Case study: Yirla 💚 Uxia

What does Yirla do, and who were we testing it for?

Yirla helps marketers make sense of paid advertising across multiple channels. A marketer connects their advertising accounts so Yirla can bring campaign data together, identify problems and recommend actions such as changing creative or reallocating budget. Its reports and alerts are meant to help the team decide what needs attention without manually pulling everything together from separate dashboards and spreadsheets.

For this study, we focused on an audience discussed with Scott Schnaars, Yirla’s founder: an in-house demand-generation lead in a small, under-resourced marketing team. They manage several paid channels, prepare reports for management and need to explain where the next dollar of advertising spend should go.

Yirla’s homepage, captured for this article on 8 October 2026.

In my original review, the “Your 24/7 Paid Media Analyst” value proposition copy already gave visitors a useful idea of the product. “No automated changes. Ever” answered a practical concern: would the software change campaigns or spend budget without asking?

I was more concerned about the claims and next steps. Results such as improved ROAS needed enough context for a buyer to assess them: which customer, what baseline, over what period? I also found identically labeled trial buttons leading to different destinations.

My recommendation was to keep the clear product description and improve the evidence and buying path. Uxia gave us a way to examine what happened next without recruiting human participants for this first pass: could someone register, understand the dashboard and see enough value to connect their advertising accounts?

What did Uxia add to the original teardown?

The teardown combined my inspection of the desktop and mobile experience with 18 synthetic copy responses. Through Expected Parrot, I ran nine relevant simulated buyer situations through two AI models to check comprehension and reactions to the claims. The models read the copy; I inspected the rendered pages and links.

With Uxia, the simulated participants could work through browser tasks. Alongside their answers, we had the sequence of screens and actions that led to them. That made it possible to investigate their actual feedback rather than take it at face value: where did the participant go, what did they see, and what happened when they tried to continue?

For these AI journeys, Uxia provides step-by-step screenshots and transcripts rather than a continuous video recording. As Borja Díaz-Roig from Uxia explained, the agents work through screenshots of the interface. You follow the screens they encountered alongside their actions and comments. This study used synthetic participants only. Uxia also supports studies with human participants in the same platform, so a team can follow up on the questions raised here with real prospects.

I supplied the audience context and study objectives. Borja’s team configured the study using its multi-task functionality, which was in beta when they delivered it in September. We were extending the earlier review into the product, with particular attention to what a relevant prospect could understand before connecting campaign data.

How do you set up the journey you want to test?

When you open a study in Uxia, the navigation at the top takes you through its setup and results. Build is where you inspect the tasks: what the participant should do, the situation they’re in and when the task should end.

We gave the participants a budget-review scenario. They worked in a small marketing team, managed campaigns across several channels and needed to make better spending decisions. This gave them a reason to explore Yirla beyond simply clicking around a new app.

Uxia’s Build view for the website task. The scenario asks a small marketing team to prepare for a budget review; the stop condition is understanding the product.

The study had three tasks and two follow-up surveys:

1. Explore Yirla’s website. Gather enough information to explain who the product is for and what problem it solves, then answer those two questions.

2. Sign up. Create an account and reach the dashboard, using the skip options instead of connecting advertising or other data sources.

3. Explore the product. Follow the product tour and visit the main sections, then answer questions about expectations, what works without connected data and what to do first.

We chose the unconnected route because the study wasn’t using a marketer’s live advertising accounts. The AI participants had no authorized campaign data to supply. They could register, explore the interface and explain what they understood; evaluating Yirla’s recommendations against real campaign performance would need a separate setup.

That choice also defined the question we could investigate: what can a prospective customer learn before granting access to their data? Yirla’s campaign reports and alerts need those account connections to show meaningful results. We wanted to see how the product explained that requirement and what it offered someone still deciding whether to proceed.

Checking these instructions first helps you work out what conclusions to draw from the results. Here, looking for a skip button was part of the assigned task.

