Adobe Target Machine Learning Setup Recommendations & Tips

[#1]

AUTO-PERSONALIZATION ACTIVITY SETUP RECOMMENDATIONS | ACTIVITY QUALIFICATION

Adobe Target's auto-personalization activity type, which is sort of the multi-variate version of machine learning, has a bunch of unique and useful features. AP, as with all Target activities, has the ability to target at the activity level. For example, if you want only visitors in California to qualify for the activity, you can set the audience qualification criteria with a geo-based audience.
With auto-personalization, you also have the ability to set what we usually refer to as “offer level targeting”. In other words, certain offers can be set up with qualification audiences. For example, you might have a credit card offer, but you only want it available to visitors that do not currently have that credit card.
One critical thing to keep in mind is that every visitor must have an available offer in every location to actually qualify for the whole activity.

TIP/RECOMMENDATION: In Auto-Personalization, a visitor needs to qualify for the complete experience. In other words, in an AP activity, visitors need to be able to qualify for an offer in every location, or they will not qualify for the activity.

SOLUTIONS:

• Confirm that the audience(s) you want to target have available offers.
• Add a fallback offer available to every targeted visitor.
• Leave as is and understand that your traffic may be lower than expected since some visitors will not qualify for offers in every location and therefore not qualify for the activity.

ACTIVITY TYPE: AUTO-PERSONALIZATION (AP)

SETUP TYPE: VISUAL EXPERIENCE COMPOSER (VEC) + FORM BASED COMPOSER

AUTO PERSONALIZATION SETUP >>

MACHINE LEARNING ACTIVITY SETUP RECOMMENDATIONS | TRAFFIC REQUIREMENTS

As with any machine learning model, Target’s auto-target and auto-personalization activity types rely on a minimum level of visitor interactions to build or learn accurate ML models. When planning for machine learning, it’s critical to consider how many visitors and conversion events will take place in the test location. Sometimes you may decide that an ML activity is not the right strategy based solely on traffic.

TIP/RECOMMENDATION: Both Auto-Target and Auto-Personalization have minimum traffic requirements for the machine learning (ML) models to build.

Simple rules of thumb can be used to understand traffic requirements for both ML activity types:

• When Conversion is your success metric: 1,000 visits and at least 50 conversions per day per experience, and in addition, the activity must have a total of 7,000 visits and 350 conversions.
• When Revenue per Visit is your success metric: 1,000 visits and at least 50 conversions per day per experience, and in addition, the activity must have a total of 1,000 conversions per experience. RPV requires more data to build models due to the higher data variance that typically exists in visit revenue compared to conversion rate.

SOLUTIONS:

• Confirm that you have enough visitors and conversions to support the number of experiences or offer variations in your activity.
• For Auto-Personalization, see the following section for information about using reporting groups. These groups consolidate visitors and conversions into fewer offer variations.
• Reduce experiences or offer variations.

ACTIVITY TYPE: AUTO-TARGET / AUTO-PERSONALIZATION (AP)

SETUP TYPE: VISUAL EXPERIENCE COMPOSER (VEC) + FORM BASED COMPOSER

AUTO PERSONALIZATION OVERVIEW >>

AUTO TARGET OVERVIEW >>

AUTO-PERSONALIZATION ACTIVITY SETUP RECOMMENDATIONS | REPORT GROUPS

Another unique feature of auto-personalization activities is the reporting group function. Reporting groups can be a powerful tool that aids in simplifying results analysis and model building by combining experiences or offers together into a single machine learning model. However, using this feature may have unintended consequences when it comes to delivering the most personalized and effective content.

