy in advance:  
    \- start near default decision threshold for balanced performance,  
    \- lower threshold for reach,  
    \- higher threshold for precision.
5.  Define activation plan:  
    \- destinations/flows/Pathfora experiences that will use the resulting audiences.
6.  Document model ownership and review cadence.

#### Examples

```text
Goal template:
- Business outcome: <what changes>
- Source audience: <who to score>
- Target audience: <who to emulate>
- Activation channel: <where action happens>
- Success metric: <how performance is judged>
```

```text
Common objective patterns:
- unknown -> known
- one-time purchaser -> repeat purchaser
- engaged browser -> converter
- recently active customer -> retained customer
```

#### Diagrams & Screenshots

![Goal definition worksheet showing source, target, threshold strategy, and activation plan.](https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/am27cf6e0d1f4a70e3/3a32acb07e6c181615cdc711/goal_definition_worksheet_source_target.png)

### Summary

Model goal definition is the highest-leverage step in the lookalike workflow. Framing goals as movement between adjacent lifecycle states — unknown to known, browser to converter, single purchaser to repeat purchaser — gives the model a learnable signal and keeps the downstream activation plan concrete. Define your expected audience size, precision target, and activation channel before you build, so you have a clear success benchmark to evaluate against once diagnostics are available.

### Documentation Links

*   [Lookalike Model Configuration](https://docs.lytics.com/docs/lookalike-model-configuration)
*   [Audience Building](https://docs.lytics.com/docs/audiences)

## Configure Model

### Learning Objectives

By the end of this section, you will be able to:  
\- Configure model parameters in the Lytics UI including seed audience and scope  
\- Select features for the model that align with your modeling goals  
\- Launch a model run and monitor its progress to completion

### Configure Model

Configuration is driven by workflow `lytics_seg_model` in the UI wizard and is split into required basic settings and optional advanced controls.

#### Key Concepts

*   **Required inputs**:
*   `source`
*   `target`
*   **Common advanced controls**:
*   `auto_tune`
*   `build_only` (training-only mode)
*   `use_scores`
*   `use_content`
*   `additional`
*   `blocked`
*   `re_run` (continuous retrain)
*   `collect` (sample size)
*   **Activation is explicit**: after build, click **Activate** to apply scoring updates on profiles.
*   **Model tabs for operations**:
*   **Summary**, **Audiences**, **Configuration**, **Diagnostics**

#### Step-by-Step

1.  Go to **Laboratory > Lookalike Models > Create New**.
2.  Set **Source Audience** and **Target Audience**.
3.  Optionally set **Custom Model Name** (recommended for governance).
4.  Decide whether to start in **Model Training Only** mode.
5.  Configure feature strategy:  
    \- start with `use_scores` + `use_content`,  
    \- add carefully selected `additional` fields if needed,  
    \- use `blocked` to remove leakage/noise.
6.  Enable **Auto Tune** if manual features are not producing healthy models.
7.  Save and monitor build status.
8.  Open the model and click **Activate** when diagnostics are acceptable.

#### Examples

```bash
curl -s -XPOST "https://api.lytics.io/ml" \
  -H "Authorization: $LIOKEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "unknown_to_known_model",
    "source": "aud_unknown_users",
    "target": "aud_known_users_with_email",
    "config": {
      "auto_tune": true,
      "build_only": true,
      "use_scores": true,
      "use_content": true,
      "collect": 5000,
      "re_run": false
    }
  }'
```

```bash
curl -s -XGET "https://api.lytics.io/ml" -H "Authorization: $LIOKEY"
curl -s -XGET "https://api.lytics.io/ml/unknown_to_known_model/summary" -H "Authorization: $LIOKEY"
```

#### Diagrams & Screenshots

![Lookalike model configuration wizard showing source/target and advanced options.](https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/am2743f4cb06638c5c/4452c4302d63af5eba247509/lookalike_model_configuration_wizard_source.png)

### Summary

Model configuration in the Lytics UI wizard requires only a source and target audience; everything else is optional but consequential. Starting with `use_scores` and `use_content` gives the model a strong behavioral signal without overfitting to niche fields. Use `blocked` to remove any fields that would introduce data leakage, and enable `auto_tune` when manual feature selection isn't producing healthy diagnostics. Activation is a separate, explicit step after build — build and activate are intentionally decoupled so you can review diagnostics before committing to production scoring.