How do you choose relevant synthetic participants?

The next section, Recruit, shows the audience settings and the individual AI persona cards. This is where I’d check whether the study is examining the people and situations I actually care about.

Our audience context came from the conversation with Scott: reporting across channels, communicating results to management and deciding how to allocate budget. Those details made the exercise more useful than asking a generic AI user to critique the site. They gave the participants a relevant job to attempt and a reason to care about the outputs.

Recruit shows the broader audience filters above individual AI persona cards. James and Jamie are shown here; set demographics and inferred details are labeled separately. These are synthetic profiles, not customer interviews.

The selected filters covered full-time marketing professionals aged 27–45 in demand-generation, digital-marketing and marketing-manager roles. The five generated cards were narrower: all Marketing Managers, aged 30–43, with advanced tech literacy.

Uxia labels which profile details were set and which it inferred. This study didn’t use custom enrichment with additional customer research. I would keep that in mind when planning a follow-up: the profiles fit our intended working situation, but we still need to hear from real prospects, including people with less technical confidence.

It’s worth looking at both the filters and the cards. The filters describe the audience you requested; the cards show the perspectives used in this particular run.

Where do you start when the results are ready?

Report brings the study together by task, with the main findings and supporting responses. It’s a good first read before opening the individual results under Analyze.

Uxia’s report brings the study together and states that no human participants completed it. Its audience paragraph describes the broader filters, not the narrower actual profiles. The displayed 8 October date is the report display date; the study was delivered in September.

The website findings were encouraging. All five participants described Yirla as bringing campaign data together and turning it into prioritized actions. That was consistent with my earlier assessment of the product description.

Their questions were more specific. Could they inspect the reasoning behind a recommendation? Could they explain it to their manager? Would the reports fit the work they needed to do?

One finding, “Decision outputs are difficult to assess or defend from the examples shown,” is a useful way to demonstrate how I’d investigate an issue. In Analyze, select the website task, open Insights, then open that finding. Inside it are three tabs: Impact, Evidence and Fix.

Open a finding to move between its proposed impact, supporting evidence and suggested fix. Impact is selected here.

Impact: why should I care about this issue?

Impact explains why the issue may matter to someone trying to complete the task. It helps you decide which findings deserve a closer look.

For Yirla, the concern was straightforward: if the software recommends moving advertising budget, the marketer needs to know why. Otherwise, how do they decide whether to follow the recommendation or explain it to their manager?

That could make evaluation harder. I’d then turn to the next tab to see what prompted that interpretation.

Evidence: what did the participants actually do or say?

Evidence connects the finding to the AI participants’ responses. Thomas P., one of the synthetic personas, said:

“I still need to understand how defensible the expected impact is, because I cannot take a recommendation to stakeholders without knowing what the estimate is based on.”

Thomas P., AI participant (synthetic persona)

That helped me separate two questions we might otherwise put under “add more proof.”

First: has Yirla helped a similar business get a useful result? A customer story with a baseline, timeframe and measured outcome could answer that.

Second: why is Yirla making this particular recommendation? If it suggests moving budget from one campaign to another, show the campaign signals behind that suggestion and explain how they led to it.

The marketer may need both before they can comfortably recommend the product to their team.

Fix: how could we address the issue?

Fix turns the finding into proposed changes and acceptance criteria: specific things the revised page or feature should do. It gives the person making the change something more useful than a broad instruction such as “improve trust.”

For this Yirla finding, the suggestions include:

• Put the supporting metrics beside an expected-impact estimate.

• Show a complete assistant question and answer, including its evidence and recommended next action.

• Explain which signals contribute to competitor scores.

• Support revenue or pipeline claims with examples that show those business outcomes.

The decision-output finding with Fix selected. Its acceptance criteria specify what an example should contain. These are proposed changes, not implemented fixes or measured improvements.

I would start with one recommendation already shown on the website. Add the context and supporting information a marketer would need to assess it, then ask someone in the intended audience to talk through what they understand and what they still need to know.

