How AI is transforming research methods

Artificial Intelligence is now an indispensable part of the world of innovation, but you, as a reader, probably already knew that, so I don’t need to tell you again. At MakerLab, we use tools like ChatGPT, custom GPTs, and our own built chatbots on a daily basis. As a result, the team's 'prompt engineering' skills have grown significantly over the past year.

‍

At the time of writing, I see three general applications of AI at MakerLab:

‍

1. Supporting our work: In addition to using ChatGPT and custom GPTs for anything and everything, our knowledge base has, for example, a chatbot that we can consult on how best to use certain experiments and methodologies.
2. Research Assistant: Think about being able to conduct desk research more quickly, analyse large datasets, and refine interview scripts using the ‘The Mom Test’ method.
3. AI-powered final concepts: We are able to create AI prototypes and deliver them as validated final concepts. For example, Equalwrites - developed by colleague Melissa - uses AI to check texts for the use of inclusive language.
   ‍

On each of these three points, we are accelerating and enhancing our processes and innovation methodologies with AI. In this blog post, I will focus on a practical example of the second point: applications of AI as an assistant in conducting research. AI offers many new forms and opportunities for conducting research.

‍

To illustrate this with practical examples, I will walk you through a project we recently completed for a client, which involved conducting a validation study for a digital platform. During this validation study, we utilised AI in two ways:

‍

• Participants from our target audience completed an AI-powered questionnaire to come up with possible features.
• Synthetic personas were created and interviewed using a self-built GPT.
  ‍

I will explain why we chose to use AI in both cases by putting it in the context of the project.

‍

Formless

‍

The core of the validation research involved conducting two experiments: online interviews with the target group and a survey among current customers, aimed at exploring potential features for the platform. For the survey, we sought a way to enable participants to interactively brainstorm features remotely, rather than evaluating pre-defined features (as in a KANO test). We achieved this through an AI-powered questionnaire in 'Formless', an AI tool from the well-known survey software company Typeform.

‍

To fully understand how this AI tool added value to this research, it's useful to know how this tool differs from a 'traditional' online survey. One of the key differences is that in Formless, you input prompts instead of questions. In other words, you don’t enter a set of questions, but rather give Formless AI instructions to ask your questions. For example: “Find out what kind of role the participant is working in and how long the participant has approximately been in this role.”

‍

The AI is trained to facilitate two-way conversations between ‘the questionnaire’ and the participant. This allows both the AI and the participant to ask questions and provide answers. Therefore, the AI not only poses questions to the participant but also responds to their answers. As a result, the interaction experienced by the participant is much more conversational than a traditional questionnaire.

‍

To provide an example: if a participant answers the question “What is your current role, and how long have you been in this position?” with “Facility manager”, the AI would then respond, “That’s an important role. How long have you been working as a Facility manager?” In this example, the AI not only acknowledges the participant’s answer but also addresses the part of the question that remains unanswered (length of service). This way, the AI continues an ongoing conversation and gathers all the insights you need.

‍

The coolest thing we built into the questionnaire is a section where participants co-create features for the proposition together with the AI. This is built as follows:

‍

• AI: Can you describe a specific challenge or problem you are experiencing within [task]?
• Participant: Expresses a particular problem.
• AI: What would an ideal feature look like that helps you solve [problem]?
• Participant: Gives an idea for a feature.

‍

The AI then generates three features based on the input within the context of the proposition and asks, “Which of these functionalities do you find most valuable for your work, and can you explain what you achieve with it?” After making a choice, the participant continues to iterate on this feature with Formless AI until they are satisfied with the result.

‍

This co-creation ensures that creative, concrete feature directions emerge from such a conversation. As a result, we were able to specify exactly what value the proposition should deliver to customers.

‍

Synthetic personas

‍

The second part where we employed AI was to enrich the insights we had already gained after conducting target audience interviews and the AI-powered questionnaire. The aim was to further explore potentially interesting features for the platform. To achieve this, we used AI to generate synthetic personas and subsequently interviewed this resulting 'additional target group panel' about their needs.

‍

To create these synthetic personas, we first collaborated with the client to develop personas for four different target audience segments. Each of these four 'seed personas' was input into a custom-built 'Persona Multiplier GPT'. Based on a single seed persona, this GPT generates five new personas, each with seventeen different attributes. These are not just standard attributes like name, age, and gender, but also include factors such as socioeconomic status, personality traits, and decision-making motivations. These attributes have been chosen to create a detailed and versatile profile for each persona.

‍

We asked the twenty synthetic personas (five from each segment) a hefty list of questions, corresponding to the questions from the interviews we had previously conducted with "real people. The personas went through all these questions one by one and answered them in great detail. Asking clear questions and follow-up questions is crucial for such an interview to avoid generic answers. This also allowed us to retrace for each thought feature from which pain or need of the personas this emerged.

‍

Conducting these interviews resulted in an immense dataset. From this, we were able to extract a list of features identified by the synthetic personas. We then compared these features with those we had previously gathered from the target group. We discovered a significant overlap, which was quite exciting to see! Even though you are comparing your 'real' results with synthetic ones, you still get a kind of validation.

‍

In addition, the conversation with the synthetic personas also yielded new features not previously mentioned. We passed these along to the client for inspiration for further development of the proposition.

‍

Conclusion

‍

Utilising Formless in this project was a revelation in how easily you can set up an AI-powered questionnaire that conducts thorough research on your behalf. It’s fascinating to see in the transcripts how participants intuitively understood the conversation flow with the AI and, together with the AI, were able to arrive at very concrete features.

‍

When using synthetic personas, I believe that, as designers, we must remain sceptical and discerning. After all, they are synthetic personas, so you will get synthetic results. These results should not always be considered the ultimate truth about real people, as that would be too simplistic. What real people say remains the guiding factor, but AI can certainly be useful for enriching results and broadening ideas. Additionally, there are other valuable applications for these personas, such as providing feedback on the initial ideas you've developed for a new proposition.

‍

In hindsight, I am very pleased with the ways we utilised AI during this project and what we have learned from it. AI tools are increasingly proving and earning their place in our work. What I have described in this blog is just the tip of the iceberg when it comes to the use of AI in conducting research. However, it is important that we remain vigilant about when to use AI and the reliability of AI-generated results.

‍