🐴 How to strategically integrate AI into design thinking

An example walkthrough using different techniques

Last month, I gave a talk for the City of Vancouver on AI and design.

It was part of the training for designers and staff from the Service Design, Customer Experience, and User Experience teams of the city.

I shared my thought process for strategically integrating AI into design thinking.

Today, I’d like to share an example I mentioned, which covered various techniques.

Hope you find it helpful as well.

---

Overview

3-10 years ago, the design field talked a lot about human-centric design. It stressed the importance of clarifying user needs, framing the user problems, etc.

But the rapid rise of AI has shifted the focus.

Designers are now expected to learn AI, adopt AI into the workflows, and think strategically with AI.

That can feel overwhelming at times.

I think the core reason is that when it comes to adopting AI, the first thing most people associate it with is AI tools or capabilities. With countless tool options and constant updates, it can feel overwhelming. I refer to this perspective as a top-down approach.

But a top-down approach is not the only way to adopt AI. There’s also what I call a bottom-up approach, which grounds us by bringing the focus back to user problems.

By combining both perspectives—top down (AI capabilities) and bottom up (user problems)—we can gain a clearer picture of how to adopt AI in a way that is both tangible and user-centric.

Top-down and bottom-up approaches

---

The Example

I walked through an example. Imagine they are tasked to improve the customer experience of getting a driver’s license in Vancouver.

Since most of the audience works in service design, I hoped the example could resonate more with them. It was a broad/vague task, involving multiple touch points and team collaboration.

So where to even start to incorporate AI into the design thinking to strategically tackle this task?

I introduced a thought process step by step.

It’s worth mentioning that there’s no single gold standard or “must-follow” approach. This is simply one process, connecting the dots across different techniques and methodologies.

---

Step 1: Map out the user problems

Map out the flow

Mapping out a simple user journey is a simple way to make vague challenge more concrete.

It is not based on imagination, but grounded in the existing information and insights you have access to.

Mapping out a simple user journey

List the user problems

Assume your team already has existing user insights for you to synthesize.

You can write down the user problems per each step of the flow.

By the way, how you frame the problem matters. Instead of focusing on solutions or being vague, frameworks like Job-to-be-Done and User-Need-Insight can come in handy. They help frame the user problems that better facilitate idea brainstorming later.

Undesirable ways to frame a problem

More effective ways to frame a problem

For example:

• A resident is trying to [show up prepared for their visit], but [they’re unsure which documents are required]. (Job to be Done)

• Applicants need [a reliable way to check wait times and office details] because [walking in without context can waste time and increase frustration]. (User-Need-Insight)

• A first-time driver is trying to [pass the knowledge test], but [the format and prep expectations are unclear and cause anxiety]. (Job to be Done)

• Multilingual users need [access to study materials in their preferred language] because [language barriers impact comprehension and confidence]. (User-Need-Insight)

Then you map the user problems to the user journey:

Mapping the user problems to the user journey

---

Step 2: List out AI’s capabilities

Step 2-4 were inspired by Dan Saffer’s talk at Bloomberg. He mentioned a mapping methodology, which was also inspired by Design through Matchmaking by Sara Bly and Elizabeth Churchill.

Instead of hype-focusing on “tools”, we can zoom out and think from “AI’s capabilities”, then list the common AI tools to use under each category.

Mapping out AI’s capabilities with AI tools

---

Step 3: Map AI’s capabilities to user problems

In this step, list each user problem and then map a relevant AI capability to it.

This helps create a clearer picture of how AI can address concrete user challenges, leading to positive outcomes for the user, and furthermore, the organization.

A lot of interesting AI-powered ideas can be generated during this step through mapping.

Laying AI capabilities on top of user problems

By the way, if you remember Impact Mapping, this kind of mapping techniques can help us connect the dots across different dimensions during idea brainstorming.

---

Step 4: Prioritize ideas as a team

After you get a lot of ideas in step 3, it is time to converge them in step 4.

There are different prioritization frameworks that the team can use.

We are all familiar with the Impact/Effort framework on the left. But Dan brought up something intriguing in his talk, that is choosing to prioritize the “Low Risk” items first when we think of adopting AI solutions. That makes more sense when we want to start adopting AI at scale, then test and iterate.

Prioritizing ideas from different angles

Prioritization itself is meant to be a collaborate process that sometimes can feel uncertain and subjective.

I know some teams like to use a matrix to quantify every idea. You can assign weights or scores to different factors. For example, if the product is high-stakes, then the risk score carries more weight in decision-making.

That said, it’s okay to keep some subjectivity when it comes to prioritization.

---

Step 5: Leverage AI to further develop an idea

Step 5 is for a deeper dive if needed.

Let’s say if the idea the team wants to prioritize next is “the conversational assistant for the personalized checklist”, then you can leverage AI to help you further develop it.

Picking one idea as an example

Below is a simple prompting framework that I like to use.

Clarifying the ask

Writing down the necessary context

After clarifying the ask and relevant context, you can combine them together to become the prompt for AI.

Design a step-by-step plan for building a simple AI-powered conversational assistant.

The assistant helps residents prepare for their driver licensing visit by generating a personalized checklist of required documents.

We are the city’s experience team responsible for improving the driver licensing services.

The biggest service issue is that residents often show up at licensing offices without the correct documents, which leads to frustration and longer wait times for everyone.

Success will be measured by fewer unprepared visits, reduced wait times, and higher resident satisfaction scores.

We want this to launch as an MVP in 4 months and integrate with existing appointment booking and notification systems.

Then you can paste it in your favorite LLMs to generate an execution plan for you.

---

Thanks for reading.

Still dealing with jet lag from my vacation in Europe. Woke up at 4:20 this morning. So if you noticed my writing style is different today, let me know :)

See you next time.

Xinran

-

P.S.

1. If you are interested in “Vibe Designing”, check out my talk next week.

2. If you are interested in a demo on creating a full-stack app with Replit, check out the lightning lesson Junaid and I had yesterday.

3. For all the paid subscribers of Design with AI, you’ll receive a separate RSVP email tomorrow on our monthly coffee chat next Thursday. Stay tuned!