That gives us a concrete piece of work to discuss with Scott and his team. We can also check whether the suggested explanation matches what the product actually does before adding it to the page.

What happened when the participants tried to sign up?

For the second task, the participants clicked the trial link, switched from the default Login tab to Sign Up, registered and went through onboarding. They used the skip controls to continue without linking advertising accounts, as instructed.

The Login default is an obvious improvement opportunity. Someone clicking “Start Free Trial” shouldn’t have to switch to the registration form themselves. All five managed the extra step, but I would still remove it.

The connection screens deserve a closer look. They emphasized syncing platforms, while the option to explore without a connection was less prominent. On one screen, the participants used the smaller “Skip This Step” link to get past a disabled-looking Continue button.

Sophie’s signup journey: Yirla’s connection screen emphasizes linking ad platforms, with “Skip This Step” at the bottom right.

Under the signup task’s Insights, the finding about exploring without connections includes a Fix recommendation to make that route easier to see, make it clear that visitors can explore before connecting an account and explain that access requires authorization. That would help a cautious prospect understand their choices before committing.

The Transcripts view lets you follow an individual participant’s actions and commentary. James, one of the AI personas, left the registration form to read the Privacy Policy, then returned and finished signing up.

The Privacy Policy screen from James’s signup journey. His transcript records the detour from registration and his return to the form.

That’s a useful place to consider a short explanation of how signup information is used, with a link to the full policy. It puts the answer beside the decision instead of making someone leave the form to find it.

Could the participants see enough of the product to want to connect their data?

Inside Yirla, the participants could find reporting categories such as Campaigns, Creatives and Audiences. These are where a marketer would expect to examine performance once their advertising data is connected.

In this study, those screens were empty. The participants generally understood why, but they couldn’t inspect a representative report or recommendation to judge whether it would be useful.

Yirla’s Creatives report in Thomas’s recorded Uxia journey, before connecting campaign data.

Thomas’s synthetic transcript, displayed step 6: he moves from Campaigns toward Creatives and explains why the empty report is difficult to evaluate. This is an AI participant’s recorded action and commentary, not a customer quotation.

Sophie S., another AI persona, described the experience in her follow-up answer:

“That made the product safe to explore, but difficult to properly evaluate.”

Sophie S., AI participant (synthetic persona)

In the platform task’s Insights, Uxia recommends sample reports and alerts. The Fix criteria ask for representative previews across the reporting sections, clearly labeled as sample data, with the route to connecting accounts still visible.

I like that as a direction to test. Before authorizing access, a marketer could inspect the format of a report, the detail behind a recommendation and whether the output would help them prepare for a team meeting. Then they could connect an account to evaluate it using their own campaigns.

Did the product tour help them work out what to do?

Yirla’s product tour introduced the location of Feed, Settings and Help. Some participants wanted more help understanding what useful work to do in the product, including the other reporting and analysis sections they could see in the navigation.

A product-tour step from Thomas’s recorded journey: Feed is introduced as an overview of connected advertising platforms. The study found that participants wanted more guidance on practical tasks.

Ask Yirla is Yirla’s built-in conversational assistant. You can open it from the product’s left sidebar and ask questions about campaign performance. Its starting screen groups suggested questions into ad spend and budget, creative fatigue, audiences, competitors, and campaign diagnostics.

Ask Yirla in Jamie’s product journey. The prompt “How does my CPL compare across LinkedIn, Google & Meta?” gives a concrete example of what a marketer could ask.

Those suggested questions offered a useful contrast to the product tour. They gave participants examples of the questions the assistant could help answer. Jamie, one of the AI personas, tried a comparison question and reported receiving “No Connected Accounts Found.” That showed both the intended use and the dependency on campaign data.

I would bring that same practical context into the tour: here’s a task you might want to complete, here’s where to do it, and here’s what you need to connect first.

