Secondary research
Last updated
Last updated
Secondary research (also called desk research) refers to the process of collecting and analyzing existing data. There are many different types of existing data and the process and insights might vary.
Data logs may be available for a similar interaction in a channel that is close to the conversational AI application, such as IVR call logs.
While it can be helpful to analyze this type of information, it is important to be mindful of the differences between channels and how that impacts how transferable the findings really are. For instance, IVRs typically represent structured workflows, and users are often conditioned to interact with a speech-enabled IVR in keywords (not natural language). This information is likely not a great fit when building a truly conversational AI.
It may also be possible to review data logs from other channels, such as a mobile app or website. Learning anything from this data that is transferable to a conversational experience may be questionable. For instance, the most popular features on a website may not translate well to a conversational experience, such as anything that involves a longer list of items, and the context of the user (where they are, what they are doing while using a conversational experience) may vary greatly.
Any prior user research that was conducted on other channels may be a source of information, specifically around expectations for functionality, level of concerns around e.g. trustworthiness, etc.
The internet is a great source of information. As always, consider the source of the information. LLMs are another way to gather information.
Market research and similar in-depth reports can be obtained from certain organizations, but they often come with a (hefty) price tag. A free resource is the .
Just like competitive research with users is a viable method, a heuristic evaluation of such an interface is a far less involved initiative. A user experience professional reviews a product and, based on their experience, rate aspects of the user experience, and they can also make note of available functionality. The main goal is to identify particularly great or poor aspects of the experience, so that the former can be considered in design, and the latter can be avoided.
The methods to select will depend on the circumstances; not every organization has access to relevant data logs or prior research. A good practice dictates that these areas are investigated; if you don't ask, you may not know that some relevant prior information is available to you.