### Documentation Links

*   [Lookalike Model API](https://docs.lytics.com/reference/lookalike-model-api)
*   [Model Configuration Options](https://docs.lytics.com/docs/lookalike-model-configuration)

## Understanding Your Model

### Learning Objectives

By the end of this section, you will be able to:  
\- Interpret model results including score distributions and audience overlap  
\- Analyze feature importance to understand what drives the model's predictions  
\- Evaluate model quality and coverage to determine if the model meets success criteria

### Understanding Your Model

After build, your job is to decide whether the model is fit for activation and how aggressively to threshold it in audience creation.

#### Key Concepts

*   **Health summary is directional, not absolute**: use multiple signals before activating.
*   **Decision threshold sets precision vs reach**:
*   lower threshold increases audience size,
*   higher threshold increases selectivity.
*   **Predefined percentile audiences speed evaluation**:
*   **Unlikely**: below 25th percentile
*   **Somewhat Likely**: above 50th percentile
*   **Likely**: above 75th percentile
*   **Highly Likely**: above 90th percentile
*   **Feature importance helps governance**: verify top predictors are meaningful and acceptable from privacy/compliance perspective.

#### Step-by-Step

1.  In **Summary**, inspect health card values for `Accuracy`, `Reach`, and `R Squared`.
2.  Use model exploration to compare source/target overlap and percentile segments.
3.  Adjust the threshold slider to estimate reachable population at different cutoffs.
4.  Inspect feature importance/correlation charts and confirm model logic is plausible.
5.  Review **Diagnostics** for error profile and threshold values.
6.  If model quality is weak:  
    \- revisit source/target adjacency,  
    \- expand relevant features,  
    \- use Auto Tune,  
    \- retrain and compare again.

#### Examples

```lql
FILTER AND (
  INCLUDE unknown_users,
  segment_prediction_percentile.`unknown_to_known_model` >= 90
) FROM user
```

```text
Interpretation rule of thumb:
- Need scale: start with 50-75 percentile
- Need conversion efficiency: start with 90+ percentile
- Then tune using observed campaign outcomes
```

#### Diagrams & Screenshots

![Model diagnostics view with R squared, AUC, error rates, and decision threshold.](https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/am8bb25208a6d75565/3fbaa83d3f77c3a4ff138897/model_diagnostics_view_r_squared.png)

### Summary

Model evaluation is a multi-signal process: start with the health summary, then validate accuracy and reach against your pre-defined success criteria, then inspect feature importance for governance. The threshold slider is your primary precision/reach control — lower thresholds expand the reachable population while higher thresholds narrow it to the highest-confidence profiles. If the model doesn't meet your criteria, revisit source/target adjacency and feature selection before retraining rather than accepting a weak model for production activation.

### Documentation Links

*   [Model Diagnostics](https://docs.lytics.com/docs/lookalike-model-diagnostics)
*   [Threshold Explorer](https://docs.lytics.com/docs/lookalike-threshold-explorer)
*   [Feature Importance](https://docs.lytics.com/docs/lookalike-feature-importance)

## Activating

### Learning Objectives

By the end of this section, you will be able to:  
\- Create audiences from lookalike scores using score thresholds  
\- Export lookalike audiences to advertising and marketing destinations  
\- Measure lookalike campaign performance and iterate on targeting

### Activating

Activation operationalizes the model by writing/updating prediction fields on profiles and enabling predictive audience workflows for downstream channels.

#### Key Concepts

*   **Activation switch behavior**:
*   **Activate**: starts profile scoring and audience eligibility updates.
*   **Deactivate**: stops future scoring; existing stored values remain until changed by future processing.
*   **Audience creation paths**:
*   quick editor in model view,
*   full audience builder using prediction/percentile fields.
*   **Model usage tracking**: **Audiences** tab lists audiences tied to a model.
*   **Activation should be tied to execution**: flows, exports, and on-site personalization should be prepared before turning on production targeting.

