WolfSellers — Adobe Experience Cloud Partner en México

Article

Adobe Target: Auto-Target vs Automated Personalization

When to use Auto-Target, Automated Personalization or Auto-Allocate in Adobe Target: how each one decides, how much traffic it needs, A4T, CJA and a pilot plan.

By WolfSellers··16 min read
Adobe Target: Auto-Target vs Automated Personalization
On this page

Auto-Target, Automated Personalization (AP) and Auto-Allocate are the three Adobe Target activity types that use machine learning to decide what each visitor sees. Auto-Allocate finds a winner and sends it more traffic. Auto-Target picks, for each visitor, the best of several full experiences you design. AP builds, for each visitor, the best mix of offers per location. This guide is for marketing, ecommerce and analytics teams who use Target, or are about to, and need to pick one and size the traffic.

At WolfSellers we implement Adobe Target alongside Adobe Experience Platform (AEP) for companies across North America and Mexico. If you're new to the product, start with our guide to Adobe Target and A/B testing. Everything we say about each activity comes from Adobe's official Experience League documentation, reviewed in November 2026.

A note on names: the Target pages we reviewed no longer mention "Adobe Sensei". It describes these activities as machine learning and, in the Target welcome kit, as Adobe AI.


Auto-Target, Automated Personalization or Auto-Allocate: a decision table

Adobe Target has six ways to deliver content. What sets them apart is what each one decides: the traffic split, the experience for each visitor, the mix of offers, or nothing at all, because you decide.

Activity What it decides Use it when What it needs License
Manual A/B test Nothing: you set the split per experience You want to measure and rank every experience The sample size from Adobe's calculator Standard
Experience Targeting (XT) Nothing: you set rules per audience You already know which experience fits which audience Well-defined audiences Standard
Auto-Allocate How much traffic each experience gets You want one winner and want to use it while the test runs 1,000 visitors and 50 conversions per experience before it reallocates Standard
Auto-Target Which full experience each visitor sees, on each visit You think different visitors prefer different experiences 1,000 visits and 50 conversions per day per experience Premium
Automated Personalization (AP) Which mix of offers each visitor sees You have several offers per location and a lot of traffic The most traffic of all; check it with the Traffic Estimator Premium
Recommendations Which products or content to suggest You want to suggest items based on past activity Visitor activity history Premium

Auto-Target, AP and Recommendations are part of Target Premium. Premium pages carry that badge on Experience League; the A/B, Experience Targeting and Auto-Allocate pages don't. With a Premium license, you can also use Recommendations as an offer inside an A/B, Auto-Allocate, Auto-Target or XT activity.

There's a seventh activity, the multivariate test (MVT), which combines offers across several page elements. It's fully factorial. Adobe warns that even a high-traffic site may struggle to finish a test with more than 25 combinations.

What is Auto-Allocate, and when should you use it?

Auto-Allocate is an A/B test where Target finds the winner among two or more experiences and sends it more and more traffic while the test keeps running. You set it up in the regular A/B workflow by choosing Auto-Allocate to best experience on the Targeting step. It doesn't personalize. It looks for the best overall result, the same for every visitor.

How the Auto-Allocate algorithm works

  1. Warm-up. Traffic is split evenly until every experience has at least 1,000 visitors and 50 conversions.
  2. Multi-armed bandit. After that, 80% of new visitors are allocated by its logic. The other 20% are spread at random across all experiences to catch changes in behavior.
  3. Hourly rounds. The models update at least once an hour with the cumulative data.
  4. Winner. It declares one when the confidence interval of the best experience no longer overlaps with any other. It uses intervals based on the Bernstein inequality, which correct for repeated checks. So you can look at results while it runs.
  5. With only two experiences. It splits 50/50 until it reaches 75% confidence. Then the winner gets two-thirds. At 95% confidence, the winner gets 90%.

If you wait for the winner badge, Adobe guarantees 95% confidence, which means a 5% false-positive rate.

When Auto-Allocate works and when it doesn't

Good fit Poor fit
Campaigns with an end date, where you can't wait for an A/B test to finish You need to rank every experience: Auto-Allocate separates the best, not the worst
You want more visitors to see the best version as soon as it shows up You don't want to count repeat conversions: it only supports Increment Count and Keep User in Activity
You expect big differences between experiences Experiences depend on timing: "sale ends today" or "Friday deal"

Three things to know before you turn it on:

  • Returning visitors aren't reallocated. They keep seeing their original experience. Only new visitors move toward the winner.
  • Frequent returning visitors inflate the conversion rate of their experience. Adobe suggests counting the goal metric only once per entrant.
  • Once it's live, you can't switch to manual allocation, change the goal metric type or change the advanced settings.

