[Workflow] Action: Split Test | Voters

[Workflow] Action: Split Test

in progress

Suresh Babu Sakthi

Even Split

splits between Path A and B till goal condition is met, then send to the winner path

Conditional Split

send to path A until the goal condition is met, then send to path B.

August 4, 2022

Ashwin Raghunandan

Hey everyone, Ashwin here (PM for Workflows).

We've picked this up along with the related A/B testing posts on the ideas board. Sharing what we're planning below -

Even Split

Contacts get distributed equally across the branches you create. Add three branches and roughly a third of contacts go down each one. This is for when you want to distribute contacts evenly, not test anything.

Weighted Split

Same as above, but you control the distribution. Set 70/30, or 50/30/20, and contacts get routed as per the percentages you define. Useful for phased rollouts, distributing load across teams, or gradually shifting volume to one path.

A/B Test

This is the one most of you have been asking for on this post. The behavior we're looking at:

Build your variants on each path with different steps or messages in each
Define a winning criteria - the event that decides the winner (eg. email event, trigger link clicked, appointment booked, set custom criteria based on field updates and more)
Define the test size - how many contacts go through the test phase before a winner is picked, can also be configured based on an end date defined
Once the test phase completes, every contact entering after that is automatically routed down the winning path
Stats per variant so you can see how each one actually performed

What I'd love your input on:

Today the 'Split' action is positioned as a way to distribute contacts across branches (random split). Even split and weighted split sit naturally under it. But A/B testing does something quite different - it validates which path wins based on a goal, rather than just distributing contacts.

So, would you prefer 'A/B Test' as its own separate action in the action list, or as an option inside the existing Split action? On naming, we're leaning towards 'A/B Test' for the new action over 'Split Test', since that's the term most people search for. If that isn't what you'd scan the action list for, or you have a name that reads more clearly, drop it in the comments.

And apart from the above, anything else you'd want to see here? Specific goal events you'd want to test against, reporting you'd need, or scenarios the above wouldn't cover - we're all ears :)

Thanks!

Johannes Meyer-Ahrens

Ashwin Raghunandan yes please an extra A-B test (in addition to the split, we would like a better distribution of contacts during the split.)

David Hall

Ashwin Raghunandan To directly answer your question, I would put the A/B Split within the Split action similar to how it is with the wait action. The last thing anyone needs is another ungrouped action to have to scroll through. As well, separately, it could also be confusing about its ultimate intention.

For the part about define a winning criteria - I would think email events like link clicked, trigger link clicked, appointment booked, contact replied, and some custom options would be very helpful.

James Campbell

Ashwin Raghunandan I would like it configurable based on all of these, with all of them optional.
Take the multivariate split. I may care about 2 different KPIs, a trigger link click AND a final purchase. I've had plenty of tests where the final goal KPI was highest in a test in which the intermediate steps were the worst. But, the final KPI has such a long sales cycle that it takes an eternity to get there (and there's a high risk of audience contamination between test cohorts).
So being able to split, defining a metric, but not automatically have traffic routed on a winner is useful. But also, setting it to route traffic on those winning conditions is also occasionally useful.
If you can give us:
Split type (multi, A/B)
Split percentage (equal, defined)
Metric to Test (single, multiple if you can make it viable with primary/secondary metric selection)
Route Future Traffic on Winner: bool
Additional bonus: define target p-value and confidence interval (probably best to include a projections of how many tests/actions will be needed to hit that goal in case someone with 100 emails/month creates a test that will take 12 centuries to complete).

Daniel Marr

Ashwin Raghunandan Definitely a separate action. We use the split extensively for non AB testing workflows in our snapshots and merging would cause unnecessary confusion.

Re: Goal event id like to see included is contact replied, by tag and custom field value.

David Hall

The current Split action is built for experimentation. We need it to support operational distribution.
Random assignment works when the goal is to test different funnel pages, messages, or offers across a large sample. It doesn't work when each path represents a person, department, appointment calendar, or workload.
If four paths are set to 25%, users reasonably expect contacts to be distributed evenly. In testing, 12 contacts were distributed unevenly.
That may be statistically valid for random assignment, but it is operationally unusable. If those paths represent four salespeople, one salesperson receives six leads while another receives one. The percentages appear equal, but the actual workload is not.
This creates several problems:
Uneven staff workloads, Slower response times, Lost sales opportunities, Internal disputes over lead distribution, Unreliable capacity planning, Manual reassignment, Reduced trust in workflow automation
The current behavior is also unclear. When a contact is removed and re-added, the contact always follows the same path. That suggests the system may use a deterministic value, such as the contact ID, to select the path. If so, the behavior is technically consistent, but it is not truly random from the user’s perspective.

David Hall

This creates two separate decisions that should not be combined:

How should contacts be distributed?
What should happen when a contact re-enters?

Those should be separate settings.

Recommended Split Methods

Random Split

Assign each contact randomly based on the percentages entered.

Best for: A/B testing, Funnel testing, Email variation testing, Offer testing, Testing different follow-up sequences

Random means the percentages should become accurate across a sufficiently large sample. It does not guarantee balance within a small group.

Even Split

Distribute contacts sequentially using round-robin logic. In the above example, after 12 contacts, each path receives exactly three.

Best for: Sales lead assignment, Customer service routing, Account manager assignment, Appointment distribution, Task distribution, Onboarding assignments, Review or approval workloads

Weighted Even Split

Use round-robin distribution while respecting different percentages.

