The three approaches to billboard pre-testing
Most tools sold for billboard and out-of-home (OOH) pre-testing fit one of three approaches. None is "best" for every question. What matters is knowing which question each one answers.
Swipe to see the full table →
| Question | Perception panels | Predictive attention models | Deterministic rule-based review |
|---|---|---|---|
| What it measures | What people say, feel, or remember after seeing the ad | Where viewers are likely to look, estimated by a trained model | Measurable properties of the artwork, such as contrast, copy length, hierarchy, and letter size, under stated viewing conditions |
| Where the result comes from | Answers from a recruited sample of people | Patterns the model learned from its training data, often eye-tracking recordings | Fixed, published rules applied to the file and settings |
| Same input, same result? | VariesA new sample can give different answers | Usually, per model versionCan change when the model is retrained | Yes, per software versionSame file, copy, and settings give the same result |
| Can you trace why? | Through the questions and responses | Harder: the logic sits in learned model weights | Yes: each score maps to a stated rule |
| OOH conditions | Only if the study simulates speed and distance | Depends on the training data. Widely used benchmarks record people looking at images on a screen, not at road speed. | Only if the rules include format, distance, and speed |
| What it can't tell you | Whether the ad is physically readable at speed, unless tested that way | Whether the copy is long enough to read in the time available | How people feel about the ad, or whether it will drive sales |
Notes: "Predictive attention models" refers to computational saliency models, which researchers evaluate against human eye-tracking data. The MIT Saliency Benchmark, for example, scores models against fixations recorded while people viewed images on a screen. Research has also found that these models can generalize unevenly to new types of images. Sources are listed at the end of this article.
Panels and attention models tell you about people's responses. Rule-based review tells you whether the message can physically be read at the speed and distance you choose. For OOH, the second question comes first. For more on panels, see algorithmic vs panel testing and why traditional billboard testing methods fail.