What it fixes
Most local pages get a few hundred visits a month. A classic split test on that traffic often never reaches a result, and a wrong conclusion costs you real jobs. We choose a method your volume can support: one big change with equal before and after periods, or a clean split when traffic allows. We note the season, ad spend and other changes in the window. We read the answer as CRM-confirmed leads and booked estimates, never clicks.
What the research says
- 12%
About one test in eight produces a significant gain on its main metric. Plan for most tests to lose or tie.
Optimizely, Top 10 takeaways from running 173,000 experiments (2026) - 60%
In a healthy testing program, roughly 60% of tests produce no clear win or loss, so budget traffic for inconclusive results.
Optimizely, Top 10 takeaways from running 173,000 experiments (2026) - 10%
Startups that adopted A/B testing had roughly 10% more weekly page views, in a study of 35,262 startups.
Harvard Business School Working Knowledge, Is A/B Testing Effective? Evidence from 35,000 Startups (2021)
What's inside
- 01
Count your monthly visits and CRM-confirmed leads
- 02
Decide what choice this test must settle
- 03
Pick a method your volume supports
- 04
Write the plan: change, baseline, target, start, end
- 05
Note season, spend and other changes in the window
- 06
Read the result as a range and decide
Deep-dives inside the playbook
- Ad platform tests
- Sample size and run length for low traffic
See it in action
Our Austin bathroom page gets about 400 visits a month. Should we split test two headlines?
Method: not a split test. At 400 visits you would wait a very long time for a clear result. Instead, change the headline and button together, then compare equal 8-week periods in CRM-confirmed leads. The plan lists baseline, target, start and end dates, and the risks, including a spring spend change. A read-out template waits for the result range, volumes, decision and lesson.
At $10M and up
How it runs in a larger operation
At $10M and above, you have the volume to test, and the risk is testing the wrong thing in the wrong place. Tests run location by location against CRM-confirmed leads and held appointments. A named person approves every change before it goes live, and leadership sees one result per test in the daily readout.
- One test calendar across all locations, no overlaps
- Results counted as held appointments, not form views
- Each test ends in a keep, change or stop decision
