AI and Design Process. #2 Data processing
This is the second article in the series of Design Process and AI.
1.Data collection
2. Data processing (you are here)
3. Generation of ideas (WIP)
4. Prototyping (WIP)
5. Testing of hypotheses (WIP)
6. Transfer to development (WIP)
In this article, we will talk about data processing and how AI can help us with it. Data processing after data collection is one of those routine tasks that take a lot of time and effort in UX research. Processing 1 hour of interviews can take 4 or more hours, so AI can speed up the work here by automating routine tasks. For convenience, we’ll divide this topic into 2 parts: Transcription and analysis, and model creation.
Transcription and analysis
Over the past few years, audio transcription has not become something unique, and many services now do it with more or less good quality. Ideally, if your audio is in English, you’ll have a wider choice of services, both paid and free. Transcription of English is now available on almost every video communication service, so let’s look at the most popular ones:
- Zoom and Teams allow you to transcribe and also summarize the meeting
- Dovetail allows you to upload and process video, audio, and text documents, create insights from them, and conduct emotional analysis of the text
- Userview boasts that it can save up to 3 hours of interview time per user
- And, of course, you can upload documents to ChatGPT and analyse the data with queries such as:
- Analyse the data I have provided you with. I want to gain insights into user preferences and behavioral patterns to understand how they interact with food delivery services.
- Take the results of a product feedback survey in the document. The goal of the research was to understand how often there is a need and what users do to create a grocery store list for family to cook based on product X’s latest feature Y. Summarize the survey findings as a table.
I also have to add that no matter how perfect the transcription is, it does not always correspond to what the person said and meant. Therefore, you always need to check what was generated yourself, plus combine these listening sessions with watching a video and see not only what the person says, but also how they say it. Accordingly, automatic text analysis can help you see some insights that you have not seen before, but it is not a complete replacement for a personal review and will never be.
Creating models or research reports
In UX research, we often use models as an example of communication or research results, it can be either personas, JTDBs, CJMs, or even something custom model. The main thing is that the model should be clear to the person for whom these documents are prepared and proper level of details and generalizations. To create these models, we make “generalizations of generalizations”.
There are not so many solutions here because the understanding of the context is bigger, and the final result depends on the person who has done the research, but still, some solutions can be found:
- In Insight7, you can create models such as Persona, or Opportunity Solution Trees based on the data.
- Notably has a set of different Insight Templates that allows you to transform data into different models as you wish
- And of course, you can use ChatGPT to process insights. Given that insights have much fewer characters than raw data, it can be used even in a free one of ChatGPT. For example, for creating a persona, or preparing a research report presentation outline
More than text
It is clear that we, the designers, will still process ChatGPT and form it into diagrams to improve the information perception. But can ChatGPT do this? There are several ways to organize textual information from a chatbot:
- Ask it to organize everything on a table. This will dramatically increase the density of information and help it to be scanned
- Ask it to create a chart using Mermaid.js. Then you can use this code, for example, in Notion to see the visualization. This works with CJM, Opportunity solution tree, task analysis, etc. They are quite primitive, but sometimes it’s enough.
What not to do
I would still recommend you to refrain from generating and validating, your own research data based ONLY on synthetic data. Products such as SyntheticUsers or QoQo will only take you further into a fantasy world where you will be solving imaginary problems of imaginary people. Remember — AI is just a tool, and you are the one who has to make decisions.
Oleksandr Valius — Designer, manager, teacher and birder. Passionate about bringing order to the world through design.
