From the “Data but Make it Personal” series – Part 8
When the Data Doesn’t Make Sense (But Matters Most)
In every dataset, there are outliers — values that don’t follow the trend, don’t fit the model, don’t behave the way you expect. Analysts are trained to spot them, question them, and often… discard them.
But here’s the thing: not all outliers are errors. Some are just rare. Special. Worth paying attention to.
This blog — the final chapter in the Data but Make It Personal series — explores what outliers can teach us about data, and about life.
Why Outliers Matter
Outliers often carry important signals in analytics. They can uncover hidden problems, reveal opportunities, or even point to entirely new patterns you weren’t aware of.
For example, a sudden spike in website traffic might not just be a bot — it could signal virality.
An unusually high sales figure in one region could reflect a brilliant but undocumented local campaign.
A single dissatisfied customer could surface a systemic flaw in your process that others simply haven’t vocalized yet.
The key is knowing the difference between noise and signal — between what should be excluded and what deserves investigation.
As analysts, it’s tempting to scrub our data clean, aiming for perfect models and predictable trends. But sometimes the most meaningful stories lie in the values we almost threw away.
The Least Likely Scenario (And What It Reminds Us)
In life, just like in data, some things simply don’t fit the expected pattern — yet change everything. Here’s my very own outlier-real-life story and what I learned from it.
Years ago, while living in Madrid on a research scholarship, I found myself desperately apartment hunting. On a student budget and with limited options, I was ready to take whatever came with natural light and at least a hint of dignity.
That’s how I ended up knocking on the door of a shared apartment I’d seen in an online ad. A tall, polite guy with a familiar accent opened the door. He was very handsome, which I noticed briefly, but it was not relevant — I was looking for a window, not romance. He showed me the available room, which did, in fact, meet my modest dignity standards. I thanked him and left hoping to get picked out of the probably long list of applicants (housing in Madrid was very complicated back then. Actually, it still is but that’s another story.)
I never got a follow-up text. Found another place. Eventually moved back to Mexico. The end, right?
Not quite.
Two years later, I noticed a random Facebook friend request from a name I didn’t recognize. The profile picture was blurry, the name vaguely familiar — and for some reason, I accepted.
Minutes later he messaged me: “You took your time.”
It turns out he’d remembered everything — the timing, the roommate search, even my knock on his door. Over the years he’d checked in from time to time with little messages. Eventually, life aligned. I was traveling for work, he was living in Brazil, and he flew to meet me. Coffee turned into dinner, dinner into plans, and eventually, into a life.
Today we’re married, living in Australia, with a son who embodies every bit of that improbable story.
And that’s the thing about outliers: a single observation can shift the entire model.
Outliers in Practice: How to Handle Them
So, what do we actually do with outliers in our work? Here are some principles and practical tips.
- Investigate before discarding. Don’t just delete or ignore unusual data points. Check their source, verify their validity, and understand their context. Outliers can be errors, but they can also reveal an edge case worth exploring.
- Spotting them visually. Charts are your best friend for detecting outliers. Some options include boxplots, excellent for quickly seeing values outside the interquartile range. Scatter plots, great for identifying unusual points in bivariate data. Time series line charts, where sudden spikes or drops stand out clearly. Histograms, which help you spot long tails or strange clusters. If something looks odd to your eye, it’s worth a closer look.
- Decide if they belong. Once validated, decide if this point is relevant to the business question. Does it distort or enrich the story?
- Model with and without them. Compare results. Sometimes including an outlier tells one story, excluding it tells another. Both stories have value — our job is to communicate that clearly.
- Use them to generate hypotheses. Outliers often lead to deeper investigation and fresh business questions. Why did this happen? Could it happen again? Should it?
The End of the Series
This is the final post in my Data but Make It Personal series — a project that started as a way to make technical concepts approachable and relatable. Along the way, it became something more: a love letter to the nuance, surprises, messiness, and occasional poetry of data.
Over this series we’ve explored how data mirrors life, how models shape stories, how aggregation, joins, granularity and even dashboards all have their human analogies. And finally, how even the smallest, rarest signals can have meaning if we’re paying attention.
Specifically, about outliers I can say that they don’t always fit. They’re not always easy to explain. But sometimes… they’re the whole point.
And with that, this series concludes — but I hope the curiosity it sparked doesn’t.
Because when it comes to both life and data, the story is never quite finished.