Data Deficiency Kills AI/ML Use Cases

Only 38% of marketers say they have the customer segment and persona data they need in the right format to make good marketing decisions. This figure is likely lower for the data accessible to AI/ML systems.

Several reasons may lead organisations to neglect investments in data collection. Some organisations may simply not be aware of the importance of data for AI/ML. Others may lack the resources or expertise to collect and manage data effectively. Still, others may be concerned about the privacy implications of collecting personal data.

Regardless of the cause, overlooking investment in data collection can severely impede an organisation’s capacity to leverage AI/ML for personalizing and testing digital experiences. Without sufficient data, AI/ML models will be unable to learn the patterns and trends that are necessary to make accurate predictions and recommendations. This can lead to several problems, including:

• Poor personalisation: AI/ML models that are trained on insufficient data will be unable to provide personalized experiences that are relevant and engaging to users.
• Missed opportunities: neglecting data collection risk missing out on the valuable insights that AI/ML can uncover.

Some missed opportunities in relation to Adobe Experience Cloud:

Adobe Analytics

Most organisations implement a solution design based on the top use cases which is a good practice until the implementation stagnates for years. Ideally, the data collection should evolve according to insights beyond traffic reports.

1. Adding user interactions along the key journeys.
2. Capturing intent signals separately.
3. Request that content authors to incorporate additional metadata.

In practice, this means to allocate more Analytics variables (props, eVars, events) to collect these interactions.

Segment IQ: Comparison

Segment Comparison seems like a simple concept. However, when combining with the flexibility of Adobe Analytics segments builder (sequential events, calculated metrics, etc) and clever criteria, you can easily uncover new insights. The main output is a difference score between 0 and 1.

• 1 means it is statistically significant
• 0 means there is no statistical significance

Segment Comparison calculates a difference score for the following in seconds:

• Metrics (events)
• Dimensions (values of props and eVars)
• Segments (Adobe Analyics Segments)

Contribution Analysis

Contribution Analysis discovers hidden patterns within your data to explain statistical anomalies and identify correlations behind unexpected customer actions, out-of-bound values, and sudden spikes or dips for selected metrics across convergent audience segments.

The contribution score in Contribution Analysis is calculated by combining the Cramer’s V test statistic and Pearson’s Residual for each dimension item, with the Cramer’s V measure used for weighting, and the final scores rescaled to provide a quantitative assessment of the association of each dimension item with the observed anomaly. This contribution score is in the range of 0 and 1 where closer to 1 has a higher contribution.

Contribution Analysis calculates a contribution score for the following:

• Dimensions (values of props and eVars)

The other output is Contribution Segments. These are automatically created Adobe Analytics Segments that join different combinations

Adobe Target

Although Adobe Target (AT) makes use of out-of-the-box parameters, what really is going to differentiate the performance of the algorithms compared to competitors is the uniqueness of the data.

For customers that have both Adobe Analytics (AA) and AT, there might be a misconception that all the AA data flows into the AT models. Adopting comparable analytics data collection principles is essential; however, there are instances where bypassing the analytics collection may be acceptable, especially when prioritizing personalisation use cases.

Despite AT requiring its own parameters to be ingested, the algorithms automatically make use of other Adobe Experience Cloud solutions such as:

Specific AT data collection page parameters also called mbox parameters using:

at.js

targetPageParams = function() {
  return {
    "profile.gender": "male",
    "user.categoryId": "clothing"
  };
};

Web SDK

alloy("sendEvent", {
  "data": {
    "__adobe": {
      "target": {
        "profile.gender": "male",
        "user.categoryId": "clothing"
      }
    }
  }
});

Auto-Target/Automated Personalization

Auto-Target activities in Adobe Target use advanced machine learning to select from multiple high-performing, marketer-defined experiences to personalize content and drive conversions. Auto-Target serves the most tailored experience to each visitor based on the individual customer profile and the behaviour of previous visitors with similar profiles.

In simple terms, Auto-Target can be used to optimise any A/B test with one click as part of the usual A/B test creation flow.

Automated Personalization (AP) activities in Adobe Target combine offers or messages and uses advanced machine learning to match different offer variations to each visitor based on their individual customer profile to personalize content and drive lift.

Insights Generated

Important Attributes show the top attributes that influenced the model and their relative importance. The quality and variety of the data collection can also produce actionable insights for opportunities and future hypotheses.

Automated Segments shows how different automated segments defined by Target’s personalisation models responded to the offers/experiences in the activity.

Recommendations

Recommendations activities automatically display products, services, or content that might interest your visitors based on previous user activity, preferences, or other criteria.

Specific AT data collection page parameters also called mbox parameters using:

Client side – Example of how to collect the entity parameters using:

at.js

targetPageParams = function() {
  return {
    "entity.id": "SKU-00001-LARGE",
    "entity.categoryId": "clothing,shirts",
    "entity.customEntity": "some value",
    "cartIds": "SKU-00002,SKU-00003",
    "excludedIds": "SKU-00001-SMALL"
  };
};

Web SDK

alloy("sendEvent", {
  "data": {
    "__adobe": {
      "target": {
        "entity.id": "SKU-00001-LARGE",
        "entity.categoryId": "clothing,shirts",
        "entity.customEntity": "some value",
        "cartIds": "SKU-00002,SKU-00003",
        "excludedIds": "SKU-00001-SMALL"
      }
    }
  }
});

More info here: https://experienceleague.adobe.com/docs/platform-learn/migrate-target-to-websdk/send-parameters.html?lang=en#entity-parameters

Feeds: CSV, AA Product Classifications, Google Product Search feed

Recommendations API: Add or manage the item metadata

Real-time CDP

Real-Time Customer Data Platform (Real-Time CDP) helps companies bring together known and anonymous data from multiple enterprise sources in order to create customer profiles that can be used to provide personalized customer experiences across all channels and devices in real-time.

The topic of data modelling for Real-Time CDP is very subjective and depends on the maturity, technical constraints and use cases.

RTCDP data collection is based on the XDM system and Adobe Experience Platform data connectors and data ingestion methods. To maximise the use of the feature below, the same principles above should be followed. Since the licensing considerations are not based on server calls but rather on the amount of data richness on the profiles, this could make data collection easier.

The quality of the data leveraged by the models is all based on the fields ingested. Both Look-alike audiences and Customer AI produce influential factors as shown below.

Look-alike audiences

Look-alike audiences provide intelligent insights on each of your audiences, leveraging machine-learning-based insights to identify and target high-value customers with your marketing campaigns.

Customer AI

Customer AI provides marketers with the power to generate customer predictions at the individual level with explanations. Customer AI can tell you what a customer is likely to do and why. Additionally, marketers can benefit from Customer AI predictions and insights to personalize customer experiences by serving the most appropriate offers and messaging.

Conclusion

If organisations want to be able to use AI/ML to personalize and test digital experiences effectively, they need to invest more in data collection. This includes collecting labelled behavioural data, content metadata, and other relevant data types. Organisations should also develop a data management strategy that ensures that data is collected, stored, and used responsibly and ethically.