Be careful during peak sales like Black Friday, Hot Sale or El Buen Fin, when urgency messages are everywhere. Adobe explains that timing effects skew Auto-Allocate more than an A/B test. The A/B test looks at a longer period, while Auto-Allocate reacts to the observed rate every hour.

Auto-Allocate works with Analytics for Target (A4T) and with CJA as the reporting source. With A4T, Adobe says to ignore the winner star in the Target report, because it uses a different confidence calculation.

What is Auto-Target, and what does it need to work?

Auto-Target also lives in the A/B workflow, under Auto-Target for personalized experiences. But it doesn't look for a single winner. For each visitor and each visit, it picks the experience most likely to convert, or to drive the most revenue per visit. It bases that on the visitor's profile and on how similar visitors behaved. It requires Target Premium and at least two experiences.

How Auto-Target picks an experience

  • Random Forest. Hundreds of decision trees built on visitor and visit attributes. Target usually builds one model per experience.
  • Random start. At first, even the personalized group sees random experiences until the models are ready. Target validates each model by its AUC. It only personalizes once at least two experiences have valid models.
  • Daily retraining. It rebuilds the models every 24 hours with the latest behavior.
  • Ongoing exploration. It always spends a small share of traffic to keep learning (epsilon-greedy).
  • Fallback. If it can't find winners per visitor, it shows the best overall experience, chosen with Thompson Sampling, while it keeps looking.
  • Consistency. The experience doesn't change within a session. It can change on the next visit, except for the control group.

The control group in Auto-Target

By default, Auto-Target keeps part of the traffic as a control. Visitors in it see a random experience, or one you pick. That group is how you measure the algorithm's lift.

Option Split What it's for
Evaluate Personalization Algorithm (50/50) 50% control, 50% personalized The most accurate lift estimate; Adobe suggests it with a random experience as control
Maximize Personalization Traffic (90/10) Adobe recommends 10% to 30% control Always-on activities; also the suggested split if the control is a specific experience
Custom Allocation Whatever you set The control can't go below 10%

How much traffic does Auto-Target need?

Adobe gives these rules of thumb:

  1. With a conversion metric: 1,000 visits and at least 50 conversions per day per experience. The activity also needs at least 7,000 visits and 350 conversions in total.
  2. With revenue per visit (RPV): the same per day, plus at least 1,000 conversions per experience, because revenue varies more than conversion rate.

These are floors, not guarantees. Adobe notes that in practice each model usually needs more than the minimum conversions before it's valid. For Auto-Target and AP, it recommends 4 to 6 locations with 4 to 6 offers each. And if you change offer content often, the algorithm needs time after each change to use what it learned.

Auto-Target with Adobe Analytics as the reporting source

Auto-Target supports A4T with the Web SDK or at.js. Conversion data from Analytics arrives 6 to 24 hours late. The effect is biggest in the first 48 hours, and behavior lines up with Analytics after five days. That's why Adobe suggests Auto-Allocate for short campaigns that get most of their traffic in the first five days. With A4T, sessions also end after six hours, and later conversions don't count.

Hands pair product cards in different combinations, like the offers Automated Personalization mixes per location.

What is Automated Personalization, and how is it different from Auto-Target?

Automated Personalization starts from individual offers, not experiences. In the Visual Experience Composer (VEC), you create several content options for each location on the page. Target then combines those offers for each visitor. It uses the same algorithm, Random Forest, and the same data as Auto-Target. It requires Target Premium and is always measured against a control group. Adobe recommends putting at least 10% of visitors in it.

Aspect Auto-Target Automated Personalization
What you define Full experiences, each with several changes Offers per location; Target builds the combinations
Editor The full VEC from the A/B workflow, with QA links No custom code editor or multiple experience audiences
Structural page changes A better fit, such as rearranging the home page A weaker fit
Number of experiences Usually fewer Usually more: every combination of offers and locations
Total traffic Often less Often more; it needs the most of any activity
A4T reporting Yes No: it always reports in Target
CJA reporting No No
Switching types No: an Auto-Target activity can't become AP, or the reverse Same: you create a new activity

Both need the same traffic per experience. What changes is how many experiences there are. With two locations and two offers each, AP already has four combinations. In Auto-Target, you can put several changes in one experience and keep it to two.

The Automated Personalization Traffic Estimator

Before you launch AP, open the Traffic Estimator in the VEC. Enter your typical conversion rate, visits per day and test duration, and it tells you if you have enough. Adobe considers an activity well sized when its models are ready within half its duration or 14 days, whichever is shorter. If traffic falls short, it suggests using Auto-Target, cutting combinations or running longer.

Size limits matter too: under 5,000 total experiences for best results, with a hard limit of 30,000.

Even with enough traffic, Target only deploys a model if it beats random selection. If it doesn't, Adobe says your offers are probably too similar. Make them more visually different, or change the message.

