Creative Testing

Results from creative testing help the creative team to revise concepts, drop ones that do not resonate with the audience, and identify which ones the audience likes best. It can be used for market research before a product launch, or to find better performing variations of an existing brand. Growth is a function of the number of experiments you run.

If you want your marketing to perform, you need to find the combination of factors that resonate with your target audience. As Ogilvy said: “Never stop testing and your advertising will never stop improving”. Your adcopy, design and product should all work together – if just one component is off, your campaign won’t work. Find the right combination of memes, and you win.

In 1923 Hopkins published Scientific Advertising, a book that covered techniques like split testing and coupon-based tracking. Ogilvy wrote of the book "Nobody should be allowed to have anything to do with advertising until he has read this book seven times”. RCTs (Randomized Controlled Trials) or A/B tests as marketers know them are at the top of the evidence pyramid – the gold standard for proving one thing caused another.

Using scientific principles and methods you can prove what works and gain an edge over the competition. If you don’t test new creative or your tests aren’t well documented, your organization is at risk of forgetting what worked (or didn’t). Winning tactics are accidentally abandoned and failed experiments unknowingly repeated. If early insights weren’t statistically valid, some misses will be miscategorized at hits (Type I error), and some hits as misses (Type II error).

To ensure your results are valid, change one variable at a time, carefully setting out your plan in advance. The group exposed should be a large enough for statistical significance, and randomly assigned the treatment or control. Publish openly to build trust in the methodology, and let others build on your work.

Most companies land on their initial winning combo by intuition and luck. Attempting data-driven decision making in the early stages of company-building is a mistake. There’s more noise than signal, and you’ll suffer from analysis paralysis. Anything that’s easily changeable or reversible “can and should be made quickly by high judgment individuals or small groups” as Bezos says. Just start trying things, and once you find something that works, double down on that area to scale.

Modern ad platforms optimize creative with AI, but that can mean one ad gets 90% of the budget, leaving us unsure if the rest were given a fair chance. There’s no substitute for split testing to learn if one creative performs better than another, but experiments are expensive and slow. So when do we prioritize human learning over machine learning?

Time in your testing schedule should be reserved for the most consequential strategic decisions – that can’t be easily reversed – what Bezos calls ‘one-way doors’. What major functions, attributes or features are most important to figure out. Once you’ve identified the biggest levers for growth, drill down into themes within each major concept. As you get more granular, small 1-2% differences won’t reach statistical significance, so you can abandon science and let the algorithm decide.

When testing it’s important to make space for scientific testing, but also cater for algorithmic optimization. You need to protect your ‘business as usual’ (BAU) campaigns from disruptions from poor performing test variations. However if you don’t test new creative your incumbent ads will fatigue and start to decay in terms of performance.

Here’s the best way I’ve found to navigate this trade-off:

I’ve found that this works across channels and industries. It lets you set your relative risk-reward by adjusting your testing budget: 5% for risk-averse brands and as much as 30% if you need a big win.

If a platform doesn’t explicitly offer split testing, then you can approximate it by targeting the same audience with multiple campaigns.

BAU campaigns target the same audiences as Test campaigns, so there is overlap, but given Test campaigns are a small percentage of BAU performance shouldn’t be impacts by much.