Log in or create a free Rosenverse account to watch this video.
Log in Create free account100s of community videos are available to free members. Conference talks are generally available to Gold members.
Summary
Experimentation can be intimidating to non-data science folk. But Erin wants to get everyone excited about A/B testing. In this talk, Erin shares the Conversion Design process. It centers A/B testing as a way to gather high-quality evidence to make highly informed decisions to improve your digital product. She also introduces the Good Experimental Design toolkit. These easy-to-follow templates usher teams through the logic needed to design trustworthy experiments that you can learn from.
Key Insights
-
•
Conversion design combines design, science, and business to intentionally create measurable improvements, not just arbitrary changes.
-
•
Ronald Fisher’s 1919 work exposed how poor experimental design led to decades of unreliable scientific data.
-
•
Conversion originates from the Latin word meaning to transform or change, which is broader than just sales or profit.
-
•
Traditional linear product development processes miss the complex, iterative nature of real-world systems.
-
•
Systems thinking offers a more accurate way to understand and manage design experiments within interconnected environments.
-
•
A rigorous conversion design process includes seven phases: understand, hypothesize, prioritize, create, test, analyze, and decide.
-
•
Experiments grounded in well-documented hypotheses based on research have higher success rates than guesses.
-
•
Randomized 50/50 AB testing is the gold standard for isolating the true effect of design changes by evenly distributing confounds.
-
•
Experimentation buckets—product foundations, content/motivation, accessibility/usability, and bug fixes—help prioritize work effectively.
-
•
Ethical considerations and guardrail metrics are crucial to ensure changes benefit all stakeholders sustainably and without manipulation.
Notable Quotes
"Conversion means change, not just sales or profit."
"Design is the rendering of intent — bringing ideas into form that solve the problem."
"Decades worth of agricultural experimental data was garbage because of poor experimental design."
"Most product teams stay on the bottom rungs of evidence, relying on opinions or observational data instead of randomized trials."
"You can never purely test an idea, only the implementation of the idea."
"Randomization is magic — it evenly distributes confounds so the observed effect is caused by your change."
"Not all changes create value; some do nothing or even make things worse."
"You have to think critically about how a change impacts all stakeholders, not just the main business metric."
"Ethics evolve faster than laws; just because something is legal doesn’t make it ethical."
"If an experiment I design made the front page news tomorrow, how would I feel about it?"
Or choose a question:
More Videos
"Scott Cook noticed his wife’s frustration balancing their checkbook and realized millions more probably felt the same pain."
Kaaren HansonStop Talking, Start Doing
June 9, 2017
"Researchers must respect participants' ownership of their own narratives and foster intersectional inclusivity."
Noah BondRedefining truth and inclusivity: Navigating data ownership and ethical research in the age of disinformation
March 11, 2025
"Inclusive design is about one size fits one, not one size fits all, especially in research methodologies."
Sam ProulxSUS: A System Unusable for Twenty Percent of the Population
September 29, 2021
"Talking about climate openly, even if uncomfortable, is a powerful way to create change."
Louis Rosenfeld Matt Jones Olga Khroustaleva Michael Leggett Karol MunozDo you want to work on climate? (Climate UX Discussion Series)
November 15, 2023
"We unapologetically move forward when high-level stakeholders miss deadlines, but also collect feedback via surveys."
Briana Thomas Christina RodriguezWhen Design Ops Comes in H.O.T. : A Tale of a Transformed Design Org
September 30, 2021
"The risk of using bad data is worse than using no data."
Ari ZelmanowDark Side of Democratization
October 13, 2023
"You can’t measure long-term relationship value with KPIs meant for short-term performance indicators."
John DevanneyThe Design Management Office
November 6, 2017
"People were doing incentives in different ways, using their own cards, which was confusing, so I centralized it."
Liza Pemstein Jane DavisScaling Research Via an Ops First Model at Clever
March 27, 2023
"I literally would have to jump up in meetings to introduce myself as the director just to be recognized."
Lisanne NormanWhy I Left Research
March 27, 2023