What if your traffic is low? What to do instead of forcing machine learning

We see this often on B2B and services sites: good total traffic, but few conversions per page. Forcing AP or Auto-Target there leads to activities that never get to personalize. Target keeps serving random experiences or the overall best one. Use Adobe's thresholds as a filter. If a page can't reach 50 conversions a day per experience, Auto-Target isn't ready for that page yet.

What does work with low traffic:

  1. Do the math before you launch. For A/B tests, use Adobe's sample size calculator. It recommends 95% significance, 80% power and rounding the duration up to whole weeks. For AP, use the Traffic Estimator.
  2. Test where the traffic is. Home, category and product pages with the most visits, not niche pages.
  3. Use fewer, more distinct experiences. Two or three clearly different ones. In Auto-Target, bundle several changes into one experience.
  4. Consider a more frequent metric. Add to cart or form submits happen more often than orders. This is our advice, not an Adobe rule. Decide before launch, because Adobe advises against changing the metric midway.
  5. Use Auto-Allocate when you expect big differences. In Adobe's internal tests, it finds a winner faster than a fixed-horizon test when one experience clearly beats the rest. It takes longer when they're close.
  6. Use Experience Targeting with Adobe Real-Time CDP audiences. If you already know what each segment wants, such as repeat customers or buyers in one industry, a rule is enough. There's no model to train.
  7. Run a manual A/B test with a larger minimum detectable lift. Detecting only big differences needs a much smaller sample.

We cover how to build the hypothesis backlog and the testing program in CRO for ecommerce with Adobe Commerce.

Customer browses on her tablet at home at dusk; device, time and location feed the models.

What data feeds Adobe Target's models

When a visitor enters an Auto-Target or AP activity, Target stores a snapshot of their data as a training record. Some of it Target collects on its own. The rest exists only if you send it.

Data type Examples Does Target collect it on its own?
Environment Operating system, browser, time of day and day of week Yes
Geography Location from IP lookup Yes
Mobile device Brand, model, tablet or phone Yes
Target reporting segments The ones you set up in reporting Yes
Session behavior Pages viewed Yes
Page parameters and Target profile Category viewed, loyalty tier, propensity score No: you send them via parameters, profile scripts or API
Customer attributes CRM data uploaded through the Customer Attributes service No: you upload them
Experience Cloud shared audiences Audiences from Adobe Analytics or Audience Manager Used automatically once shared with Target
Adobe Real-Time CDP audiences Audiences shared through the Target destination Used once you activate them there

If some data shouldn't influence personalization, you can block it with the Models API so no Auto-Target or AP model uses it.

How Real-Time CDP audiences reach Target

  • Who can use it. The Adobe Target destination is available to organizations with Real-Time CDP and, since August 2026, with Adobe Journey Optimizer (AJO). Having the destination doesn't mean you have Target. You need a Target license and, for same-page personalization, a datastream with the Target service.
  • Same-page and next-page personalization. It requires the Web SDK or Edge Network API, a datastream with Target and Experience Platform, edge segmentation and an active-on-edge merge policy.
  • With at.js and no datastream. Next-session personalization only.
  • Which audiences. People audiences only, not B2B account or prospect audiences.
  • Expiration. Edge profiles expire after 14 days of inactivity. The first visit after that may use a historical view of the profile.

We explain how those audiences are built in Adobe Real-Time CDP. If identity doesn't link the web visitor to the store or CRM customer, the model learns from split profiles. We cover that in data governance in Adobe Experience Platform.

How to measure: Analytics for Target (A4T), CJA4T and Target reports

Each activity uses one reporting source, and you can't change it once the activity is live.

Reporting source What it is Supported activities
Target Target's native reports All; AP always reports here
Analytics for Target (A4T) Results are reported in Adobe Analytics, with its metrics and segments A/B, Auto-Allocate, Auto-Target, XT, MVT and Recommendations; not AP
Customer Journey Analytics (CJA4T) Activities are analyzed in CJA with the Experimentation panel Manual A/B, Auto-Allocate, XT, MVT and Recommendations; not Auto-Target or AP

CJA4T is the short name Adobe uses in its Experience League webinars. The documentation calls it Target reporting in Customer Journey Analytics (CJA). It's comparable to what A4T does in Adobe Analytics, but connected to AEP. It has no prebuilt report like A4T. You build it in CJA with the Experimentation panel, and you can pick the success metric after the activity launches.

What CJA4T requires:

  1. Access. CJA and Target for you and your company, the Approver or Editor role in Target, and a role in AEP.
  2. Implementation. The Web SDK is preferred. The alternative is an Adobe Analytics source connector, but Adobe warns that Analytics and CJA results likely won't match.
  3. A CJA connection. Your event dataset plus the Adobe Target Classification Events lookup dataset.
  4. Persistence. On the activity dimensions. Without it, a purchase on a later page isn't credited to the experience the visitor saw.
  5. For Auto-Allocate, the data view and the CJA metric are required.

