Key Takeaways
- He measured four things, not forty — Speed, lactate, heart rate and how it felt. That is the whole dashboard he described to COROS, and it was enough to win a World Marathon Major on debut.
- The data set the ceiling, not the target — His double threshold days were capped at about 1.5 to 2.5 mmol/L. The number told him when to stop pushing, which is the opposite of how most of us use a watch.
- Same lesson, different scale — My failed FTP test said the same thing at 289 W: I managed the number on the screen, and the signal was in the one I ignored.
A data scientist won New York
On 1 November 2026 Andreas Almgren won the TCS New York City Marathon in 2:06:01. It was his first marathon. He went through halfway in 1:04:04, ran the second half in about 1:01:57, and beat Yemaneberhan Crippa by 13 seconds after the lead group had stayed together to 35 km. Six weeks earlier he had won the world half marathon title in Copenhagen in a European record of 58:06.
The headline on my feed called him "the self-coached data scientist". Both halves are roughly right. He worked as a data scientist at Svexa, a Swedish sports technology company, before running full time. He writes his own plan. Nike's Jordan Donnelly told LetsRun he is "99% self-coached", and his longtime coach Urban Aruhn mostly argues with the plan rather than writing it.
I am not a 2:06 marathoner. My best is 3:23:41, from Fürth in 2025. But I run my own training the same way I run a technology organisation: measure the system, not the ego. So I read everything I could find on how he built this race, to see where his method and mine line up.
He checks four things, not forty
Asked by COROS how he judges a session, Almgren said: "I usually check four things: speed, lactate, heart rate, and overall feeling." That is the whole dashboard. No readiness score, no proprietary strain number, no twelve-tile app home screen.
This is the part that lines up most with how I work. Every race I have finished lives in one place, my athlete training database, and the race results on this site are copied from it rather than typed from memory. Garmin, Strava and Intervals.icu all hold my sessions, but they are inputs. One source of truth beats three dashboards that disagree.
My own list is just as short: heart rate variability for readiness, lactate threshold, and thresholds I set myself from lab tests rather than the defaults my watch suggests. I have taken five lab lactate tests since May 2024, in Singapore, Saigon and Erlangen. On the treadmill at Sportpraxis Erlangen in January 2025, my aerobic threshold came out at 1.3 mmol/L (149 bpm, 11.2 km/h) and my anaerobic threshold at 2.8 mmol/L (166 bpm, 13.6 km/h).
Put that next to Almgren's numbers and the overlap is the point. His 16 x 1 km session read about 1.5 to 2.5 mmol/L, the same band that sits between my two thresholds. The physiology he steers by is the physiology I steer by. Only the speed on the watch is different.
Where my setup goes further than his published one is the morning. Before I decide what a day's training looks like, I check my sleep score, resting heart rate and blood pressure, plus how well I fuelled the day before. Those four inputs set the breakfast: more carbohydrate on a hard-session morning after a thin fuelling day, less when the day ahead is easy. It is the same idea as Almgren's lactate cap, moved earlier in the day: let the data decide before the ego does.
The data set the ceiling, not the target
Almgren's signature session is the double threshold day: two controlled threshold sessions, morning and afternoon. For New York he made the morning a 50-minute run at the lower threshold and the afternoon longer, for example 16 x 1 km. According to the COROS data Marathon Handbook reported, he read about 1.5 mmol/L of lactate in the first half of that session and about 2.5 mmol/L in the second.
Read that again. The world half marathon champion, eight weeks out from a Major, was holding himself down. The lactate number was a ceiling. Most of us use a watch the other way round: we pick a pace target and chase it, and the data becomes a scoreboard.
I learned this the expensive way in October. My 20-minute FTP test ended after 7 minutes and 11 seconds at 289 W, exactly on target. The power trace looked perfect. Heart rate had gone from 142 to 174 bpm in under six minutes, and it was telling me the target was wrong before my legs did. I was steering by the number on the screen, and the signal was in the one I ignored. Almgren's lactate cap is the disciplined version of that lesson: decide in advance which number is allowed to stop you.
