What should you test first?
A practical framework for prioritizing test ideas that drive real business impact.
Split Test Pro team · 6 min read
More ideas than time
Once you start testing, ideas pile up fast — a competitor’s page, a support question, a funnel drop-off, a debate in a meeting. You’ll always have more of them than traffic to test them with. Picking well is what separates a program that compounds from one that spins its wheels.
Start where it matters most
For most stores and SaaS sites, the highest-leverage places to test are the ones every buyer passes through:
Your main call to action
The add to cart or sign-up button: its label, size, colour and position.
Headlines and pricing
Headline wording, and how the price and compare-at price are presented.
Trust and friction
Reviews and trust badges near the decision, and optional form fields you can remove.
Avoid starting with low-traffic or low-leverage changes, like a footer colour or a 404 page redesign. Even a winner there won’t move your overall numbers.
Score every idea with ICE
ICE is a quick way to rank a backlog. Score each idea from 1 to 3 on three questions, add them up, and test the highest scores first:
| Factor | Question | 3 means |
|---|---|---|
| Impact | If this wins, how much could it move the primary metric? | Major |
| Confidence | How strongly do we believe it will work? | Very sure |
| Ease | How simple is it to build and test? | Trivial |
ICE isn’t precise — two people can score the same idea differently. That’s the point: it surfaces the disagreements worth talking about. Re-score from time to time, because an idea that was a 7 last month may be a 5 once the page around it changes.
Let Splitty suggest ideas for a page
Not sure where to begin on a page? Open Splitty on it and describe what you want to improve. It suggests up to five ideas, each with a hypothesis, the change, the metric, the audience and the effort — so you can start with a low-effort idea aimed at the metric you care about.
Or let Test Plans keep the list for you
On Pro and Agency plans, Test Plans turns prioritizing into a ranked queue. Pick a north-star metric and Split Test Pro maps your funnel and generates a backlog of hypotheses, ranked by projected impact, effort and a feasibility check — so an idea your traffic can’t decide never crowds out one it can.
You stay in control: remove anything you disagree with, and build the next test when it reaches the top. A built test is a draft — nothing starts on its own.
One test, one change
Pick the top idea
Take the highest ICE score — or the top of your queue — on a page with real traffic.
Change one thing
One change per variant, so a win tells you exactly what worked.
Run it to a decision
Wait for 95% and at least a full week, then apply the winner before the next test on that page.
Most tests won’t win — even strong programs see 10–30% of tests produce significant improvements. That’s normal, and every result sharpens the next idea. For the full method, see A/B Testing Methodology and Building a Testing Program.