If you report in Adobe Analytics today and plan to move to CJA, see how to migrate from Adobe Analytics to Customer Journey Analytics. For questions to ask that data, read Customer Journey Analytics analysis examples. For ecommerce measurement, see Adobe Analytics for ecommerce.

How to read the Auto-Target and AP report

  • The top two rows are an A/B test between control and algorithm. That's the lift that matters for your decision.
  • Don't compare experience A with experience B. Compare each experience served by the model with the same one served at random. The ones with the highest lift found a segment that likes them most.
  • Personalization Insights. The Automated Segments and Important Attributes reports show which segments the model built and which attributes mattered most. They appear after 15 days live with traffic, for conversion goals only, in the default environment only, and they use personalized traffic only.

Common mistakes with Auto-Target, AP and Auto-Allocate

  1. Changing the goal metric or reporting source midway. Adobe doesn't guarantee what happens. Create a new activity instead.
  2. Using Reset Report Data. It clears what you see, but not the model's training records. Adobe recommends deactivating the activity and creating a new one.
  3. Picking a metric without the volume it needs. RPV in Auto-Target needs at least 1,000 conversions per experience. With A4T, calculated metrics can't be the primary goal; use a custom event.
  4. Shrinking the control group to "not lose sales". Without a large enough control, you can't tell if the model helps.
  5. Offers that are nearly identical. The model doesn't beat random, so Target doesn't deploy it.
  6. Picking the Auto-Allocate winner by eye. Wait for the badge. Before that, it's easy to pick the wrong experience.
  7. Sending significant traffic to non-default environments. Auto-Allocate models only the default environment, and so does Auto-Target's overall comparison model.

E-commerce manager and analyst review the printed week-by-week plan for the 90-day pilot.

A 90-day Adobe Target pilot plan

The timing is our working proposal, not an Adobe rule. Adjust it to your traffic.

  1. Weeks 1 and 2: assessment. Compare daily visits and conversions on your pages with each activity's thresholds. Set the goal metric and the reporting source (Target, A4T or CJA). Check whether you have the Web SDK. You need it for CJA4T and for same-page personalization with Real-Time CDP.
  2. Weeks 3 to 6: first test. Launch an A/B test or Auto-Allocate on your highest-traffic page, with two or three very different experiences. The goal is to learn which messages move your visitors.
  3. Weeks 4 to 6: first data-driven rule. Activate the Target destination in Real-Time CDP and start an Experience Targeting activity for one clear segment, such as repeat customers.
  4. Weeks 7 to 12: Auto-Target. If the page clears the thresholds, build an Auto-Target activity with the experiences that worked, at 50/50 to evaluate the algorithm.
  5. Day 15 of Auto-Target: insights. Review Automated Segments and Important Attributes to see who each experience works for and to feed new hypotheses.
  6. Weeks 12 and 13: decision. If the algorithm beats the control, move it to 90/10 as an always-on activity. If not, go back to Experience Targeting and A/B tests with what you learned. Consider AP only if the Traffic Estimator gives you the green light.

Frequently asked questions

What's the difference between Auto-Target and Automated Personalization?

Both use the same algorithm (Random Forest) and the same data. In Auto-Target, you design full experiences and Target picks one per visitor. In AP, you define offers per location and Target builds the mix. That's why AP usually needs more total traffic, and why Auto-Target is better for structural page changes.

Does Auto-Allocate personalize the experience?

No. Auto-Allocate is an A/B test that finds an overall winner and sends it more traffic, the same for everyone. If you want each visitor to see their best experience, that's Auto-Target or AP.

How much traffic does Auto-Target need?

With a conversion metric, Adobe's rule of thumb is 1,000 visits and 50 conversions per day per experience, plus 7,000 visits and 350 conversions across the activity. With revenue per visit, add 1,000 conversions per experience.

What is CJA4T, and which activities does it support?

It means using Customer Journey Analytics as the reporting source for Adobe Target. It supports manual A/B, Auto-Allocate, Experience Targeting, MVT and Recommendations. Auto-Target and Automated Personalization don't support it yet.

Do I need Target Premium for these activities?

For Auto-Target, Automated Personalization and Recommendations, yes. Auto-Allocate, A/B tests and Experience Targeting don't carry the Premium badge in Adobe's documentation.

If this topic is relevant to your business, these services from WolfSellers can help you implement it:

Want to dive deeper?

Let's talk.

We're an Adobe Gold Partner in Mexico with experience in Adobe implementations and integrations. If anything in this article applies to your operation, the first consultation is on us.

Or email us at contacto@wolfsellers.com

Keep reading

Chat with us on WhatsApp