For example:

-Senior salesperson: 50%

-Salesperson B: 25%

-Salesperson C: 25%

The distribution could follow: A, B, A, C, A, B, A, C

This allows businesses to distribute work based on staff capacity while still maintaining predictable ratios.

***Re-Entry Should Be a Separate Setting

Each split should include a re-entry behavior:

Keep Previous Path - The contact returns to the same path previously assigned.

Best for: Consistent account ownership, Continuing an experiment, Maintaining the same service representative, Preserving customer experience, Assign Again

The contact receives a new assignment based on the selected distribution method.

Best for: Recurring task distribution, New sales opportunities from an existing contact, Repeated appointment requests, New support cases, Workload balancing over time

The user should decide whether assignment is tied to the contact or to each workflow entry.

David Hall

Advanced Use Cases

Sales Lead Distribution - A company has four salespeople. Every incoming lead enters a workflow and is assigned evenly. With Random Split, one salesperson could receive six of the first twelve leads.

With Even Split, each salesperson receives three. This affects response time, commissions, workload, and trust in the system.

Opportunity Distribution - An existing contact may create multiple opportunities over time. If the split is permanently tied to the contact, every new opportunity returns to the same salesperson. That may be appropriate for account ownership, but it does not work when the business wants to distribute each new opportunity separately.
The split may need to operate by: Contact, Opportunity, Workflow entry, or Appointment Routing - A business has three service providers with equal availability. An Even Split can rotate appointments across their calendars. A Weighted Even Split can send more appointments to someone with greater capacity.
Customer Support Routing - New support requests can rotate between available support representatives. This prevents one person from receiving a disproportionate number of cases because of random variation.
Client Onboarding - New clients can be distributed between onboarding specialists. Balanced assignment helps maintain: Consistent response times, Predictable workloads, Comparable service capacity, Clear accountability, or Quality-Control Reviews. Completed work can rotate between managers or reviewers. This prevents one manager from becoming the default bottleneck.
Vendor or Location Distribution - A company may distribute requests between: Multiple branches, Service locations, Contractors, Fulfillment teams, Installation crews, or Partner companies. The distribution may need to be equal or weighted based on capacity.
Controlled Workflow Testing - Even Split can also improve certain tests. If a user is testing four workflow paths with only 40 contacts, random assignment may produce groups of very different sizes. Even Split guarantees 10 contacts per path, giving the user cleaner comparison groups.

Random assignment should remain available for statistically valid experimentation. Even assignment should be available when equal sample size matters more than randomness.

David Hall

Recommended Improvements
The Split action could offer:
Distribution method
-Random
-Even
-Weighted Even
Re-entry behavior
-Keep the contact on its previously assigned path
-Assign a path each time the contact enters
-Assign once per opportunity or record, where supported
Paths
-Path name
-Percentage or weight
-Assigned destination
-Current distribution count
-Minimum Acceptance Criteria
For an Even Split:
Four equal paths receiving 12 entries must receive three entries each.
The difference between the highest and lowest path counts should never exceed one for equally weighted paths.
Weighted paths must maintain the configured ratio across each completed distribution cycle.
Re-entry behavior must follow the user’s selected setting.
The system must handle simultaneous entries without assigning several contacts to the same next path.
Editing path percentages should begin a new distribution cycle or clearly explain how existing counts affect future assignments.
Workflow logs should show the split method and why a path was selected.

Divyam Bhadoria

Merged in a post:

A/B testing workflows

Yon Elosegui

It would be great to make a/b testing withing workflows.
For example: You create a workflow with all its steps (emails, sms...) that triggers when someone submits a form. But you want to test the same workflow with the same trigger with different steps or messages. And also the end having the possibility to check the stats to see wich workflow worked best, for example: if the conversion objective of the workflow was a lead to book an appointment the ability to check that would be great.
You could also check which workflow worked best in the pipeline by using some filters but the stats could also work great

January 25, 2026

Brahim Scaling Adventures

This might be a little out of context.

But a more useful implementation would be split testing for funnel pages. Currently, there is that functionality, but it's not very useful as it only shows conversions if there's an opt-in form. It would be more useful if we could actually run tests by defining a goal (a funnel page). Once a variation reaches that goal, it gets +1 point for the variation, and we have a percentage of success to determine which is the winner. Currently, we don't have this capability.

Prasoon Dadhich

We have Split action with Random Split, the same can be used for Even Split. Conditional Split was de-prioritized as there was less usage for the Split action. Can you please share some examples which cant be achieved today and we will relook the new features for this action.

Billy Wood

This is such an important feature we need

Kennedy Email

This would be a stand out feature as i dont know another platform that offers this.

Steve Day

hi GHL! any news on the ETA for this - we can't move our evergreen event registration to GHL until we can split test in automations.
Marketing / Sales is all about split testing to incrementally improve - without it, we're just guessing! Please prioritise this essential feature.
thanks so much for all your amazing work!

Billy Wood

Steve Day Yes the GHL team need to look into this. The AI features sound great for newbies. But there's serious businesses using GHL that need basic functionality like this, rather than AI stuff we don't use.

Alex Varnals

Any updates on when A/B split tests will be available in Workflows?

abdo hainna

The ability to make changes inside split test after a contact go throw the automation

→