How does it all come together?

How can AI support research preparation? Rethinking a real project (artificial intelligence)

The following article is based on the presentation of the same title by Balázs Dombóvári and László Horváth, prepared for this year's Service Design Day conference (SDD 2025).

At SDD, the Works. team devoted an entire morning to presenting the opportunities we see in AI in autumn 2025, and where we currently stand in optimising our own research processes with its help. (artificial intelligence)

Balázs and Laci examined the beginning of the research process, with particular attention to interpreting the brief and selecting the right methodology.

An enquiry from spring 2024

Last spring, we received a brief from one of our telecommunications partners that was roughly as follows:

At the time, following the traditional approach, we:

• interpreted the brief,
• formulated hypotheses and then research questions,
• selected the methodology,
• then developed the tools used for the study (discussion guide, questionnaire, interview guide).

It was a creative process on a human scale. We were also familiar with researcher loneliness: that state when you are scrolling through a list of hypotheses on screen, wondering whether these really are the most appropriate questions.

What if we were starting today?

What would we do differently if we were embarking on the same project today? How would artificial intelligence support our work? This thought experiment was not about automating research and removing people from the process entirely. Rather, it was about exploring how AI could become an assistant and thinking partner, and how it might affect the researcher’s role. (artificial intelligence)

1. Brief

The very first phase of research is interpretation. A brief always originates from the client's perspective, and our task is to uncover its underlying layers, real expectations and assumptions. Of course, an experienced researcher can draw these out of the business in a workshop – they simply need to keep asking: “Why?”

While asking follow-up questions remains important, using AI as a question generator could shorten the time between receiving the brief and preparing a proposal. How? (artificial intelligence)

The response gave us what we expected: new perspectives – similar to ones we had prepared for ourselves, but not all of them and not quite in this form.

A few examples of the questions:

• Is the remote-control development a standalone project, or part of a broader product or service development initiative?
• What is the client's definition of success: cost reduction, user satisfaction, fewer customer service calls…?
• In what situations is the remote control used: traditional TV viewing, streaming, gaming, smart home control?
• Which devices does the remote control need to operate?
• Are there any brand strategy or technological constraints?
• Who makes the decision, and how many levels of approval are required?

2. Hypotheses

Next, we explored what hypotheses AI would form based on our partner’s personas. (artificial intelligence)

What both approaches had in common:

• they assumed different usage patterns (background viewing, intentional content consumption, etc.),
• they assumed different levels of user knowledge,
• and they recognised that the role of streaming has become more important compared with linear TV viewing.

The expert thought at a systems level

• examining the service experience and the entire ecosystem,
• taking temporal and social patterns into account,
• identifying specific usability challenges,
• and considering how a device could replace other functions within the household.

AI’s focus was operational and role-based (artificial intelligence)

• it wrote hypotheses assigned to personas, and
• focused on the remote control’s specific features and modes of use.

So AI did not think for us; instead, it opened up a new perspective on the same problem. While we were thinking at a systems level, it was looking for patterns at a micro level. (artificial intelligence)

A prompt is like research

For both, the quality of the input determines the output. The more precise the prompt (or research question), the more reliable the response (and the research findings).

3. Methodology

We originally used a mixed-methods approach for the project:

1. Think-aloud videos with users, based on a defined scenario
2. Co-creation workshop with users to prototype the remote control
3. Co-creation workshop with the expert team to narrow down and select potential directions

AI also introduced alternative methodologies, including: (artificial intelligence)

1. Interviews in the home environment We had considered this, but ruled it out for cost-efficiency reasons.
2. Clickable prototype We had considered this, but felt that, as it is a physical object, it would be difficult to assess virtually, potentially skewing the results significantly. If complemented by testing a physical prototype, however, it could be worth considering.
3. Online questionnaire validation of the prototype This had not occurred to us, and we still do not consider it a good idea today; see the previous point.
4. Expert remote-control analysis workshop This had not occurred to us at all: our client specifically wanted to amplify the voice of customers, rather than hear expert opinions.

Final thoughts

Using artificial intelligence can be a huge help even in the early stages of research:

1. It broadens perspectives effectively: it brings new considerations and alternative approaches, while enabling dialogue and rapid iteration.
2. It eliminates researcher isolation – a companion when working through dilemmas, if there is nobody to bounce ideas off.
3. For less experienced researchers, it can provide – sometimes perhaps misleadingly – confidence, helping them structure their questions and recognise missing connections.
4. It can formulate questions tailored to the industry and business context.
5. Does it make work faster? Yes. But more importantly, it can improve quality.
6. The key principle of working with AI is this: weak input leads to weak output. AI does not automate; it assists – supporting, helping to structure and inspiring. (artificial intelligence)
7. Responsibility still lies with people. For now, the researcher is still the one who makes the decision at the end of the day.

This article was written by Horváth László, Senior Experience Designer at Works.