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.
Garbage in, garbage out? Measuring error rates to get ready for AI
Summary
We’re all aware of a big push to implement AI everywhere, including in the services that many of us are working on. It seems only fair to try to give the AI some good quality input in the hope of getting decent output from it. Or, being more pessimistic: we probably expect to get some level of errors from the AI, but what do we know about the error rates in what we’re putting into the AI? In this session, we will compare our ideas on identifying errors and measuring error rates, including thinking about errors in six ways: 1) Problems along the way 2) Wrong result 3) Unnecessary action 4) Delayed-impact problem 5) Non-uptake or over-uptake 6) Technology problem We’ll wrap up with “tips and next steps”: an opportunity to consider what we now need to find out or do differently.
Key Insights
-
•
Errors in data collection and user input are foundational issues that compromise AI and service outcomes.
-
•
Users often 'fudge' answers due to ambiguous questions, privacy concerns, or to achieve a desired outcome.
-
•
Non-uptake, where users abandon a form or process, is a major source of error but is rarely published or measured.
-
•
Mistakes can be categorized as problems along the way, wrong results, unnecessary actions, and delayed impact issues.
-
•
Multiple accounts creation often occurs due to users forgetting existing accounts, leading to data duplicates and service inefficiencies.
-
•
Measuring error rates is complex; different metrics (per person, per attempt, completion vs. start) yield different perspectives.
-
•
Elections provide a useful model for measuring data quality, using turnout, participation, and eligibility rates.
-
•
Data quality deteriorates over time due to changes like moving, name changes, loss of documents, or organizational restructuring.
-
•
AI initiatives can provide a compelling rationale and funding opportunity for improving longstanding data quality problems.
-
•
Frameworks like the UK Government Data Quality Framework help organizations systematically assess and address data issues.
Notable Quotes
"If we get garbage in, we get garbage out — this is true for AI as much as for surveys or forms."
"People can make all sorts of inventive mistakes on their forms that AI struggles to interpret."
"Sometimes a form forces you into a wrong answer by giving inappropriate options."
"I’ve seen people fudge their date of birth so their child can attend a summer camp they aren’t technically eligible for."
"A major error in many services is users creating multiple accounts because they can’t find or reuse existing ones."
"An error might not be immediate; data can be fine when collected but deteriorate over time and cause problems later."
"Completion rates (conversion rates) and dropout rates are simple metrics but often not tracked or shared."
"Organizations rarely know their error rates, which limits their ability to improve user experience or data accuracy."
"Linking data quality efforts to AI initiatives can help secure attention and budget for necessary improvements."
"Data quality involves accuracy, completeness, uniqueness, timeliness, and representativeness—not just error reduction."
Or choose a question:
More Videos
"You’re not supposed to sell to the US public sector unless you’re WCAG 2.0 AA compliant."
Sheri Byrne-HaberAccessibility at Scale
June 9, 2021
"If you skip iterative testing, you risk building features nobody really wants."
Prayag Narula Hannah HudsonEmpowering Designers to do Good Research
March 11, 2022
"Don't just flip the switch and say go talk to customers; you have to set guardrails and train people."
Janelle EstesUX Research Trends
January 28, 2021
"Start with content first wireframing and avoid lorem ipsum to prevent treating words like decoration."
Craig Brookes Andreas Huebner Morgan Quinn"Just Make it Look Good" and Other Ways We're Misunderstood
June 11, 2021
"Everyone goes through quite an extensive application process, but it was stress-free and informal."
Marc Fonteijn Ru ButlerIncrease your confidence, influence, and impact (through a Professional Community)
December 3, 2024
"Managing vendors is a full-time job, and research ops teams often coordinate numerous vendors for tools and recruitment."
Kate TowseyThe State of ResearchOps: More Than Just Theory
June 20, 2019
"Psychological safety is the number one factor determining team effectiveness and performance."
Alla WeinbergDesign Teams Need Psychological Safety: Here’s How to Create It
September 9, 2022
"We have over 30 contracting teams spread across roughly 10 vendors all working on parts of the veteran experience landscape."
Shawna Hein Kevin HoffmanCreate a Cohesive Civic Design Practice Across Agency, Vendors, and Contracts
November 17, 2022
"When ideas don't make it into the product, I do a post-mortem to share opportunities missed and improve the process."
Tricia WangSCALE: Discussion
June 15, 2018
