Introduction: Using AI as a Thought Partner in Product Design

So we are part of the product design team of Talental EMS here at the Pygnosis with Maria and today we will show you how we use AI as a thought partner for our day -to -day tasks.

The Persona Adoption Problem

So a couple of months ago I created our very first user personas and I quickly ran into a problem because great personas personas cannot help if no one uses them actually.

What a Persona Means in This Context

And so a quick note on what I mean persona in this work. It is a compressed story of a real cluster of users with shared context, behavior, needs and constraints. It's grounded in quantitative signals such as usage patterns and events and qualitative

inputs, like support tickets, calls, past research, and ultimately it is a shared language that helps our teams align on who our users are, so who we build for. But, you know, like any other tool, adoption takes time, but people need personas when, you know, in the moment that they have to make decisions.

From Static Personas to Workflow-Friendly Tools

And so We grounded our personas in those roles. And so I asked, how could we make our user personas usable in day -to -day work immediately?

Because we had solid personas, great analysis, but they were living in Figma, easy to access, but they were still passive. And so we needed them to show up inside our workflow immediately, basically.

Turning Personas into Custom GPTs

And that's where the custom GPTs came in. So to build these GPTs, I first referenced the great work of Yannis from Yodek. He also showcased at a previous Mindstone meetup how they brought into life their marketing personas through custom GPTs, so it was a great starting point for me.

Framework: Role, Goals, Constraints, and Reasoning

1And then I followed Mindstone's framework, like who the GPT is simulating, what it's optimized to help, their world constraints and stakes, mistakes, how to reason, like clarifying the scenario, evaluating against goals, stress testing failure modes.

Guardrails: Traceable Outputs and Reduced Hallucinations

1And I added guardrails to reduce hallucinations and keep outputs traceable, so each GPT has a clear purpose and voice.

And of course I wrote the instructions using only what we had evidence for, so the model doesn't fill in the blanks with false assumptions and then I realized the real power as our personas evolved with new data so the GPTs evolved too.

Keeping Personas Current as Data Evolves

For example for our learner -accused personas I added a new lens let's say the accessibility lens, so that inclusive design is considered deadly.

And so now when we ask, for example, me at GPT about the flow, it doesn't feel like we are talking to a generic chatbot. It feels like we are consulting a strategic assistant that stays in character and translates translates personas' needs into concrete product and UX implications, and I'm really proud because

our team really likes to think of its GPT as a calculator for empathy, like you feed it a flow or a feature, and it runs it through Mia's or Sandra's perspective in a consistent way. And so today we have 10 persona GPTs, one per persona, plus three combined GPTs for admin instructors and learners.

What Changed for the Team Day to Day

And this is where these GPTs made a really huge difference for us. Instead of our user personas living in static documents, the GPTs turned them into something the team could use immediately.

So since day one, we are using the GPTs as an extra lens to stress test ideas or flows and they've been really helpful for us for asking better questions and make edge cases and and trade -offs more visible at an earlier stage.

And of course they don't replace research for any other activity. They just help us ask better questions earlier.

Hands-On Demo: Designing with Persona GPTs

And so now Maria will show a hands -on example of how a designer uses our GPTs as a teammates. Thank you.

Just give me a second. Okay, perfect. Perfect.

So together we'll see how we really use those chat CPTs that Christina made for us. Ah, I hear my voice. Okay, so let's get into it.

So I am Maria. I introduced myself earlier. I'm a

Scenario Setup: Building an Admin-Focused Chatbot

product designer here in Epignosis and we'll make a hypothesis for this example. Let's say that we are requested to build a chatbot.

Everybody more or less knows what a chatbot is. It's the virtual assistants that usually help us when we have a question or to perform some specific tasks that we have in mind.

Usually they don't work that well and we end up calling for the representative, but let's say that we will build a good one. Okay, so what do we build?

The Design Process (From Problem to Final UI)

we build a chatbot as we said what this chatbot will do will help users perform some tasks but whom whom specifically we are going to help admin role so the users that have more managerial tasks to do in our platform so let's say that these are the steps that is in a a designer needs to follow in order to come to a final solution.

Usually, we have a lot of circles going around. It's not like that waterfall, but let's stick to that.