#### Step-by-Step

1.  Open model view and click **Activate**.
2.  Wait for scoring to populate/update profile fields (`segment_prediction*` maps).
3.  Create predictive audiences:  
    \- start with one precision audience (for example percentile >= 90),  
    \- optionally create one reach audience (for example percentile >= 60).
4.  Connect these audiences to destinations in [Integrations & Connections](/academy/courses/lytics-course/integrations-connections) and personalization surfaces in [Pathfora](/academy/courses/lytics-course/pathfora).
5.  Track performance against your goal and tune thresholds/features.
6.  If needed, enable `re_run` retraining cadence and repeat evaluation.

#### Examples

```bash
curl -s -XPUT "https://api.lytics.io/ml/unknown_to_known_model" \
  -H "Authorization: $LIOKEY" \
  -H "Content-Type: application/json" \
  -d '{"is_active": true}'
```

```text
Two-audience activation pattern:
- High intent: percentile >= 90 (email/high-cost channels)
- Expansion: percentile >= 60 (onsite/email nurture)
```

#### Diagrams & Screenshots

![Activation workflow from model status toggle to predictive audience exports and campaign measurement.](https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/am0fb32c4c0bfc4477/d45f06c90163a56e686b0445/activation_workflow_model_status_toggle.png)

### Summary

Activating a model starts profile scoring and makes prediction fields available for audience rules. The recommended pattern is a two-tier approach: a high-precision audience at the 90th percentile for high-cost channels and a broader reach audience at the 60th percentile for nurture and onsite experiences. Prepare your destination connections and flow configurations before activating in production, and establish a review cadence to tune thresholds as campaign performance data accumulates.

### Documentation Links

*   [Activating a Lookalike Model](https://docs.lytics.com/docs/lookalike-activation)
*   [Audience Export](https://docs.lytics.com/docs/audience-export)
*   [Flows](https://docs.lytics.com/docs/flows)

## What You've Learned

In this section you covered the full Lytics lookalike model lifecycle. You learned how propensity models score profiles in a source audience based on their similarity to a target audience, and why clear source/target design is the highest-leverage step in the workflow. You walked through feature selection and configuration, learned to read model diagnostics — accuracy, reach, R Squared, AUC, decision threshold — and practiced the threshold strategy that balances precision against reach. Finally, you saw how activation writes scores to profile fields and enables the two-tier audience pattern that serves both high-intent and upper-funnel channels.

### Key Terms

> 📘 **Source Audience** — The audience you want to score and take action on. All profiles in the source audience receive a prediction score after the model is activated.
> 
> 📘 **Target Audience** — The audience that represents the outcome you want more users to resemble. The model learns what distinguishes target members from non-members.
> 
> 📘 **segment\_prediction** — A profile meta field (map) that stores raw model score values, typically on a 0–1 scale, keyed by model name.
> 
> 📘 **segment\_prediction\_percentile** — A profile meta field (map) that stores integer percentile ranks for each model, used for relative threshold-based audience creation.
> 
> 📘 **Decision Threshold** — The score cutoff used to classify a profile as target-like or not. Lowering the threshold increases audience reach; raising it increases precision.
> 
> 📘 **Auto Tune** — A model configuration option that allows Lytics to automatically select and weight features rather than relying on a manually specified feature set.
> 
> 📘 **AUC (Area Under Curve)** — A model quality metric measuring the classifier's ability to distinguish target-like from non-target profiles across all possible thresholds. Higher is better.

#### Key takeaways

- Connect **Lookalike Model** back to your stack configuration before moving to the next module.
- Capture one concrete artifact (screenshot, Postman call, or code snippet) that proves the step works in your environment.
- Re-read the delivery versus management boundary for anything you changed in the entry model.

## Supplement for indexing

### Content summary

Lookalike Model. Lookalike modeling in Lytics extends your rule-based audience strategy by scoring profiles on their similarity to a target audience. When manual segment logic is too coarse and you need the platform to learn high-signal combinations of behavior and profile attributes, a lookalike model gives you a precision tool that writes directly back to profile fields — making its output available everywhere audiences are used. This section covers the full model lifecycle: framing goals, configuring and training a model, interpreting its diagnostics, and activating predictive audiences downstream. Overview Learning Objectives By the end of this section, you will be able to: \- Explain what lookalike mode

### Retrieval tags

- Lookalike
- Model
- lytics-implementation
- lesson 12
- Lookalike Model
- lytics-implementation lesson

### Indexing notes

Index this lesson as a primary chunk tagged with lesson_id "12" and topics: [Lookalike, Model].
Parent course slug: lytics-implementation. Use asset_references URLs as thumbnail hints in search results when present.
Never surface LMS quiz content or assessment answers from this file.

