Design Thinking for Data Scientists — Need-finding Planning

8 min readSep 27, 2020

Author(s): Tara Su

Abstract. I would like to redesign the “all inboxes” interface of the mailboxes software for the iPhone. Mailboxes are a relatively mature product. It comes as a default with the phone therefore has a very large user base. Nevertheless, the “all inboxes” interface is less than ideal as a modern cell phone app interface sitting in between the users and their tasks. In this report, I will describe the need-finding plan to help me shape my redesign.

Problem space

Mailboxes are one of the most actively used iPhone apps in the world. It comes as a default with the phone. It was integrated since the first generation iPhone and is considered to be a mature app. Nevertheless, the “all inboxes” interface of mailboxes is less than ideal as a modern cell phone app interface. It is intuitive to understand but lacks the affordance one would expect with the advancement of Artificial Intelligent (AI).

A usability gap is revealed when the user is active in both professional and personal domains and has a short time span to grasp the important information from emails. For someone who receives work-related emails, important personal emails, and advertisements all through the cell phones, it is very hard to quickly check the “all inboxes” during busy working hours and grasp the important messages without being buried by a high number of promotional messages. Thus, the scenarios I would focus on is the information-rich environment, such as public area and workspace, where users don’t have the luxury to dedicate a large amount of time on combing through their emails.

The goal of this report is to form a plan on finding the need of improving this commonly used app and interface. Through this plan, I hope to deeply understand the users’ pain points, which will shape the direction of my redesign.

User type

Figure 1: user type of need-finding

In the need-finding phase, I would like to focus on users who are active in both professional and personal fronts (Figure 1B). They receive a large number of important emails for work and important personal matters, such as bills, personal contacts, and receipts. They also receive a large amount of social media and promotional information. The users who fit this scenario are adults in the workforce. They are educated, have a decent income, and use cell phone email app regularly, although not necessarily know the app inside out. They want to extract important and time-sensitive information when they are busy and catch up on personal, social, and shopping activities when relaxed (Figure 1A).

Needfinding plan 1: participant observation

I normally start need-finding with naturalistic observation. In this project, I did start from here and formed the idea of tackling this specific part of the mailboxes interface. I see people checking their emails on the train, in the queue waiting for a lunch order, during meetings, in between keystroke, walking in the hallway, and even in the bathroom. I see them half concentrating on something else and scanning through their screen. I see their facial expression changes. I know checking email is something they do very often and not completely hassle-free. But I can’t really closely observe them and nail down the pain points. This is such a private activity, I can’t get enough information by just naturalistic observation. So I think the first step for me is to try it out myself.

I am already one of them. I am a working professional who receives work-related emails and personal important emails through the app. I also got too much promotional information and social media pushes that I can’t care less during busy hours. I scan through my emails whenever I have the opportunity but still miss important information and got annoyed by the promotions. I also got yelled at by my husband when I missed big promotions. I am already a participant in this activity. What I need to do is actively “observe” and take notes.

  • Environment settings:

Morning in the car. Started the engine and waiting for it to warm up.

Between meetings. Walking in the hallway.

During meetings. Somebody is talking something that’s not very relevant but could becomes relevant any minute.

Lunch time. Have 10–20 minutes to catch up on the personal matters.

Coffee time with teammates. Don’t want to miss important work emails

Bed time. Make sure I didn’t miss anything important.

  • Concentrate on collecting data including but not limited to:

The context of the task

What is my major purpose?

What do I need to achieve the purpose?

What specific tasks I do? What are the subtasks?

What’s my thoughts and complaint?

What hacks I used to help achieve the goal?

  • Be aware of the bias:

I am just one of the users I am going to design for. I may not be representative.

Let the observation shape the design. Don’t fall into the confirmation bias trap.

Needfinding plan 2: interviews

Figure 2: conducting user interviews

It is very well established that “I am not my user”. One of the best ways for me to gather the data to help form my design is to directly talk to my users. The second need-finding exercise I plan to do is user interview (Figure 2).

I would like to conduct 2–3 rounds of user interviews, with the first round focusing mostly on high-level questions and the following round on relatively specific questions. I plan to recruit 5–6 people for each round of the user interview. Each of these interviews with taking 30–40 minutes.

