Airline Delays · The 7 a.m. Domino
When a plane lands late with under an hour before its next flight, that flight leaves late about 9 times in 10.
- Role
- Sole analyst
- Stack
- SQL (SQLite), Power BI
- Data
- U.S. DOT flight records · July 2026
Imagine you’re an airline executive reviewing schedules. Why are delays happening, how much are they costing us, and how can we improve?
- Why are we late? 42% of delay minutes start with a plane that was already late (on flights arriving 15+ minutes late), and the delay grows on the ground.
- What does it cost? Late planes with under an hour on the ground added 734,000 minutes in July, about $74M at stake.
- How do we improve? Give late-arriving planes more turn time where delays snowball most, starting with Atlanta.
Why flights run late
Share of delay minutes by cause, flights arriving 15+ min late
A late first flight makes finishing the day an hour or more late 1.7× as likely.
Where it costs the most
Millions of dollars at stake from tight turns after a late arrival, top 10 airports
At stake = minutes added × $100.76 per minute, Airlines for America’s 2024 average aircraft block-time cost. A yardstick, not an exact bill.
Source: U.S. DOT Bureau of Transportation Statistics, On-Time Performance, July 2026
The question
I wanted a real business question, answered with real data. So I took one month of U.S. government flight records, every domestic flight the major airlines reported in July 2026, about 632,000 of them, and followed each plane through its day by tail number. The summary above has the short answers; the rest of this page is how I got there.
One plane’s journey
What turns one late flight into a late day?
Before the big numbers, one plane. The data has a tail number on every flight, so you can follow a single aircraft through its day. On July 1, N1776R hit an airline problem on its second flight, landing 92 minutes late, and the next flight inherited almost all of it. Two weeks later, the same plane picked up small delays and still landed in San Diego 25 minutes early. What made the difference?
July 1
Arrival delay at each stop, in minutes. An airline problem on flight 2 rides into flight 3.
July 15
Arrival delay at each stop, in minutes. Small delays the schedule absorbed.
The domino effect
Does a late start ruin the day?
Across all 138,730 plane-days with two or more flights, a first flight that left 15 or more minutes late made finishing the day an hour or more late 1.7 times as likely: 30.9% of late starts, against 17.8% of on-time starts. But it wasn’t destiny. Almost 30% of planes that started late still landed their last flight early. So where does the time get made up, or lost?
How each plane’s day ended
Share of plane-days by how late the last flight arrived, July 2026
- First flight left 15+ min late
- First flight on time
The lever: time on the ground
How do late planes catch up, or fall further behind?
My first guess was that late planes catch up in the air: faster flying, quicker taxi. The data said partly. Planes that recovered won time back in both places, 37 minutes in the air and 45 on the ground, but planes that fell further behind roughly broke even in the air and lost 65 minutes at the gate. And the schedule decides a lot of that. After a late arrival with under an hour on the ground, the next flight left late about 9 times in 10; with 90 minutes or more, about half the time.
Where time is won and lost
Average minutes won back (+) or lost (−) over the day, planes whose first flight left 15+ min late
Next flight leaves late
Share of turns after a 15+ min late arrival where the next flight also left 15+ min late, by scheduled time on the ground
Where to make changes
Which airports should get buffer time added?
Adding time to every schedule isn’t free: a plane sitting at the gate isn’t flying passengers. So the question is where extra time pays off. Atlanta is the clear first place to look, with the most delay minutes added of any airport (32,110, about $3.24M at stake) and the most added per turn among the 17 busiest. Denver adds a lot too, but mostly because it’s so busy. Smaller airports can add more per turn, but they have far fewer turns, so less is at stake.
Size vs. severity
Airports with 100+ tight late turns. Bubble size is dollars at stake; hover a bubble for its airport.
What I’d recommend
Give late-arriving planes more time on the ground, but only where delays snowball, starting with Atlanta.
Blanket buffers would cost flights. The next step would be testing this on more months of data, including winter.
About the dollars: I converted minutes at $100.76 each, Airlines for America’s 2024 average direct cost of aircraft block time (taxi plus airborne). It’s a yardstick, not an exact bill, which is why the $74M is “at stake” rather than lost or saved.
How I did it
Data. The Bureau of Transportation Statistics’ on-time records for July 2026: every domestic flight by the reporting airlines, with tail numbers and delay causes. Real, public data. I loaded it into a SQLite database and kept 38 of the 110 columns.
Questions first, then SQL. I wrote seven queries, each answering one question, from “what causes the most delay minutes?” to “which airports add the most delay on tight turns?” I used window functions (ROW_NUMBER, LAG, COUNT OVER) to put each plane’s flights in order and follow delays from one flight to the next.
Checking the work. Each query had checkpoints: row counts and totals that had to match before I moved on. I checked one plane’s turn times by hand against the query. A few decisions changed the analysis, and I logged each one with what I gave up, for example leaving out about 2,100 late-arrival turns that looked like aircraft swaps.
AI-assisted. I built this with Claude as a guide: it explained the concepts and set up exercises, and I wrote, ran and checked the queries myself. When an autocomplete suggestion quietly changed a filter and inflated the headline number by about two-thirds, the checkpoints caught it.
Visuals. I built the report in Power BI with one accent color and takeaway titles; the PDF above has the original pages. The charts on this page are rebuilt in HTML from the same exported numbers, so they stay sharp and readable on any screen.
Caveats
- One month, July, at peak travel season. These are July 2026 findings, not a full year.
- Times are local, and each plane’s flights are ordered by scheduled departure, which can misorder red-eyes. As a check, the previous flight’s destination matched the next flight’s origin 96.9% of the time.
- The dollar figure is a yardstick: a block-time cost applied to minutes at the gate, from Airlines for America, an industry trade group.
- The turn-time results show a strong pattern, not proof that more buffer would win back every minute. Congestion, crews and gates matter too.
- Cancelled flights and aircraft swaps are left out.