Decision, Random, Engagement, Path: SFMC Journey Builder Splits

If you've shipped more than a couple of Journeys, you've seen this: someone drags a Decision Split onto the canvas because they want to "send different emails to engaged users," the flow runs, and the numbers come back weirdly wrong. The fix is almost always that they picked the wrong split type.

Journey Builder has five splits, each answering a different question. Here's the field guide we hand to every new team member.

Decision Split — "What does the subscriber's profile say?"

Decision Split branches based on subscriber data at the moment they reach the block.

Typical use: MemberTier = "Gold" → branch A (priority content), else → branch B.

The catch: the data the Decision Split checks has to be linked to the Contact record through an Attribute Group in Contact Builder. If the value is sitting in a Data Extension that isn't linked to Contact, the split will never see it. First-time users trip over this and spend an afternoon wondering why the "Gold" branch is always empty.

Use when: you want to branch based on profile data, preferences, or a pre-computed segment.

Engagement Split — "Did they interact with the last email?"

Engagement Split branches based on whether the subscriber opened or clicked the email you sent earlier in the same Journey.

Important limit: Engagement Split only sees engagement with emails inside the same Journey. It can't check whether the subscriber opened a campaign from another Journey or from Automation Studio. If you need cross-channel engagement checks, that's a Decision Split on a pre-computed attribute instead.

Use when: you want to follow up based on engagement with an email you just sent in this Journey.

Random Split — "Chop the audience into random groups"

Random Split assigns subscribers to branches by percentage, completely randomly. No data, no behavior — just a dice roll.

Use when: you're A/B-testing content or flow structure. Classic case: NTO wants to know whether a one-email welcome converts better than a three-email welcome. Random Split 50/50, each branch ships the variant, measure conversion on Goal completion.

Path Optimizer — "Random Split that learns"

Path Optimizer is Random Split's smarter cousin. It runs the test for a defined period, then automatically sends the rest of the audience into the winning branch based on a metric you chose upfront.

Configuration you pick before it runs:

• Winning metric: Open Rate, Click Rate, or Web Conversion
• Test duration: how many days before the winner is chosen
• Holdback: optional % kept as a control group

Realistic setup: test two subject lines on a welcome email. Path Optimizer runs 7 days with 20% of audience on each variant, then funnels the remaining 60% into whichever subject line had the higher open rate. No one has to watch the dashboard and flip the switch manually.

Einstein Scoring Split — "Ask the model"

Scoring Split branches based on Einstein Engagement Score — a per-contact score Salesforce's model produces predicting the likelihood they engage with future email.

Use when: you want different content for highly-engaged vs. unengaged contacts, but you don't want to hand-define the thresholds. Einstein does the segmentation for you.

Requires Einstein for Marketing Cloud (not bundled with every edition — check first).

Picking the right one — the short version

Question you're answeringSplit to useWhat does the profile say?Decision SplitDid they engage with the last email here?Engagement SplitI want to test, no data ruleRandom SplitI want to test and auto-scale the winnerPath OptimizerI want to segment by model-predicted engagementEinstein Scoring

The mistake most teams make is using Decision Split where Engagement Split belongs. They build a branch on "has the subscriber ever opened an email" — that has to live on the Contact as a boolean attribute, updated by an Automation upstream. Without that setup, Decision Split sees an empty field and routes everyone into the default branch. Engagement Split would have answered the question in one block.

Takeaway

Every Journey we ship in production touches at least one of these splits. Picking the wrong one is a silent bug: the Journey runs, the emails go out, but the reporting lies. When you inherit a Journey with odd reporting, start by listing every split and asking what question it's answering. Seven times out of ten, one of them is lying to itself.

---

Auditing an inherited SFMC setup? Our Salesforce team ships production-grade Marketing Cloud engagements - from Journey design to deliverability tuning. Get in touch ->

See our full platform services for the stack we cover.