In Figure 2, I provided a flow chart of how I am going to conduct the multiple rounds of user interviews (Figure 2A). I also drafted a sample script for the first round interview based on Six W’s (Figure 2B). These questions are relatively general. I would like to go into the interview process with a broad view to minimize bias and keep open to opportunities for new findings. Figure 2B is just the very first cut of the script. Before I conduct the real user interview, I will consult the professional user research experts (I, fortunately, have a few friends in this domain) and practice the scripts with my friends (Figure 2A). This will help me make sure I will have a productive and ethical interaction with my users. I will also re-exam my user profiles (Figure 1A) and decide whether I need to fine-tune it for this specific interview. Upon finalization of the script and interviewee list, I will conduct the first round interview. I plan to take notes during the interview. I also would like to make video recordings if possible, so I can revisit and observe details such as users' facial expressions and body language. After summarizing the result and reflecting on it, I would like to form the script for the following round of interviews. I will follow a similar workflow for all the rounds of interviews (Figure 2A).

Other things I want to bear in mind:

Dress casually. Similar to the users.

Make sure I explain the purpose of the interview and how the data collected will be used.

Make the interviewee feel comfortable.

Allow some tangent lines for interesting discovery. But also be conscious about keeping on track.

Take notes, but don’t be just a notetaker. Eye contact!

Let the interviewee do most of the talking.

Make sure to thank the interviewee at the end. Wholeheartedly!

Consider contextual interview after the first round interview.

Needfinding plan 3: think aloud and post-event protocols

At one point, simply asking users to talk about “checking emails” is not going to be enough. I would like to observe them doing the tasks and ask them to think aloud.

Think aloud is one of the most popular, if not THE most popular usability testing methods. It is easy to conduct, robust (poor facilitator proof), and convincing to stakeholders (in this case, my teacher, my TA, and classmates). It also has its limitations though. For example, talking aloud while doing the task is not the natural setting for the users, so it may deviate from the “reality”. It is a brain dump instead of carefully reflected thoughts. Interruption from the facilitator can also bias users and deviate them from natural behavior.

To compensate for the downside of think aloud, I also would like to utilize post-event protocols. I plan to conduct an interview right after the user finishes the tasks. This way, the experience is still fresh in their mind, but they won’t be too disturbed by my study.

I plan to follow the below steps:

  • Recruit representative users. I plan to have 10 users for a think aloud and 5 out of these 10 users for post-event protocol.
  • Ask them to check emails with different contexts. To think aloud, I will ask them to audio record their thoughts. For post-event protocols, I will conduct 20 minutes interview as soon as they finish the task while the circumstance allows.

In the car, check email for 2 minutes (both think aloud and post-event protocols).

Walking in the hallway (both think aloud and post-event protocols).

During meetings. This is a difficult setup for a think aloud. For the users’ convenience, I will just conduct the post-event protocols after their meeting.

Lunchtime (both think aloud and post-event protocols).

Coffee time with teammates (both think aloud and post-event protocols). In this case, I will be the teammate.

Bedtime. Think aloud is more realistic in this setup.

  • While thinking aloud, I will let users do a brain dump to talk about whatever comes into their minds. While conducting the post-event protocols, I will let the users lead the talk, but I will also ask them to talk about their purpose, the tasks they try to achieve, what they needed to achieve the tasks, and the pain points.
  • I will collect all the data and perform my analysis to help inspire my design to solve the true user problem.


In the above sessions, I described my 3 plans for need-finding. I believe they will give me a good starting point to design a better “all inboxes” interface to truly bring users closer to the task and make the user interface less visible. These 3 plans have a natural flow according to the sequence I presented them. I briefly explained the logic of the flow. But they don’t have to be executed strictly sequentially. A large part of them can be paralleled to save time, although some level of sequence would provide insights from one to the other.

Although I proposed only 3 plans, I will not limit myself to just these methods. Other methods such as surveys can be also very helpful. I plan to use them when necessary. For example, a survey is good for quantitative measures. I currently intend to use it once I have a low or intermediate fidelity prototype, so I can gather a lot of feedback on different designs.