Journey Optimizer: profile and context features for AI models

Adobe's November 2026 Journey Optimizer notes add three read-only orchestrated campaign skills to Coworker, profile and context features for Personalized optimization models, and token substitution for Experience Manager fragments. The two bigger items are Limited Availability, so most teams will need to ask their Adobe representative before they can touch them. For most programs the practical work is deciding which of these deserve a place on this quarter's roadmap and which can wait.

The October 2026 notes sit right behind them and contain changes that many teams can use sooner, including sender identity overrides and self-serve high throughput setup. This post covers both sets, in the order a busy team should think about them.

Key takeaways

• Personalized optimization models can now draw on up to 50 profile attributes and up to 20 context attributes, but access is Limited Availability.
• Context features apply only to decisioning policies used in inbound channels.
• The new Coworker skills for orchestrated campaigns can list, retrieve and render campaigns. They cannot create, edit or publish.
• Experience Manager fragment token substitution covers fragments added directly to a message, and excludes fragments served through Experience Decisioning.
• October's sender overrides and self-serve high throughput setup need no special access request according to the notes, though the high throughput one needs the add-on.
• Treat every model or content change as a test with a baseline, not a switch you flip for everyone.

What changed in the November 2026 Journey Optimizer release

Three items landed in the November 2026 section, each with its own availability date.

Adobe also notes that its continuous delivery model means rollout is phased. A feature with a date on it is not necessarily visible in your sandbox that same day. Check your own environment before you plan a sprint around anything here.

What the Coworker orchestrated campaign skills give you

Coworker is the chat interface, and the new skills let you ask it about orchestrated campaigns instead of clicking through the user interface or calling an API. There are three. Orchestrated Campaign List lets you list, filter and count campaigns, for example by status or channel. Orchestrated Campaign Get returns a specific campaign's details, lifecycle history and execution history, and can visualize its activity graph. Orchestrated Campaign Workflow Renderer draws a workflow graph as a diagram.

All three are read-only. Adobe says plainly that creating, editing or publishing campaigns through Coworker is not yet supported.

That limit shapes who benefits. A marketing operations lead who gets asked every Monday how many campaigns are live on a given channel can now answer in one question. A support analyst chasing why a campaign behaved oddly can pull its lifecycle and execution history without needing access to the build screens or a developer to write a query. An approver who needs to understand a workflow before signing off can ask for the rendered diagram.

The flip side is that nothing here changes how campaigns get built. If your bottleneck is authoring speed, this does not help yet. If your bottleneck is visibility, it does.

A sensible first step is small and low-risk:

1. Pick three questions your team asks repeatedly about campaign state.
2. Try each in Coworker and compare the answer with what you see in the interface.
3. Note where the answers disagree or feel incomplete, and raise it through your Adobe contact.
4. Only then tell the wider team it is a reliable shortcut.

Permissions deserve a look too. Conversational access is still access to campaign data. Confirm who can use Coworker and what they can see, using Adobe's own documentation, before you announce it broadly.

Profile and context features for Personalized optimization models

This is the most consequential line in the notes. Personalized optimization models can now be enriched with profile and context features, in addition to or instead of audiences. When you create an AI model, you pick up to 50 profile attributes from your profile schema, such as loyalty tier or churn propensity score, and up to 20 context attributes sent with the decision request, such as device type, language or page URL. The model uses those signals to score decision items in real time and aim for more relevant offers.

Two conditions matter. First, it is Limited Availability, so you contact your Adobe representative to get in. Second, context features are only available for decisioning policies used in inbound channels. A team that decides offers only for outbound email gets the profile side but not the context side.

Why does this matter? Audiences are a blunt instrument. Putting someone in a high-value segment or not is a yes or no. A loyalty tier or a churn propensity score carries more information than a segment flag, and a model can weigh it against the offer in front of it. Context goes further. A person on a phone reading in one language at a particular page is in a different situation from the same person on a desktop, and audiences rarely capture that. The decision request is the only place that information exists at the moment of choice.

That also hints at the real work. The model can only use what you send and what exists in the profile. Teams often discover that the attribute they most want, say a propensity score, is stale, partly populated or computed in a system that only syncs weekly. A model fed unreliable features will score confidently and wrongly.

Before you request access, do this homework:

• List the profile attributes you would pick, and check how complete and how fresh each one is.
• List the context attributes your front end could send with the decision request, and check whether engineering can add them without a release freeze.
• Decide which inbound placements will use the model. Context features do nothing for you elsewhere.
• Agree on a success metric and a holdout before the model goes near live traffic.

One trade-off is worth stating. More features is not automatically better. With a ceiling of 50 profile and 20 context attributes, the temptation is to fill the slots. Resist it. Start with a short list you can explain to a colleague, and add only when a test shows lift.

Dynamic token substitution for Experience Manager fragments

The third November item is aimed at content teams. You can now use attribute paths such as {{profile.person.name.firstName}} directly in Experience Manager Content Fragments, with no token map needed. It lists an availability date of 5 November 2026, and it is Limited Availability, so access again goes through your Adobe representative.