So, first, we understand the problem. Usually, for this, we are using also the PRD. The PRD is the document that has all the feature, all the requirements, actually, that the feature needs in order to be made.

Then, in the second step, we need to clarify clarify those requirements, so to really understand what we are building. The third step is to have some initial designs. Usually these are low fidelity.

Then we come all together and we iterate on them. We gather feedback and then we have the refined solution, the final designs.

For this example, we are going to see how personas will help us in the first and into the third step.

Step 1: Clarifying Requirements with a Persona Lens

so regarding the clarification what do we have we have the prd what do we do we ask persona some question in order to get some feedback so this is the prd actually we generate it through chat gpt it's not that heavy text usually it's a long long text to be read so as a designer i may have some

questions reading through it for example what is this really about or what does assist actually mean for a user or have we missed any edge case so memory again with the prd and i go to the persona and we have a chat what i need to do is to translate those questions that i have into my head into prompts that persona is easier for it to actually give a feedback for so

Example Feedback: Edge Cases, Trust, and Risk Visibility

for example have we've missed any edge cases it can be translated into in what situation would this assistant fail you let's see what actually a personas respond with they said that copilot must guide not just answer so if we have the flow that requires a lot of steps, we need to have a step -by -step procedure, not to give

to the user a ton of text to read. The second one is that trust requires context. We cannot give the same instruction to different users, it really needs to be differentiated

based on what roles they have into our platform in order to know for us that they can perform those actions that we suggest. And the third one is that the risk must be visible if we're going to suggest some actions that they have major They can change major things in our platform that needs to be To be outlined

Step 3: Stress-Testing Early Designs with Personas

So let's go to the third step how personas will help us during design and And what do we have? We have a design.

For the design, since this is a hypothetical scenario, we have created a lovable design. Lovable is a UI helper through AI. And again, we will ask some questions. You will see now actually how we are going to do it through ChatGPT. And then we get the feedback.

So, this is the design. On the bottom part, there is the chat but as we see we have some common tasks that are suggested from us that users may feel stuck with. Let's go with the help of an SSO and what we see here is that we have split the task into five steps and we have also the caution message that Persona also remarked before. Perfect, we can go to the next slide.

again as a designer we have a lot of questions so we need feedback also for the stage of the workflow and now for example does it actually save time that I have a question in my head could be transformed into would you this or still go to support you will actually use it or you would still go to support in order to solve this what are we going to see next is how we are truly using the

personas so now i am on the chat gpt on the left part of the screen we see the personas that christina made for us so since we are referring to the admin we go to the admin lens which is sandra and carmen and here i have created a prompt with the questions that i had and i also attach touch the screens and the video of lovable. I'm not that quick.

So what happens is that Sandra and Carmen will give quite a lot of feedback, which is really useful for me. We'll see that in a short. We are not going to, first, of course, we take a deep breath, as mindset taught us to, in order to get better results. And here is Sandra, and next is Carmen.

As I said, we're not going to read all through that, but we'll share with you what actually they suggested. They suggested to offer more tasks when you first open the chat GPT, but those tasks to be structured by categories in order for the user not to feel overwhelmed.

The second one is to turn steps into interactive states. We'll see what do we mean by that. but let's say that we need a more interactive stepper and the third one is that if someone follows instruction and needs help in a point then they need

Design Iterations Driven by Persona Feedback

to be able to ask for it so the final result as we see here we have more options for the user to choose from but we have structured them by categories categories and we are going to do the same flow, help me set up an SSO.

And what happens here is that every step has been split into a different element. So I can, of course, it has a link if it needs to get me somewhere. And what I can do is I can mark it as done. And if I have a warning message, I can see that as well.

And if I need help, I can do that too. too.

Conclusion: Personas as Fast, Consistent User Representation

So how personas helped me through the day is that they reveal what we forgot. So we made spot any blind spots we had.

And then they simplify decision if I am between A or between B, they can offer some suggestions and be able to say what A offers regarding or what B offers.

And the third one is is that test if users actually trust it.

And if it was to say one thing, it would be that personas don't replace users, but design moves fast, and we need to make a lot of decisions really quickly.

So what personas do is that keep the user in the conversation even when we don't have the time to reach them. Thank you.