Many organisations have rushed into “integrating” AI into their processes and products. However the successful integration of AI into existing workflows requires careful planning and an intentional approach. Without this, organisations risking disrupting the user experience, potentially becoming an additional tool that users must manage, rather than genuinely helpful enhancement.

A common implementation approach involves deploying chatbots, which require users’ input queries and validation of AI-generated responses. While this can be effective for certain workflows, it is often not the ideal approach across all contexts.

The Chatbot Dilemma

While implementing a GenAI chatbot is a common approach for integrating AI into products, it introduces a new user flow that may not have existed previously. Additionally, the probabilistic nature of AI means that output can vary significantly based on user input, which can create inconsistencies. As a result, users are often required to construct and articulate prompts carefully to achieve the most accurate outcomes.

Considering the wide range of user interactions, like users whose first language is not English (in cases where the app is in English), users typing on mobile devices, or those prone to typos, the experience and results may vary significantly based on input factors outside the control of the solution or product.

Moreover, in many implementations, AI is integrated in a passive manner, e.g. positioned at the bottom of the screen, waiting for the user to initiate the AI interaction or request information, which can limit the proactive engagement that AI could potentially offer.

Context-aware Assistance

Following an approach with passive intelligence in the background, the user experience can be improved as AI provides assistance when and where needed.

For example, AI could function unobtrusively, monitoring user actions and anticipating user needs. With smart triggers it can suggest help or resources only when it detects signs that the user may benefit from them.

Automatically triggering AI assistance requires confidence that the AI will be genuinely helpful to the user. There will, however, be instances where our confidence in its usefulness or accuracy may be lower. This is where appropriate UX patterns can be used to make a subtle suggestion. When confidence is lower, a reveal-on-click option, similar to the passive implementation many applications use, can give users control over whether to engage with the AI’s suggestion. When the confidence is higher, AI could intersect with existing the user’s workflows to guide them more quickly toward the desired outcome.

In this approach, contextual understanding from AI is crucial, and user feedback can help refine AI’s responses over time. This strategy increases the relevance of AI within a product or workflow, respects users’ autonomy and minimises disruptions, ultimately contributing to higher user satisfaction.

Conclusion

In summary, while I am a strong advocate of AI and firmly believe it can add valuable insights to numerous workflows, organisations must be careful to integrate their AI solutions in ways that enhance rather than disrupt user experiences.