The embarrassing part is that the right number already existed. About five weeks before that test, a bike test at iQ-Move in Erlangen put my anaerobic threshold at 249 W, at 164 bpm and 4.0 mmol/L. The provisional FTP I set after failing, 248 W, landed within a watt of it. The 275 W in my Garmin profile was the odd one out. I had a lab-measured threshold and trained off a default.
Fixed blocks, then change one thing
Almgren told COROS he keeps "a set structure for maybe six weeks, and adjust if needed". His New York build ran about ten weeks from mid-August, totalled about 2,006 km and peaked at about 241 km in a week. Before it, he had never run further than about 32 km. He did not reinvent his training for the marathon. He kept the double threshold structure and changed the parts the distance demanded: longer long runs, heat sessions, fuelling practice at about 120 g of carbohydrate an hour, and a week in New York running the last 6 km of the course.
That is an engineering habit, not a running one: hold the system stable long enough that a change shows up in the data.
My own marathons in 2025 show what happens when the system does not stay stable. Fürth in June was my best, 3:23:41. Between Fürth and Berlin I picked up posterior tibial tendon dysfunction (PTTD), an injury to the tendon that holds up the arch of the foot. Berlin in September took 3:55:07, and Valencia in December 3:51:01. No dashboard rescues a block like that. What the data can do is tell you honestly that the plan changed, so the race target changes with it.
Where my method cannot copy his
Volume is the obvious gap. Almgren's peak week, about 241 km, is more than my monthly average: I ran more than 2,500 km in the whole of last year. The less obvious gap is the lab. He tests lactate in sessions as routine. For an amateur a lactate test is an occasional event, and heart rate has to carry most of the load in between.
The other gap is time. I fractured my right femur on 21 April 2026 in a low-speed cycling crash, and my recovery log runs on a different kind of data: six weeks without weight bearing, then a rebuild measured in days and milestones rather than kilometres. The comeback target is Challenge Roth 2027.
Where we do line up is who writes the plan. I coach myself, and I build the sessions with aitrainingplan.app, anchored on the thresholds from my lab tests. Nobody else writes my training. Almgren has Urban Aruhn to argue with his plan. My equivalent is the next lab test: it is the one opinion my plan cannot talk its way around.
What I am taking from it
- Fewer metrics, read every time. Speed, heart rate, feel, and lactate when I have a reading. Anything else is optional.
- Every hard session gets a ceiling before it starts. A heart rate or lactate number that ends the rep, decided the night before.
- Blocks of four to six weeks. Change one variable, then let the log show whether it worked.
- Rehearse the finish. Almgren ran the last 6 km of New York several times in race week. For Roth 2027 that is the run course.
The race-side breakdown of how he won New York is on findmynextrace.com, and the double threshold sessions, scaled for a hybrid training week, are on hybridtraining.guide.
Sources: Marathon Handbook, men's results; Marathon Handbook, how Almgren trained for New York; Canadian Running; World Athletics, Copenhagen report; COROS, Andreas Almgren training (January 2026); Read The Profile (citing LetsRun). All checked 3 November 2026.
Who is Andreas Almgren?
A Swedish distance runner and former 800 m specialist who won the 2026 TCS New York City Marathon in 2:06:01, his first marathon. Six weeks earlier he won the world half marathon title in Copenhagen in a European record of 58:06. He worked as a data scientist at Svexa, a Swedish sports technology company.
Is Andreas Almgren self-coached?
Largely. He writes his own training. Nike's Jordan Donnelly described him to LetsRun as "99% self-coached", and his longtime coach Urban Aruhn acts mostly as a sounding board.
Can an amateur train like Almgren?
You can copy the method, not the volume. He peaked at about 241 km a week. The transferable parts are a small set of metrics, hard sessions capped by lactate or heart rate, fixed blocks of several weeks and a long training log you actually read.
What data should a self-coached runner track?
Start with what Almgren named: pace, heart rate and how the session felt, plus lactate if you have a meter. Log them the same way every time so you can compare blocks, not just days.
Ready to Train Smarter?
Get personalized training zones, race predictions, and performance insights with our free calculators.