### Asset references

| Label | URL |
| --- | --- |
| Lookalike model workflow from source/target selection to scoring and predictive audience activation. | `https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/amcab82ebd9a8794e1/984a3252933be7c0494877cf/lookalike_model_workflow_source_target.png` |
| Lookalike model summary view with audience cards, threshold explorer, and feature importance charts. | `https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/amde3193a79426d72a/bbba365fe3b7975fb7a0aa4e/lookalike_model_summary_view_audience.png` |
| Goal definition worksheet showing source, target, threshold strategy, and activation plan. | `https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/am27cf6e0d1f4a70e3/3a32acb07e6c181615cdc711/goal_definition_worksheet_source_target.png` |
| Lookalike model configuration wizard showing source/target and advanced options. | `https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/am2743f4cb06638c5c/4452c4302d63af5eba247509/lookalike_model_configuration_wizard_source.png` |
| Model diagnostics view with R squared, AUC, error rates, and decision threshold. | `https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/am8bb25208a6d75565/3fbaa83d3f77c3a4ff138897/model_diagnostics_view_r_squared.png` |
| Activation workflow from model status toggle to predictive audience exports and campaign measurement. | `https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/am0fb32c4c0bfc4477/d45f06c90163a56e686b0445/activation_workflow_model_status_toggle.png` |

### External links

| Label | URL |
| --- | --- |
| Contentstack Academy home | `https://www.contentstack.com/academy/` |
| Training instance setup | `https://www.contentstack.com/academy/training-instance` |
| Academy playground (GitHub) | `https://github.com/contentstack/contentstack-academy-playground` |
| Contentstack documentation | `https://www.contentstack.com/docs/` |
| Lookalike model workflow from source/target selection to scoring and predictive audience activation. | `https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/amcab82ebd9a8794e1/984a3252933be7c0494877cf/lookalike_model_workflow_source_target.png` |
| Lookalike Models Overview | `https://docs.lytics.com/docs/lookalike-models` |
| Predictive Audiences | `https://docs.lytics.com/docs/predictive-audiences` |
| Lookalike model summary view with audience cards, threshold explorer, and feature importance charts. | `https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/amde3193a79426d72a/bbba365fe3b7975fb7a0aa4e/lookalike_model_summary_view_audience.png` |
| Model Diagnostics | `https://docs.lytics.com/docs/lookalike-model-diagnostics` |
| Feature Importance | `https://docs.lytics.com/docs/lookalike-feature-importance` |
| Behavioral Scoring | `https://docs.lytics.com/docs/behavioral-scoring` |
| Goal definition worksheet showing source, target, threshold strategy, and activation plan. | `https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/am27cf6e0d1f4a70e3/3a32acb07e6c181615cdc711/goal_definition_worksheet_source_target.png` |
| Lookalike Model Configuration | `https://docs.lytics.com/docs/lookalike-model-configuration` |
| Audience Building | `https://docs.lytics.com/docs/audiences` |
| Lookalike model configuration wizard showing source/target and advanced options. | `https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/am2743f4cb06638c5c/4452c4302d63af5eba247509/lookalike_model_configuration_wizard_source.png` |
| Lookalike Model API | `https://docs.lytics.com/reference/lookalike-model-api` |
| Model diagnostics view with R squared, AUC, error rates, and decision threshold. | `https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/am8bb25208a6d75565/3fbaa83d3f77c3a4ff138897/model_diagnostics_view_r_squared.png` |
| Threshold Explorer | `https://docs.lytics.com/docs/lookalike-threshold-explorer` |
| Activation workflow from model status toggle to predictive audience exports and campaign measurement. | `https://images.contentstack.io/v3/assets/bltebc53cfaf0dd6403/am0fb32c4c0bfc4477/d45f06c90163a56e686b0445/activation_workflow_model_status_toggle.png` |
| Activating a Lookalike Model | `https://docs.lytics.com/docs/lookalike-activation` |