November 11, 2026 · 6 min read
Try it on your own accounts — 2,500 free credits a month, no credit card. By default every change waits for your approval.Start free
Playbooks

Adobe Target A/B tests with an AI agent: weekly readouts, clean-ups and test prep

TL;DR

Adobe Target teams spend their week on three chores an AI agent can take over: reading every live test's lift, cleaning up activities and offers nobody uses, and preparing the next test. This playbook shows each with the goals to type, what the agent reads and what stays a person's call.

The weekly readout

Every A/B and Experience Targeting activity has a performance report by experience: visitors, visits, conversions and lift. Reading them one by one in the Target UI is where most of a Monday goes. An agent connected to your tenant lists the live activities, pulls each report and writes one readout.

A useful readout says what is winning AND what is too small to call. The agent reports the visitor and conversion counts beside the lift, so "+18% on 40 conversions" is not mistaken for a result.

Goal you typeWhat the agent does
"Readout of every live Target test: winner, lift, and whether it is significant enough to call."Lists activities in state approved, reads each performance report, ranks experiences and flags low counts.
"Which tests have run more than six weeks without a clear winner?"Compares start dates and lift, and lists the stalled ones with their numbers.

Cleaning up

  • Ended but still live: activities past their end date that are still approved — the agent proposes deactivating them (an ordinary change, by default waiting for your approval).
  • Orphaned offers: offers no activity references — listed for you to decide; deleting one always waits for a person.
  • Audiences with no activity: listed, never removed on their own.

Preparing the next test

  1. 1Ask for the test in plain words: "A/B test the spring banner on the homepage hero: current vs a new JSON offer, for returning visitors."
  2. 2The agent proposes a JSON offer, an audience and a new activity created saved — nothing a visitor sees changes.
  3. 3You review the approval card. An HTML offer, being code on your site, is held for a person every time.
  4. 4When you are ready, make the activity live — the agent can ask, but that step always waits for a person.

What we measured on a real Target tenant

  • Target's resources each want their own media type (offers v2, activities and audiences v3, the rest v1); a wrong one answers 415, so the tools set it per path.
  • An HTML offer was created, changed, read back and deleted; an A/B activity created saved, deactivated and deleted — never made live.

Frequently asked questions

Can an AI agent start an Adobe Target test on its own?
It can prepare everything up to a saved activity; making the activity live waits for a person every time.
Does the agent decide which experience won?
It reports lift with visitors and conversions and says when the numbers are too small; the call stays yours.
Run your ad ops with an agent you can trust

Start free — 2,500 credits a month, no credit card. By default every write waits for your approval.

Keep reading