AI & Design Process. #1 Data collection

This is the first article about AI in Design process. Today’s article is the first — Data collection, or sometimes this phase is also called: research, requirements elicitation, emphasize

There are many methods of data collection at the beginning of a project. Some of the most popular ones are: domain research, audience research, user interviews, survey interviews, and analysis of existing documents. For this purpose, we will mostly use ChatGPT, since it contains a dump of the Internet, right?

Warning. The results may vary greatly depending on how you have Custom Instruction set up. Be sure to read the answers carefully. You can’t use the entire output of the AI in your project, so think of it more as “inspiration” and “preliminary work” rather than “generated and in production”.

Planning

I did not cover the topic of “planning” because it is too early to trust AI with this. To plan, you need to know a lot about the context of the project: available resources and budget, access to users and stakeholders, opportunities to be on-site or online. You can use queries for inspiration, but to get the most out of what’s available, each case needs to be considered separately.

Domain research

Domain research can be started but never finished. To get a quick start on a project, we need just the right depth of understanding to be able to start working on it. Here, I can share a few recipes on how to write a request so that it contains the maximum amount of useful information. The knowledge of different frameworks for market assessment is very useful. For example:

• Make an overview of external factors. Do PESTLE analysis of the proof of stake cryptocurrencies
• Ask him to generate a swap analysis. Give me the general overview of the domain of using drones in the agriculture. Describe the strengths, weaknesses, opportunities and threats of the industry in a table form

Also, when you start drowning in unfamiliar terms and words, don’t forget about the classic ELI5. Explain me like I’m five carbon credits

Audience research

First, let’s start with audience clustering to understand who we should focus on more. We use ChatGPT queries:

• Categorize types of tourists in Greece. Create a table with columns: name, description, volume. Sort by volume. Result
• What are the types of B2B customers that are looking for project management tools. Result
• Map types of customers and how they are dealing with problems with the car. Result

User interview

When creating questions for a user interview, I recommend that you generate the entire text from start to finish: greetings, permission to record, and warm-up are often forgotten due to stress, and when you ask in the middle of the interview if you can record, it doesn’t look very professional. Also, the more detailed you describe what you want to talk about and what aspects you would like to discuss, the better questions you will generate.

• Create a remote user interview script for an hour to gather detailed insights from people staying in a hotel in Dubai for the summer vacation. Include interviewer introduction, ask for the record permission, warm-up, and closing. Covers key topics such as user needs, preferences, and experiences with similar products. Result
• Prepare script for remote user interview of the user of the internal flood risk analysis tool. Include interviewer introduction, warm-up, set of open questions and closing. Ask about user role and position in the company, use cases for the tool, their pains and expectations on the user experience. Result
• Generate a script for on-site user interview of the food delivery app. Include interviewer introduction, warm-up, and closing questions. Topics to ask:
  - general expectations from the app and its user experience,
  - how the search and filtering is helpful
  - restaurant page information
  - price communication
  - delivery tracking
  Result

Surveys

Here it is important to work with the fact that the AI does not generate too many questions and to indicate in what form you want the result (the last example). And of course, the more specific answers you need, the more detailed you need to describe everything.

• Design a survey questions that capture user feedback on the experience of staying in a hotel in Dubai. Present the result as a table. Add a free form comment at the end. Result
• Develop a Questionnaire to collect qualitative data from bleisure tourists about their experience using amenities in hotel. Result
• Create a short 5-minute user feedback survey for users of product Twitter, which is a social media for microblogging. Twitter provides features that include sending tweets, sharing photos, videos, and voice messages, creating lists of users The survey should cover aspects such as ease of use, satisfaction with features, areas for improvement, and overall user experience. Please include a mix of question types, such as multiple choice, Likert scale, and open-ended questions. Result

In any case, it’s up to you to decide what to use and what to delete

Analysing documents

If you have ChatGPT Plus, this is not a question at all. You upload a document and ask questions as many times as you need. If there are a lot of documents, you can even create a custom GPT to talk to them about different topics, it’s also useful if you have people in your team with a paid version, then you can distribute it. The only thing is to ask control questions to check that it’s not a hallucination, but real data. The main thing is to torture data until it confesses.

If you don’t have a paid version, other tools will be useful here. I tried

• https://www.chatpdf.com
• https://docanalyzer.ai
• https://askyourpdf.com/

And they all give more or less the same results, the first one is a little better

Thank you for reading. Have any questions? You are welcome to ask!

Oleksandr Valius — Designer, manager, teacher and birder. Passionate about bringing order to the world through design.