The boundary is specific. It works only for fragments added directly to a message. It does not work for fragments served through Experience Decisioning.

In practice, this removes a layer of plumbing. Previously, as the notes imply by saying no token map is needed, personalizing a fragment meant maintaining a mapping between tokens in the content and profile fields. Authors had to know the mapping existed, and someone had to keep it current. Now an author writes the attribute path where the first name should go, and it resolves.

The risk is that the same fragment behaves differently depending on how it is used. A fragment dropped straight into an email will resolve the path. The same fragment pulled in through Experience Decisioning will not get this treatment. If your content library reuses fragments across both patterns, an author could preview one and be surprised by the other.

A few guardrails help:

• Label fragments by how they are meant to be delivered, directly or through decisioning.
• Test with a profile that has the field empty, since a missing first name is the common failure.
• Keep a fallback in the copy so the sentence still reads properly without the value.
• Include one rendered preview of each delivery pattern in your content QA checklist.

The October 2026 changes that may be easier wins

The October section is worth a second look, because several of these do not require a Limited Availability request.

Email header fields and delivery settings can now be overridden at the journey or campaign action level, without a separate channel configuration. From name, From email prefix, Reply-To, recipient address and list-unsubscribe values can each be set using profile attributes or contextual data. The availability date is 26 October 2026. If one channel configuration serves several regions, brands or teams that each want their own From address or reply routing, you no longer need a configuration per variant. The caution is governance. Letting every journey set its own sender identity means someone has to own the rules for what is allowed, or your sender reputation story gets messy.

Also on 26 October 2026, teams that bought the High throughput transactional messaging add-on can set up High throughput for API-triggered email and push campaigns themselves in the product UI, through a dedicated channel configuration. Adobe says you no longer need to request a manual setup for each resource. If you waited on those requests before, this is time back.

October also lists a set of Coworker skills: Channel Builder, Business Performance Analysis, Journey Hygiene Analysis, Trace Journey Profile, Journey Execution Timeline, and a decisioning rule generation skill that produces eligibility rules and ranking formulas from natural language. The notes describe each in its own section, and the page's custom outbound channel entry is cut off in the text available, so check Adobe's page for that one directly. The generated rules and formulas are worth reviewing by a person before use, the same as anything written by hand.

A rollout plan with owners

Most teams do not need a project for this release. They need a short list with names next to it.

Sequence matters. Do the audits first because they cost nothing and they decide whether the access request is worth making.

How to measure whether it worked

For the model change, compare against what you run today. Keep a holdout that continues on the existing audience-based model, and judge on the outcome you already report: offer click-through, conversion, revenue per decision, whichever your program lives by. Let it run long enough to cover your normal weekly pattern. A model that looks better on a Tuesday afternoon has told you very little.

Watch for segments that behave differently. An enriched model may help returning customers and do nothing for new visitors with thin profiles. That is useful to know, and it points to where better data matters.

For fragments, measure errors rather than lift. Count messages where a name or other value rendered blank or showed a raw path, using test sends before launch.

For Coworker, the measure is time. How long did a routine status question take before, and how long now? If nobody saves time, drop it.

Where a partner helps and where you can go solo

You can do a lot of this yourself. Checking what is visible in your sandbox, requesting access, auditing attribute freshness and trying the Coworker skills are ordinary admin and analyst tasks. So is adding a fallback to a fragment.

A partner earns their fee on the parts that cross teams. Designing a model test with a defensible holdout, deciding which profile attributes are trustworthy enough to feed it, and getting context values from a front end into the decision request all involve data, engineering and decisioning at once. If your in-house team is stretched across those, outside help makes sense. The Adobe Journey Optimizer partners directory lists firms that work on it, and the shortlist of top Journey Optimizer partners is a reasonable starting point for comparison.

If you do go that route, write down the scope before the first call. The Journey Optimizer RFP template gives you a structure for stating goals, constraints and what success looks like, which keeps proposals comparable.

What to ignore, and what not to do

Ignore the urge to rebuild everything because a feature exists. If your current audience-based decisioning performs well and you have no inbound placements, context features offer little. If your fragments never use personalization, token substitution is not for you.

Don't announce Coworker as a way to manage campaigns. It reads them. It does not build or publish.

Don't assume Limited Availability means soon for everyone. Adobe says access is for a set of organizations, and you get in by contacting your representative. Plan as though you might not have it this quarter.

Don't push a changed model to full traffic with no baseline. And don't let sender overrides spread without someone owning the rules.

Finally, read Adobe's detailed documentation for each item before you build, since limits and behavior can change as features move out of limited release. If you want to compare how other teams approach this work, the MarTech Partners blog tracks Adobe changes as they land.

Sources

• Adobe Experience League: Adobe Journey Optimizer release notes — November 2026 and October 2026 updates, page last updated 10 November 2026