TIP/RECOMMENDATION: In AP setup we have the option to set up reporting groups. Reporting groups have several notes, drawbacks and benefits, including:

• Don’t use reporting groups unless necessary: Letting Target’s machine learning capabilities deliver content without constraints allows the machine learning models to deliver the most effective content possible to each visitor profile. In some cases, using report groups may negatively impact activity effectiveness.
• Fewer machine learning models: One of the primary benefits of using reporting groups is that there will be fewer ML models. In situations where there is low visitor traffic and/or a low conversion rate, using reporting groups will consolidate  the number of offer variations and therefore increase visitors and conversions to offer groups that share the same ML model. The models will learn faster and be able to deliver personalized content sooner.
• Updates on the fly: Another benefit of reporting groups is when you plan to update content offers while the activity is running. Reporting groups are a way to collect similar content offers into a single ML model. If you plan to update content offers while the activity is running, using reporting groups will allow the same model to be used for updated content and therefore not require the traditional learning period. This should only be done if the new or updated content offer is materially the same.
• Simplified reporting: Using reporting groups may simplify activity reporting since there are fewer ML model variations to analyze. Not typically a reason to use reporting groups, but could be considered a side benefit.

ACTIVITY TYPE: AUTO-PERSONALIZATION (AP)

SETUP TYPE: VISUAL EXPERIENCE COMPOSER (VEC) + FORM BASED COMPOSER

REPORTING GROUPS DOCUMENTATION >>

AUTO-PERSONALIZATION ACTIVITY SETUP RECOMMENDATIONS | ALLOW/DISALLOW DUPLICATES

TIP/RECOMMENDATION: In AP setup, there is the option to allow or disallow duplicate content offers across activity locations. Details of this feature include:

• Duplicate content is defined by offers having the exact same name but in different selected content locations. Note that unless the content name is exactly the same, this function will not recognize the duplicated content.
• This function may come into play where, for example, you have the same content on different pages and you want to allow duplicates to show multiple times. The opposite use-case might be where you want to disallow duplicate content from appearing in multiple locations on the same page.
• One of the most interesting use-cases may be where you want to let machine learning deliver a personalized content order and inclusion of content at the same time. I'll be posting a future article on personalized content sort order.

ACTIVITY TYPE: AUTO-PERSONALIZATION (AP)

SETUP TYPE: VISUAL EXPERIENCE COMPOSER (VEC) + FORM BASED COMPOSER

DUPLICATE OFFERS DOCUMENTATION >>

AUTO-PERSONALIZATION & AUTO-TARGET ACTIVITY SETUP RECOMMENDATIONS | TRAFFIC ALLOCATION SETTINGS

TIP/RECOMMENDATION: In both Auto-Personalization and Auto-Target, you can use the preset traffic allocations or set your own allocation using customize allocation. There are nuances to allocation settings that only apply to machine learning activities in Target:

• Machine learning activities in Target require an initial learning period before delivering personalized content to each visitor profile. During this learning or model building period, Target needs to deliver content randomly to understand how visitors react.
• To learn quickly, Target will deliver up to 85% of traffic to random content during the learning period regardless of the traffic allocation setting. This means that from the initial activation of the activity, the allocation can be set at 10/90 or 10 random and 90 personalized. There is no need to set different allocations for the learning period or the personalization period after the models have built. In most cases, this is the suggested approach.
• ML activities have a minimum traffic and conversion requirement. In situations where visitors and/or conversions are low, you can use a custom allocation of 99/1 (99 control/1 personalized) to help the ML models learn slightly quicker. Keep in mind that if you don’t switch to a more personalized traffic allocation once the models are built, you may be giving up personalization and related lift. If you use this approach, you’ll need to check the activity on a daily basis to see when the models are built and adjust traffic allocation to deliver personalization to more visitors.

ACTIVITY TYPE: AUTO-PERSONALIZATION (AP) & AUTO-TARGET (AT)

SETUP TYPE: VISUAL EXPERIENCE COMPOSER (VEC) + FORM BASED COMPOSER

ALLOCATION DOCUMENTATION >>

---

Are you looking to optimize your website's performance and drive conversions?
Look no further than Adobe - the industry leader in A/B testing. With more A/B tests under our belt than any other organization in the world, we have the expertise and knowledge to take your online presence to the next level. Adobe Target is unrivaled in the industry, and with Adobe Consulting, you can have access to this powerful tool and the expertise behind it. Don't miss out on the opportunity to unlock the full potential of your website. Contact us today and discover how Adobe can help your business thrive. If you would like to understand how Adobe can help your business reach out to Adobe Professional Services.