How can you manage risks and issues in IA projects?

For sure key risks include data quality and availability, as AI systems rely heavily on large volumes of high-quality data. Ethical and privacy concerns are paramount, given the sensitive nature of data and potential biases in AI algorithms and change based on the industry. In my experience it is also relevant managing stakeholder expectations and preparing for the potential impact on employment and workflows is essential. Without the proper adoption phase, also the most powerful tools can be unhelpful.

A better way to identify risk is understanding the type of outcomes you anticipate, for instance when building navigation items and hoping to leverage Information Architecture, there is a need to consider the type of method which can range from simple card sorting with internal teams to advance card sorting which requires recruiting users either remotely or in internal labs. A bigger part is also identifying if open-card sorting or closed sorting will be leveraged based on the type of aggregation and architecture to be done. Involving the right audience or subject matter participants may make all the difference in objective or subjective data quality as well.

A personal experience while trying to architect a list of user menu for a banking app had us acknowledging that we are dealing with a wide range of users crossing multiple levels of financial literacy, we had to do an open card sorting, while also introducing an additional description for items specified. This may sound a bit giving off too much information and unconventional, it helped to educate the users on financial terms that may otherwise cause the users to just randomize their choices of categorization in the interest of time constraint. However, this allowed us access to analytical results with over 70% participant's understanding of card items and helping categorize data effectively. Recruiting the right users is quite important.

Depending on the level of organization's design maturity and data-driven penchant, this may determine the risk evaluation outcome, while budget may be an issue with recruiting users, other factors may be driven by time, tools, bias etc. For a limited team architecting internal tool in an on-prem situation, X-sort is a free and good tool, though interface looks old but it will still get the job done, for remote teams, "optimal workshops" has worked great for me in the past and grants access to similarity matrix and other analytically sane data points.

I believe an additional area to look out for is bias, we sometimes see data that is not reflective of the post-implementation behavior of users, this may make us force the data in a particular direction and this will distort user experience and sometime make discoverability an issue. While users have ways of associating data, we will have to leverage UX research instruments in understanding the delta between post-implementation user behaviors and analytical IA output.