There are shiny AI features everywhere, but building AI experiences that people actually trust and rely on? That's where the real challenge lies. At monday.com, we've learned that AI success isn't measured by how exciting features are, but by how confidently people use them to move their work forward.
In this session, Senior UX Researcher Shai Passal shares the ongoing journey she and her team have taken to understand and measure what trust means in AI products, putting humans at the center of AI development. Through research insights and real-world examples, she'll explore how focusing on trust transforms the way teams think about and build AI experiences.
Watch this WaysConf session, then continue with related talks or explore the current programme.
At WaysConf 2025, Shai Passal, Senior UX Researcher at monday.com, shared how her team moved beyond initial launch hype to systematically understand, measure, and design for user trust in AI. When monday.com first released AI tools for work management tasks like categorizing feedback, high engagement metrics masked a deeper issue: anxious users were only testing features in self-made sandbox environments out of fear of breaking real data. To address this, Passal's team combined qualitative contextual inquiries, in-context surveys measuring reliability and control, and behavioral usage data. By identifying these friction points, monday.com introduced low-risk preview environments ("try before you apply") and transparent explanations that clarify why the AI made a specific decision. Ultimately, this research transformed the team's culture, establishing trust as a core product metric and shifting focus from pure technical capability to user confidence.
High engagement metrics can be highly misleading if users are only testing AI features in fake sandbox environments due to a fear of breaking real work.
Watch from 5:24Unlike traditional deterministic software, AI requires a dedicated trust strategy because it makes decisions, hallucinates, and produces non-deterministic outputs.
Watch from 8:00Measuring trust requires combining qualitative contextual inquiries to observe subtle hesitations, in-context surveys to capture immediate sentiment, and behavioral data as proxies.
Watch from 9:48Providing a "try before you apply" preview allows users to safely iterate, edit, and verify AI outputs without risking their actual work.
Watch from 16:56Users do not need complex algorithmic explanations; they need simple, contextual reasons explaining why an AI made a specific decision to feel in control.
Watch from 18:35These concise notes were edited from automatic captions and checked against the talk structure. They are not a verbatim transcript.
When monday.com first launched its AI features, the initial quantitative metrics looked incredibly promising, leading the team to celebrate a successful release. However, during follow-up qualitative interviews, a troubling pattern emerged. Users repeatedly expressed fear of breaking their actual work and admitted to setting up fake sandbox environments just to test the AI. This meant the high engagement data the team celebrated did not represent real, productive adoption, but rather hesitant experimentation by users who felt anxious and out of control.
Watch from 4:50Building trust for AI is fundamentally different from traditional software. With standard applications, clicking a button yields a predictable, deterministic result. AI, however, acts as an agent making decisions on behalf of the user. It is highly unpredictable, prone to convincing hallucinations, and non-deterministic, meaning the same input can yield different outputs. Because the underlying technology also changes on a weekly basis, users struggle to build a stable mental model, making active trust-building essential.
Watch from 7:53To understand user trust, monday.com relied heavily on contextual inquiries where researchers observed users interacting with AI features in real time. This observational approach is crucial because trust signals are often incredibly subtle, such as a brief hesitation or a quick double-check. Users rarely mention these micro-behaviors in standard interviews because they seem too insignificant to report. Observing them directly allows researchers to identify friction points and ask targeted follow-up questions.
Watch from 10:02Rather than trying to force users out of their testing habits, the product team embraced this behavior by designing a low-risk preview environment. This "try before you apply" feature allows users to see exactly what changes the AI proposes on their actual work management boards without applying them permanently. Users can safely review, edit, and refine the AI's suggestions in a sandbox-like preview, keeping the human in the loop and giving them complete control before committing to any changes.
Watch from 16:07Research showed that users consistently wanted to know why the AI generated a specific result, regardless of whether the output looked correct. Instead of complex algorithmic explanations, they needed simple, straightforward reasoning. Monday.com addressed this by adding hover-over explanations. For instance, if the AI categorizes a piece of customer feedback as a feature request, a simple tooltip explains that the user asked for a filtering option we do not currently have, allowing the user to easily verify the logic.
Watch from 18:05One of the most significant outcomes of this initiative was the transformation of the AI team's internal culture. Once trust was defined and measured through concrete metrics, it ceased to be a vague, abstract concept and became a core KPI for the engineering and product teams. The team began prioritizing dedicated initiatives focused solely on earning user trust rather than just shipping new technical capabilities, turning trust into a shared, daily conversation across all disciplines.
Watch from 20:34