Another finding concerned File Vault, a section for stored files and marketing context. It displayed a card labeled “Yirla File Vault Guide.pdf,” which appeared to offer guidance on using the feature. Four participants selected the card without opening the document. I’d reproduce those clicks and check the card’s behavior, including whether the download icon works separately. A guide is only helpful if people can open it.

How can you investigate a finding beyond the summary?

In Uxia, I would use three views together: full survey answers, Transcripts and Journeys.

Full answers preserve useful differences. James and Jamie proposed connecting an account next; Sophie wanted sample outputs first; Alex wanted to inspect Feed and Analytics before deciding. Those differences suggest questions to explore with real prospects: who needs to see a sample output first, and who is ready to connect an account?

Transcripts show the actions and associated synthetic commentary. Journeys lets you inspect the sequence visually and open individual frames. Together, they help you check what the participant actually encountered before raising an issue.

In Thomas’s product journey, for example, you can follow his movement through Campaigns, Creatives, Audiences and other reporting areas. The Journey preview also has a Heatmap mode.

Uxia’s Journey preview in Heatmap mode, Thomas’s frame 7 of 14. The underlying screen is Yirla’s unconnected Analytics–Creatives view. The colored overlay belongs to this synthetic trace; it is not human gaze, eye tracking or a customer click distribution. No quantitative color legend is shown. Account details are redacted.

The overlay belongs to the simulated journey. I’d use it alongside the screen and transcript to inspect that step.

This is especially useful when a finding concerns something the participant couldn’t find. The report notes that an Integrations destination was visible, but some participants didn’t visit it. I would check where they looked and whether the page they were reading could have pointed them toward the answer.

What does the success rate tell you?

Uxia’s Analytics tab summarizes task completion and steps. All three tasks showed 100% success among the five participants.

Uxia Analytics—not Yirla’s own analytics—shows completion of the assigned exploration task by five AI participants. Success here is not a purchase or account connection. UOS is Uxia’s displayed score, not a human usability result; a dash is not a measured zero.

To interpret that, go back to the task endpoints. The participants successfully registered and explored the platform. The unresolved questions about report quality came up during that successful exploration. That’s why I’d read the completion figures together with the comments and journeys before choosing what to improve.

How I’d use this study with Scott

I’d work through it with a short list of decisions to make:

1. Confirm the setup. In Build and Recruit, check the tasks, budget-review scenario and actual profiles so we agree what the study covers.

2. Preserve what’s clear. Use the Report and full survey answers to see what the participants understood about Yirla. Keep the paid-media analyst description unless stronger evidence gives us a reason to change it.

3. Choose one proof improvement. Open the website decision-output finding, compare Impact with Evidence, and use Fix to decide what information to add to one recommendation example.

4. Check the signup and guide interactions. Follow the transcripts and journeys, reproduce the extra steps or failed clicks, and fix what the team confirms.

5. Try a more useful first visit. Explore clearly labeled sample reports, an obvious optional connection route and a product tour built around practical tasks.

6. Take the important questions to real users. Can intended customers understand the recommendations, use them in their work and get useful results after connecting their campaign data? We could run that follow-up in Uxia with our own participants or through its partner panels, provided we can recruit marketers who fit Yirla’s audience.

Our original teardown gave us reasons to look more closely at proof and the trial path. Uxia helped us follow those questions into the product and gave us specific screens, comments and suggested fixes to discuss.

When would I use Uxia again?

I’d use Uxia again after changes to the signup flow, recommendation examples or product tour, to inspect how the revised journey works and identify questions worth taking to real prospects. It’s also useful when a team needs a first look while arranging access to customers. Yirla’s redesign would be a natural moment for that next pass.

For Yirla, I’d pair that next pass with conversations about the decisions this study couldn’t settle: is the report useful for your work, do you trust the recommendation, and what would you need before connecting your accounts?

Explore the full Yirla study in Uxia to see the findings, participant responses, and recommended fixes for yourself.