In order to get the best data from Vizit, here are some tips for creating benchmarks:
- Quality is better than quantity. A minimum of 30 images is required for a benchmark, and the upper limit is 1,000. But if you include too many low-quality images, or images that don’t fall within the benchmark category, your comparison points will be weak.
- Get at least 30. A benchmark with fewer than 30 images is considered a draft and cannot be added to a project. A project without a benchmark can be audience-ranked, but you won't be able to view any detailed scoring or comparisons.
- Quantity is also important. Just as standard market research relies on large sample sizes, a benchmark should have enough images to represent the actual market it represents. We recommend at least 50 assets for a robust benchmark.
- Compare apples to apples. Content segmentation should be similar in a benchmark. For example, a benchmark should contain only scenery images, or images with people, or lifestyle images. Comparing a hero shot with below-the-fold content will not be useful.
- Use design-centric or concept-based images. The Vizit AI engine evaluates imagery, not text. So avoid including text-heavy images such as content labels, instructions, etc. If you do want to evaluate text-heavy images, be sure to create a benchmark to compare similar text-heavy designs.
- Include images from similar sources. Use separate benchmarks for social media images, carousel images, product shots from Amazon or other retail sites, etc.
- Keep the numbers consistent. If a benchmark contains images from multiple sources, it’s a good idea to include a similar number of assets from each source.
- Benchmarks need regular updating. Content on the digital shelf is constantly changing, so do your best to make sure your images are current, and that you know the current competition. Your Vizit strategy team can help formulate an update timeline, depending on your product categories.
- Similar images are fine; duplicates are not. The AI engine is great for comparing similar images - even tiny differences in font size or image location can produce different scores. But a benchmark that contains exact duplicates may produce skewed results.