I drove 158,940 miles
and found a story in the data.

What does an Uber driver actually earn?

The number Uber shows you in the app is a fiction. So is the figure on your 1099. Both are technically true but neither answers the question that matters: per hour of your life, what does this work pay? The honest answer requires reconciling at least three different ledgers, none of which agree with each other.

I started driving for Uber in July 2018 to supplement income during a career transition. What began as a side gig became a multi-year operation that produced — without my realizing it — one of the richest personal datasets I'd ever encounter. Uber records every trip with timestamps, distances, surge multipliers, and per-line-item payments. Lyft tracks sessions and payouts. Gridwise logs total clocked time including the unpaid waiting and deadheading that Uber's data conveniently omits.

In May 2026 I requested the full archives. What follows is what I found when I joined them together.

Reconciling six exports that disagreed with each other.

The single hardest part of this analysis was getting four different platforms to tell the same story. Each export had its own definition of a "trip," its own version of what counts as earnings, and its own set of inexplicable gaps. The 2024 Uber data showed me a $1,920 day — except I'd only completed four rides that day. The Lyft rides file had 1,484 trips but the payments file showed I'd been earning Lyft money during years where no rides existed in the rides export. Gridwise had hours but no platform attribution.

The advertised hourly wage isn't wrong. It's just measuring the wrong thing.

If you look at Uber's data alone, my "on-trip" earnings averaged $95 per hour. That sounds great. Now subtract the time spent driving toward riders, waiting at the airport queue, repositioning after a long trip, sitting through dispatch lulls — all the time the meter doesn't run but the work continues. The honest rate is less than half.

Annual breakdown

The headline rate hides a more interesting story: my hourly compensation has been remarkably stable over three years, despite Uber raising its take rate from 17% to 44% over that span. I've been running faster to stay in place.

Three years of data, told one month at a time.

A monthly view reveals patterns invisible at the annual level: the autumn surge that drives every year's earnings curve, the spring trough that I haven't successfully fought against, and the steady upward drift in my month-to-month consistency as I've learned the Pittsburgh market.

Saturday is worth more than Monday and Tuesday combined.

If you've spent any time driving, this won't surprise you. What might surprise you is the magnitude. The average Saturday produced $453 in net earnings; the average Tuesday produced $272. Over three-plus years that delta has compounded into something like $50,000 of difference between my weekend earnings and my early-week earnings — for the same hours of work.

The deeper finding is what the day-of-week pattern reveals about cost of complacency. Tuesday is the cheapest day to take off, but Tuesday is also the day where my marginal hour earns the least. The optimization isn't to drive more on Tuesday — it's to never drive past 8 PM on Tuesday, and to always be driving by 5 PM on Saturday.

While my hourly rate stayed flat, Uber's take rate doubled.

In 2018, Uber kept about 17 cents of every dollar a rider paid. By 2026, they're keeping 44 cents. This isn't a secret — Uber acknowledges it in their public filings — but the lived experience of it is striking. To earn the same $40 hour I earned in 2023, I have to be substantially more efficient in 2026, because Uber sits on a larger share of the rider's payment.

67% of my trips have zero surge. The other 33% pay disproportionately.

Conventional wisdom says successful drivers chase surge. My data says something more nuanced: the vast majority of trips happen at base price, and those base trips form the backbone of consistent income. Surge isn't a strategy — it's a windfall that occurs alongside the strategy. The 248 trips I've completed at 3.0× or higher pulled an average fare of $32, more than double the no-surge average.

Three patterns that drove real money.

Airport trips pay 2.7× a normal trip — but I rarely take them.

Pittsburgh International Airport runs averaged $42.69 per trip, against $15.76 for the typical city trip. Over the analysis period I only completed 139 airport trips out of 24,744 — about 0.6%. If I doubled that share, my earnings would rise by roughly $2,000 a year at no additional time cost.

Tipping rate of 39% — well above industry benchmarks.

Across the period I received tips on 39.1% of trips, totaling $35,911. Industry research suggests typical rideshare tip rates of 16% to 25%, so this represents one of the strongest signals in the dataset that customer experience matters. At an average $3.60 per tipped ride, tips alone funded the equivalent of roughly 2,800 extra trips' worth of base fare.

Uber and Lyft pay almost the same per hour — but I run 97% Uber.

When measured properly, Lyft pays $37.21 per clocked hour to Uber's $41.52 — a gap of just 10%. The per-trip narrative ($12.06 Lyft, $12.77 Uber) tells the same story. Lyft is a viable supplement, especially during the late-night bar-close window. I've under-utilized it.

The point isn't the rideshare numbers. The point is the method.

Most data-analysis projects look the same: a corporate dataset, a familiar business question, a chart that confirms what management already suspected. This project was the opposite. The data was my own. The reconciliation was non-obvious. The conclusions changed multiple times as new sources surfaced. That's what real analytics work looks like.

The skills demonstrated in this case study — ingesting messy multi-source data, identifying anomalies that warrant investigation (like the $1,920 day that turned out to be a settlement payment), reconciling ledgers that don't agree, computing the right unit economic rather than the obvious one, and presenting the findings narratively — are the same skills that turn data into business value at any organization.

I taught myself this stack over the course of the analysis: Python, Pandas, time-series reconciliation, financial waterfall analysis, statistical pattern detection, and the communication discipline of writing for an audience that needs to understand the numbers, not just see them.