If you build funnels in GoHighLevel, you’ve probably used the built-in split test on a step. It does what it says: two versions, a traffic percentage, a conversion number for each. For a quick headline test, fine.
The trouble starts when you ask what “conversion” means. In GoHighLevel, it means a form submit. For an agency, a coach, a clinic or a home-services company, a form submit is a lead, and a lead is worth nothing until it becomes an appointment, a show and a closed deal. The page that wins on form submits is not necessarily the page that wins on revenue. Quite often it’s the opposite, because the page that promised the most pulled the most submits and the worst-qualified leads.
This guide covers how to split test in GoHighLevel so the lead is followed through the pipeline and the variant that gets the credit is the one that produced closed business.
Key takeaways
GoHighLevel's native split test is one control against one variant, scored on form submits, with no statistical readout and no pipeline revenue.
For service businesses, the real conversion is a won opportunity. Measure that, and use form submits as a fast guardrail.
Tie the visitor id captured at form submit to the GHL contact, and the won deal weeks later is credited to the landing page variant that created the lead.
Agencies: one Splitlab workspace per client sub-account keeps every client's data separate.
What the native GHL split test gives you
At the time of writing, the funnel split test in GoHighLevel lets you add one variation to a step, choose how traffic divides between original and variation, and compare conversion rates. It's quick, it needs no code, and it lives inside the builder you already use.
What it doesn't do, if you sell a service:
More than one variant. Testing three offers means three tests in a row.
Anything after the form. Conversion is the submit. Appointments, shows and deals are invisible.
Statistics. Two percentages, no probability, no interval, no signal for when to stop.
Pipeline revenue. Opportunity value, won stage, deal size: none of it flows back to the page variant.
The lead-quality trap
Here's the pattern that costs agencies money, over and over.
Variant B of a landing page offers a "Free $500 audit" instead of "Book a strategy call". Form submits jump 40%. The native test declares B the winner. The client is thrilled.
Sixty days later the pipeline tells a different story. Variant B leads were tyre-kickers. Show rate dropped, close rate dropped, and the average deal came in smaller because the audit anchored the relationship on “free”. The original page produced fewer leads and more revenue.
Nobody caught it because the test ended at the form.

The fix isn't to stop testing. It's to measure the thing the client actually pays for (revenue per visitor, not conversion rate), which is won opportunities and their value, and to credit that value back to the page variant that produced the lead.
How to set up revenue-attributed split tests on GoHighLevel
1. Install the snippet on the funnel
Paste the Splitlab snippet into the tracking code section of your GoHighLevel funnel or website so it loads on every step. One paste per funnel. The snippet assigns each visitor to a variant, renders it without a flash of the original, and keeps the assignment as the visitor moves between steps.
Then add a hidden field to your GoHighLevel form and set its Query Key to mvt_visitor_id, exactly. The snippet fills it in, so every new contact carries the id of the visitor who submitted the form.
2. Connect GoHighLevel
In Splitlab, open Integrations, choose GoHighLevel and connect your sub-account through the HighLevel marketplace app. Splitlab only asks for read access.
Splitlab then gives you a webhook URL. In GoHighLevel, add workflows that send Contact Created, Invoice Paid and Opportunity Won to it. Before counting a sale, Splitlab reads the invoice or opportunity back from GoHighLevel to confirm it's really paid or won, and uses the amount GoHighLevel reports. Details on the GoHighLevel integration page.
3. Decide what "conversion" means for this test
For a service business you usually want a ladder, not a single event.
Event | Role in the test |
|---|---|
Contact created (form submit) | Guardrail (catches broken forms, fast feedback) |
Call booked (Calendly or Cal.com) | Secondary metric |
Invoice paid | Primary metric, revenue |
Opportunity won, with value | Primary metric, revenue |
There’s no deadline to configure. The credit follows the contact, so a deal won six weeks after the form submit still lands on the variant that produced the lead.
4. Build the variants
Point a test at the landing page URL and build variants in the visual editor. Test the offer, the headline, the proof, the form length, the CTA. You can run more than two, and you can combine changes across elements in a multivariate test if traffic allows.
5. Launch and read the result in revenue
The report shows revenue per visitor and probability of being best for each variant, with form-submit rate sitting there as a guardrail. The scenario above now reads: “Variant B: +40% form submits, 71% probability of lower revenue per visitor.” That’s a very different conversation with the client.
