Newsletter 7: key human behaviours for successful AI products

Artificial intelligence is still both a trending topic and a big issue — or at least, that’s how we think at Cosmic Velocity. At Websummit, one of the largest European tech events, over 50 talks and discussions were about AI with topics ranging from future regulation to digital transformation across PepsiCo’s end-to-end value chain. After our own presentation on design, leadership, and fundraising for impact with Zeilen Van Vrijheid founder Veronika Mutsei , attendees were eager to catch the next one about AI, exhibiting a mix of excitement, desire to learn, and anxiety.

At Cosmic Velocity we have been helping startups shape their generative AI product offering and experiences, solving related design problems, and are dedicated to addressing current and future developments, and sharing insights gained from practical experience.

Embracing the inevitable

When we think of generative AI tools, we usually mean systems that make something new based on its previous experiences, like Midjourney, ChatGPT, Stable Diffusion, DALL-E or Civitai. Usually they come with a free tier or a low-cost subscription.

We are merely at the beginning stages of this technology, making it impossible to fully grasp its potential or predict its impact on our lives. However, what we've seen so far is remarkable, and also often hard to navigate, especially when researching and designing for AI.

Despite numerous challenges in privacy, ethics, transparency, and unintended consequences, adopting a “wait and see” approach towards AI does not seem feasible anymore. It seems that virtually everyone is either building AI products or at least incorporating AI-driven features. No one wants to be left behind! If your competitors are already integrating ChatGPT (or other generative AI), you will be pressed to respond quickly with an innovative design or solution.

Challenges of building and integrating AI products

The first challenge is a lack of AI-specific mature strategy across the organization. Product owners often have no idea where to start and may be looking for frameworks to help address pressing challenges quickly. There are countless invisible features, like Spotify playlists or spell checkers, that can impact product adoption and become a competitive advantage. Incidentally, it may be promising to use AI to drive product adoption or reduce churning metrics — but will it work?

Another significant obstacle in AI integration lies in its high stakes for engineers, product teams, and leadership. For engineers, the sheer complexity and rapid advancements of the machine learning space make it hard to build. Product teams face a scarcity of real-world AI expertise, making it difficult to design for it. Additionally, small to medium-sized organizations may find the costs of implementing AI prohibitively high.

Lastly, the cost of development, training, and maintenance can be significant, therefore, mistakes become threatening. Recently, Air Canada was forced to give a partial refund to a grieving passenger who was misled by an airline chatbot hallucinating an airline's bereavement travel policy that did not exist in reality. No business ever wants to be in this place!

How does one even start? We start with the most important thing: people first. We always recommend starting with problem space research and identifying user needs and problems, and here are some practical learnings from our work.

5 key insights you need to know about people when working with AI

User groups gap

When it comes to AI user experience and adoption, the divide between experienced and non-experienced user groups is truly massive. Even though balancing different audiences is a common task for a digital product, it is the scale of the divide that makes it particularly different in the generative AI space. People who are already using and integrating AI into their workflows, and have seen the benefits of this approach, will have dramatically different requirements, mental models, and information needs.

Have I done this before?

Previous experience and frequency of use will dictate first impressions and adoption. People will evaluate a generative AI product based on other products and services they have tried or seen and will try to understand it using their own specific expertise.

• “Experts” will compare it with what they are already using, and existing workflow fit
• Less experienced people or those without a clear task fit need to first understand the idea and mentally evaluate whether it fits their needs.
• Complete newbies need to be reassured of the promise of the innovation as they may not be able to figure out the task fit.

Task is a core behaviour

Despite seemingly infinite possibilities, generative AI is exceptionally task-based, and its output is rarely a complete final outcome. The tasks usually range from very specific with an outcome in mind, e.g., creating a graphic novel illustrated by Midjourney, to a fuzzy idea without a pre-defined outcome, e.g., what happens if I use my ChatGPT as a therapist on my own journal notes.

Even formulating the prompt itself is a specific task and instruction for the machine, which means that exploration and discovery occur within a task-based context. Begin by identifying and modeling a subset of core tasks for different user groups and master them.

Don't Make Them Think!

When user tasks are clear, ease of use becomes paramount. It's a fundamental principle of usability, particularly with generative AI, encompassing overall goals, completion goals, and behavioral goals.

• Overall Goals: What your users aim to ultimately achieve.
• Completion Goals: What users expect as an outcome after using your product.
• Behavioral Goals: Actions users would take to achieve their goal without your product.

Focusing on and designing for your audience's behavioral goals is a key factor in achieving success.

Analogous thinking and tinkering

Most people won’t even attempt to grasp the details of how an AI application functions; the domain is simply too complex. Instead, they be using metaphors and analogy-rich thinking to achieve the desired output. We observed people employing a variety of strategies to bridge the gap between their judgments and the AI's predictions. Recognising these tendencies can significantly drive adoption of a product.

Tinkering is regarded as an integral part of the process and it embeds a mindset of experimentation even among people who might not typically engage in such behaviour.

AI products are often solutions in search of a problem. Innovation, therefore, becomes a function of strategic product decisions. It demands deep understanding of human complexity and an embrace of its unpredictability to build lasting products that are used and loved.

If you are working on similar problems, drop us a line to to see what your organization can do to resolve your challenges to adopting artificial intelligence as your product strategy