A useful paid social test starts with a decision you will make afterwards. Choose one question, define the result that matters, keep other important conditions comparable and set the spending limit before launch. If the evidence is too small or inconsistent, record an inconclusive result instead of inventing a winner.
Write the decision first
“Which advert is best?” is usually too broad. Try “Does showing the working process produce more suitable enquiries than showing the finished result?” That tells you what must differ and what should stay comparable.
Write the action you would take after a clear result. You might commission more process footage, change the main message or stop using an unsupported assumption. If no practical decision follows, the test may not be worth the spend.
Keep the question relevant to the business. A colour preference is unlikely to be the most important uncertainty when nobody understands the offer.
Choose a fair comparison
In a controlled test, keep the audience, offer, destination and measurement consistent unless one of those is the variable being tested. Changing everything creates two different campaigns, but does not tell you which change mattered.
LinkedIn’s A/B testing documentation, for example, describes a comparison between ad sets differing in one variable. Use the current experiment tools and requirements for the platform you are running, rather than assuming all platforms work identically.
Simply placing two adverts in a normal campaign is not necessarily a fair experiment. Delivery may be uneven, and the people seeing each version may differ. That can still provide useful operational information, but describe the evidence accurately.
Decide what counts as success
Choose the primary measure before seeing the results. For an enquiry campaign, that might be cost per suitable enquiry, with a written definition of suitable. Clicks and reactions can help explain behaviour, but they should not quietly replace the business measure.
Also choose safeguards. A variant should not win simply by attracting unsuitable requests, creating misleading expectations or overwhelming the people handling responses.
For an illustrative appointment-based service, a low-cost form submission may be less useful than an enquiry that fits the service and can be followed up. Use the same quality check for both variants.
Make a short test sheet
- Question: the specific uncertainty the test will address.
- Variants: what changes and what stays comparable.
- Primary measure: the result used to judge the test.
- Quality check: how suitable enquiries will be identified.
- Spending limit: the authorised total and who monitors it.
- Timing: planned start, review point and allowance for delayed outcomes.
- Decision rule: what evidence would justify a change.
- Stop conditions: broken tracking, incorrect content or an operational problem.
The sheet is there to prevent the rules moving halfway through. Save the creative files and destination versions with it so you can reconstruct what people actually saw.
Give the result enough time to exist
A click can happen immediately. A serious enquiry may arrive later, and its suitability may take longer to establish. Choose the observation period around that reality and the platform’s current test requirements.
There is no universal number of days or enquiries that makes every test conclusive. The expected difference, variation and volume all matter. If the decision is expensive, get the test design checked before committing the money.
Do not end a test just because one version briefly looks ahead. Keep monitoring for faults, but separate operational safety checks from repeated attempts to declare a winner.
Allow an inconclusive result
LinkedIn’s test-results guidance distinguishes clear results from directional or inconclusive ones. That is a useful discipline whatever platform you use.
If the evidence is weak, say so. You might continue within a newly approved limit, test a larger difference or use customer interviews to understand the uncertainty. “We have not established a difference” is useful information.
Do not choose a different metric after the event just because it makes one version look better. You can record an interesting secondary observation and design the next test around it.
Write down what you learnt
Record the conditions, result, uncertainty and resulting decision. Keep the conclusion narrow. A message working for one audience and offer does not prove it will work for every service next year.
Review the quality of the enquiries after the platform report is complete. If apparent performance disappears when real suitability is checked, that belongs in the conclusion.
Our paid social service joins creative decisions with practical reporting. Prepare the assets with a paid social creative brief, and use the guide to social engagement figures to keep supporting metrics in their proper place.