Behavioral Drift
-
98%
of organizations have shadow AI usage within. (Verizon)
-
>45 minutes
of course content spread across 5 modules and 9 videos. Learn at your own pace!
-
3
downloadable artifacts that allow you to assess where behavioral failure modes are occurring in your organizations’ own AI adoption.
-
>97%
of previous participants in masterclasses, workshops, and executive education recommend the instructor to others.
Why this course exists.
AI programs can appear successful—high usage, completed training, satisfied leaders—while quietly failing where it matters: whether people use them correctly once no one is watching. The most dangerous breakdowns are invisible ones: employees route around approved tools (shadow AI now appears in 98% of organizations, with detections up fourfold in Verizon’s 2026 report), human reviewers start rubber‑stamping outputs due to the recognition bottleneck, and people stop escalating concerns as small deviations become the new normal, echoing Diane Vaughan’s Challenger analysis. These behaviours look different but share a single mechanism: each is a rational workaround when “doing it right” carries friction and the shortcut doesn’t trip an alarm. This course treats shadow use, performative review, and failure to escalate as forms of behavioural drift—and shows you how to diagnose which hidden failure mode is emerging on your team and design an intervention that closes the real gap rather than adding another unenforced policy.
What you’ll leave with.
-
A documented instance of shadow AI use on your own team, diagnosed against the friction/capability-gap framework.
-
A Review Reality Check—an honest audit of whether human review on one real workflow is substantive or performative.
-
A redesigned escalation path for one real decision point, with explicit triggers and a reduced cost to speaking up.
-
A Behavioral Drift Audit that combines the three into a single diagnosis-and-action plan you can run on any workflow going forward.
-
-
-
Explains why employees bypass sanctioned AI tools even when given legitimate ones, grounding the diagnosis in BJ Fogg's Behavior Model and real adoption data. Breaks the behavior into three drivers—friction, capability gaps, and distrust of the process—so viewers can name exactly which one is happening on their team.
-
Shows why banning shadow AI use backfires, then redesigns the response using choice-architecture principles from Thaler and Sunstein. Walks through three targeted fixes—faster approval paths, closing real capability gaps, and safe disclosure—matched to each driver from the diagnosis video.
-
-
-
Diagnoses how human review of AI output quietly turns into a rubber stamp, using real override-rate data and the psychology of automation bias and alert fatigue. Gives viewers three concrete questions to test whether a review process on their own team is actually functioning.
-
Redesigns oversight so it can't be faked, grounded in Kahneman, Sibony, and Sunstein's research on judgment noise and decision hygiene. Introduces three structural fixes—reversing the review order, routing by confidence, and requiring a named check—instead of just adding more approval steps.
-
-
-
Examines why people who spot a real problem often choose to stay silent, tying the behavior to asymmetric cost, pluralistic ignorance, and Diane Vaughan's normalization of deviance. Gives viewers a way to spot this pattern before it becomes the accepted norm on their team.
-
Designs an escalation path people will actually use, grounded in Amy Edmondson's research on psychological safety as a structural—not motivational—condition. Covers three interventions: lowering the personal cost of speaking up, making escalation criteria explicit, and resetting the baseline on a schedule.
-
-
-
Reveals how the three failure modes cascade into each other on a real team, and formalizes the course's recurring three-question lens into one diagnostic sequence. Applies that sequence live to a workflow the course never covered, proving it generalizes beyond the three named failure modes.
-
Brings the three artifacts built across the course together into a single, recurring audit practice rather than three separate exercises. Closes the course by returning to its opening idea—that a clean adoption dashboard was never proof of correct use—now with a concrete practice to check instead.
-
Your instructor.
Eugene Chan, PhD
Eugene (PhD Toronto; MA Chicago; AB Michigan) is a behavioral scientist whose work focuses on the moments when customers are uncertain, skeptical, or under scrutiny, translating insights from consumer psychology into practical strategies that strengthen confidence, reduce friction, and improve adoption across products, services, and communication. His academic work is published in Financial Times Top 50 journals and has been covered by media outlets include Globe and Mail, Men’s Health, and the Wall Street Journal. He has conducted workshops and masterclasses for leading companies in Canada, Australia, and the United States, and has served on tenured faculties at Tyndale University, Toronto Metropolitan University, Purdue University, Monash University, and the University of Technology Sydney.
You’ll want to enrol if you want to…
-
... diagnose which of the three hidden failure modes is occurring in a given team or workflow, rather than treating every workaround as the same problem.
-
... explain why shadow AI use persists after official tools exist, and identify the specific friction or capability gap driving it.
-
... distinguish real human review from performative review, recognizing automation bias and the recognition bottleneck before they produce a rubber stamp.
-
... identify the cost/visibility asymmetry that turns a legitimate concern into silence, and catch normalization of deviance while it’s still small.
-
... design an intervention matched to each failure mode’s actual driver, instead of defaulting to more policy or more training.
-
... apply one diagnostic sequence across all three failure modes—and any future one you encounter—to determine whether drift is happening and where to intervene.