Starting with Observation

Simply defining a problem does not mean that you have a solution. To begin to explore solutions, you must first spend time observing human behavior in context, as this presents the richest empirical insight. Even if you have a solution in mind, and even if your solution is mostly behind-the-scenes, observing human behavior will uncover crucial patterns and edge cases. Regardless, observation is best conducted without a clear solution in mind at all.

Let’s take the same example scenario as before. For the house price estimation problem, you probably need to find a few real estate experts to interview as well as home buyers at various stages in the process. You should take notes on how experts interact with buyers to find clues about the required data and experience. In addition, you should study what a given buyer does with that information once they have it. You may find a host of unexpected uses for a house price. Perhaps someone was simply curious and didn’t intend to use that information at all (are you misunderstanding pain points?). Perhaps they were interested in renting a property they own (are you solving the right problem?). Maybe they were looking for nearby houses with a lower price or looking for similar houses far away (are you providing the right insight?). All of these findings would point to vastly different user interactions and experiences to design. Be careful in assuming that each of your findings correlates to a different ‘feature’ required by one super-product. Instead, your entire product could revolve around a single feature such as one unique use of a house price.

Especially in the context of AI, it is crucial to observe humans making use of the kind of information you want your product to provide. Users will often require more than the information itself. They may need an agent on hand to answer questions or to clarify ambiguities (see Embodiment). They may want multiple interpretations (see Transparency), or to challenge the decision once they see it (see Errata). They may simply want the information presented in a different way (see Intuition). For example, instead of a fixed house price, perhaps they want a percentage difference between that house and the price of houses in its neighborhood.

Observation may seem to lack the rigor of a scientific investigation (e.g. instead of interviewing one person, why not send out a survey?) However, surveys tend to exchange observed (or revealed) behavior for verbal (or expressed) behavior. This difference can be significant. Someone may claim on a survey that they use less salt in cooking, but turn out to use more. However, simply observing them cook may show their thought process in context. Similarly, a person may claim that they want a housing price estimate to be highly accurate, but in reality they just want to use it for reference. Use observation to study nuances in behavior, while using surveys to collect broad information like demographics.

There is, however, a possibility for the rich use of data in your design process—something we dive into further in the next section.