Institutional Trust and the Hidden Failure of Enterprise AI Adoption

概览

This episode of Data Science with Sam argues that enterprise AI adoption often fails because organizations confuse tool deployment with real behavior change. The central claim is that the missing question is not whether the data, budget, or model is ready, but whether users trust the institution enough to change how they work.

Guest Sumayya Shravani connects AI adoption to institutional trust, cognitive diversity, and the experience of people who have had to learn systems from the outside. She argues that immigrants, career switchers, and others with outsider experience often notice adoption barriers that insiders treat as invisible.

The conversation moves from diagnosis to practical measurement. Instead of relying only on dashboards, training completion, or accuracy metrics, Sumayya recommends asking frontline users what is actually hard about their work, tracking whether users overwrite AI outputs, and studying the people who quietly stop using the tool.

分段落总结

[00:04] AI Adoption Gap and Guest Introduction

[事实] The host says McKinsey reports that 88% of organizations are using AI, but only about 6% are getting value from it.

[事实] The episode frames the gap as a problem caused by treating tool deployment as adoption.

[事实] Sumayya Shravani is introduced as a technical program manager and data infrastructure lead at the University of Colorado Denver, managing 200 data pipelines across a 530-user analytics environment.

[事实] She is also described as the builder of Data Ready, a four-agent system for auditing AI readiness in data environments.

[01:53] Institutional Trust as the Missing Adoption Question

[事实] Sumayya argues that enterprises usually ask whether they can afford the tool, whether the data is ready, and whether leadership will sponsor it.

[事实] She says the more predictive question is whether users trust the institution enough to change how they work.

[事实] She describes being invited to an AI rollout meeting where the scope had already been defined and she was only informed, not included in the discussion.

[推测] Her example suggests that AI rollout teams can overlook affected users without intending harm, because decision-makers may assume institutional trust is already present.

[03:52] Learning Systems from the Outside

[事实] Sumayya says she came to the U.S. from Bangladesh at 19 and learned institutional systems as an adult from the outside.

[事实] She says trust was never free for her and had to be earned with each system she entered.

[事实] She recalls early corporate experiences where she avoided asking questions because doing so felt like it might mark her as not belonging.

[推测] This outsider experience becomes a lens for noticing when AI users may silently avoid asking basic questions about a new tool.

[06:12] Deployment Is Not Adoption

[事实] Sumayya says organizations often deploy a tool and call it adoption because deployment is visible to the people writing scorecards.

[事实] She compares this to measuring someone’s gym membership and calling it their fitness.

[事实] She argues that outsiders may build AI systems, while adoption strategies are often written by insiders who have never had to learn the institution from scratch.

[事实] She identifies “silent departure” as an important signal: users try the tool once, decide it is not for them, and never return.

[推测] The episode treats non-use as valuable adoption data rather than laziness or lack of enthusiasm.

[09:34] Resistance as Honest Data

[事实] Sumayya says she built Data Ready to check semantic layers, documentation, model quality, and governance posture before copilot testing.

[事实] She later realized the tool measured whether data was ready but missed whether people were ready.

[事实] She says employee resistance should prompt the question: what did users see that the implementation team did not?

[事实] She gives three reasons people may not use an AI tool: they do not know it exists, it does not fit their workflow, or they do not trust the institution enough to change their workflow.

[推测] Her framework implies that training mainly addresses awareness, while workflow fit and institutional trust require deeper organizational work.

[13:15] Cognitive Diversity Before the Plan Is Finished

[事实] Sumayya says many corporate cognitive diversity efforts are theater, such as panels, listening tours, or reviews after the plan is already done.

[事实] She says this is like QA on a product that has already shipped.

[事实] She argues that people who learned systems from the outside should help define what “working” means before success metrics are set.

[事实] She cites a 2017 Harvard Business Review study saying teams with higher cognitive diversity solved complex problems significantly faster.

[推测] The practical value of cognitive diversity in her argument is not symbolic representation alone, but reducing shared blind spots in AI adoption strategy.

[17:04] Outsider Experience as Diagnostic Skill

[事实] Sumayya says immigrants and career switchers learn unwritten rules, real power structures, and gaps between stated process and actual process.

[事实] She describes this learning as expensive because it can involve embarrassment, uncertainty, and feeling like one does not belong.

[事实] She says this experience trains people to ask who actually decides something, not just who is supposed to decide it.

[事实] She includes career switchers from fields such as nursing, teaching, or the military as people who may develop similar diagnostic skills.

[推测] The episode frames outsider experience as a practical capability for detecting hidden adoption barriers, not just as a diversity category.

[20:20] Concrete First Steps for Leaders

[事实] Sumayya recommends that before the next AI readiness assessment or board scorecard, leaders pick 10 people whose daily work the tool is supposed to change.

[事实] She says leaders should ask those users what is actually hard about their jobs before the tool exists.

[事实] She says leaders should ask whether existing success metrics make sense from those users’ seats.

[事实] She recommends tracking overwrite rate: how often users receive an AI suggestion but go back to the old way of working.

[推测] Her advice shifts adoption measurement from implementation activity to observed workflow change.

[22:54] Measuring Success Through Quiet Departures

[事实] Sumayya says leaders should ask who stopped using the tool, not only who complained.

[事实] She says complaining means people still care, while quiet departure is a real readiness signal.

[事实] She compares this to employee attrition: companies often do exit interviews when someone quits, but not when someone stops using an AI tool.

[事实] She argues that if the people thriving with the tool are only the designers or people similar to them, the organization has built a club rather than broad AI adoption.

[推测] The central success measure becomes whether the tool works for people outside the design group’s assumptions.

[26:03] Closing and Further Contact

[事实] Sumayya says listeners can connect with her on LinkedIn.

[事实] She mentions Ground Truth, a weekly newsletter where she does AI tool teardowns in real enterprise workflows.

[事实] The host says he will share Sumayya’s LinkedIn profile, Substack articles, and Data Ready platform in the show notes.

播客点评/总结

This episode’s strongest value is its reframing of AI adoption as an institutional and human trust problem rather than a purely technical rollout problem. The discussion is especially useful because it gives concrete signals to examine, such as silent departures, overwrite rates, and whether frontline users recognize the stated success metrics.

A major highlight is Sumayya’s use of personal experience to explain why outsiders often notice barriers insiders miss. The episode avoids treating diversity only as a moral or symbolic issue and instead connects lived experience to specific diagnostic skills that matter in enterprise AI.

The main limitation is that the conversation is conceptual and example-driven rather than based on a detailed case study with measured before-and-after outcomes. [推测] Listeners looking for implementation templates, survey designs, or analytics instrumentation may need additional material beyond this episode.

[推测] This episode is best suited for AI leaders, data leaders, product managers, transformation teams, and practitioners responsible for enterprise AI adoption, especially those whose dashboards show activity but whose tools are not clearly changing day-to-day work.