By Sonia Koesterer
The world is imperfect. Every “happy path” intersects with dozens of crappy paths caused by typos, technical errors, and data that goes missing, is mis-assigned, adulterated, or is otherwise compromised/ stolen by evil data pirates. While you can’t prevent all data fails, you can avoid catastrophic failures, design graceful recoveries, and even turn the weakest points of your service into a strategic advantage. In short, you can create great services despite bad data.
The impact of data failure can be a humorous accident, minor inconvenience, or completely detrimental. For example, each year, the U.S. government falsely declares over 12,000 people dead due mostly to typos. In sheer percentage this is a rarity of a corner case of an edge case… but for those 12,000 individuals who suddenly lose their social security benefits, health insurance, bank accounts, and can’t easily prove they are alive, it’s catastrophic.
So design for the the edge-case! Understand the weakest points of your service, learn from them, and turn your failures into great experiences.
23. When you say
edge case, you're
just defining the
boundaries of what
you care about.
BETH DEAN
24. We need a process for
understanding & fixing
failures in design.
Solution
Problem Space
FAIL
FAIL
FAIL
FAIL
25. Why not borrow concepts
from Manufacturing,
Software Engineering,
Systems Engineering &
Safety Engineering?
Solution
Problem Space
FAIL
FAIL FAIL
FAIL
67. 1. First be fat.
Pilot new offerings at a small scale to learn where the
weak points are and gracefully recover.
2. Know your anti-use case.
Define what you don’t want your service to enable, and
create forcing functions to prevent disastrous mistakes.
3. Get ahead of changes.
Future-proof your service by continuously gathering
updated data from multiple sources.
4. Do the right thing.
Verify the validity of key transitions where major status,
account, or behavior changes occur for your users.
68. INTERESTING
THOUGHTS
FROM SMART
PEOPLE
1. Fat Thinking and
Economies of Variety
Venkatesh Rao
2. Go Home, Data, You’re
Drunk
Danielle Malik
3. Goodhart’s Law and Why
Measurement is Hard
David Manheim
69. If you torture the
data long enough,
it will confess.
RONALD H. COASE
Thank you.
@SoniaKoesterer