How the lead gets tied back to the variant
This is where most attempts at external testing on GoHighLevel fall apart, so it's worth understanding even though the tool does it for you.

The snippet assigns the visitor to a variant and stores a visitor id in the browser.
When the visitor submits the form, the hidden
mvt_visitor_idfield saves that id on the GoHighLevel contact, linking the contact to the Splitlab visitor.Weeks later the opportunity moves to Won. Your workflow sends the event to Splitlab, which confirms the opportunity with GoHighLevel, looks up the contact, finds the visitor and the variant, and credits the deal value to it.
No pixels on thank-you pages, no UTM spreadsheets, no guessing which page a closed client saw six weeks ago.
Agency setup: one workspace per client
If you run GoHighLevel sub-accounts for several clients, mirror that in Splitlab with one workspace per client. Each workspace has its own domains, tests and currency, and connects to its own sub-account with its own webhook URL.
Team members can be invited to one client's workspace or across all of them, which keeps a client's data out of another client's view.
Mistakes that manufacture fake winners
Non-sticky assignment. If a visitor can see the variant on one visit and the control on the next, your data is scrambled. Use cookie-based assignment on the funnel's domain and don't clear it between steps.
Counting conversion in two places. A form-submit event on one variant and a thank-you-page event on the other won't match, and the gap will look like a lift. One source of truth per event, ideally the GoHighLevel contact and opportunity records themselves.
Running the native split test and an external test on the same step. Two tools assigning independently gives you four experiences and no interpretation.
Testing through a promotion and reading the result after it ends. Traffic mix changes. A bandit that adapts continuously handles this better than a fixed test that averages the promo and the lull together.
Stopping when the number looks good. With two percentages and no probability, you’ll stop early and ship noise. Decide the rule up front: a probability threshold and a maximum acceptable expected loss. Splitlab reports both for every variant, so the rule is easy to apply.
Which allocation method to use
For a landing page that runs all year on paid traffic, multi-armed bandit allocation makes sense: it moves traffic toward the variant producing more revenue while the test is still live, which is money you'd otherwise spend on the loser. For a test you need to learn from and present to a client, a fixed even split gives the cleaner story.
One caution: with a long sales cycle, the bandit learns from a lagging signal. Point it at booked calls while the test runs, and let the final report use won revenue once deals close.
Frequently asked questions
Can I split test GoHighLevel pages without touching code?
Yes. The native split test needs no code. Splitlab needs one snippet pasted into the funnel's tracking code once, and variants are built visually after that.
How do I attribute a closed deal in GoHighLevel back to a landing page variant?
The visitor id assigned by the testing snippet is captured with the form submission and linked to the GoHighLevel contact. When that contact's opportunity is marked Won, the deal value is credited to the variant the contact originally saw.
Does this work across multiple sub-accounts?
Yes. Connect each sub-account to its own Splitlab workspace. Each workspace gets its own connection and its own webhook URL, so one client's deals never land in another client's reports.
What if my sales cycle is 60 days?
Nothing to change. The credit follows the contact, so a deal won 60 days later is still credited to the right variant. While the test runs, watch an earlier event such as booked calls, and read the final result on won revenue once deals close.
Can I test the booking calendar step?
Yes. Any URL the snippet loads on can be tested. For the calendar step, booked calls is the natural primary metric.
My final thoughts
GoHighLevel’s built-in split test is fine for a quick headline check. Where it falls short is the part that matters most to an agency: which version brought in leads that actually closed.
If you run funnels for clients, that’s the number your client cares about. A variant that doubles form fills but sends the sales team junk leads is a loss, however good the report looks. So tie each lead back to the version they saw, judge on won deals and revenue, and keep one workspace per client so the numbers never mix.
More leads is nice. More closed deals is the point.
Stop optimizing for leads that never close. Connect GoHighLevel to Splitlab and credit won deals to the page that produced them. Start testing or book a demo.
Built for revenue
Test for revenue, not just clicks.
Run A/B, multivariate, and multi-armed bandit experiments — with real revenue tied to every variation.
We’ll increase your conversion rate by 30% or you don’t pay. Terms apply.
✓
Revenue per variation
✓
A/B & MVT
✓
Multi-armed bandit
Test for revenue, not just clicks.
A/B, multivariate and multi-armed bandit experiments, with real revenue tied to every variation.


