The biometric edge: Can algorithms help you crack the three-hour marathon?

Its very genesis is rooted in the tragic effort of Philippides, the 5th-century BC courier who, according to the legend, died immediately after completing his desperate run from Marathon to Athens to declare an Athenian victory in war.

Yet, my own recent pursuit of this ancient benchmark was a distinctly modern endeavour: a 14-week sprint where I treated my body as a living dataset, which ended up involving running not one but two full marathons.

The question was not how much pain I could endure, but whether a stack of consumer-grade, artificial intelligence-powered devices — from an AI watch coach to a biometric tracker engineered by Harvard alumni — could engineer a personal best performance over the 42.2km distance.

Crucially, I set out to determine if the final margin of victory still belonged to the runner’s heart, or if human ambition has, at last, been outsourced to the code.

The conservative algorithm and the need for a human push

The experiment began when I got the chance to enter the TCS Sydney Marathon about eight weeks before race day. This was a significantly shortened training window for a marathon, which normally takes 12-16 weeks to prepare for.

Initially, I was equipped with a new generation of wearable technology: a Samsung Galaxy Watch 8 Classic, featuring an AI running coach. The promise of an algorithm trained on millions of data points, capable of generating an optimal, personalised training path, was appealing.

But the reality was a significant disconnect between data science and raw athletic ambition.

Based on my initial fitness level, Samsung’s AI confirmed I could complete a marathon. But its generated plan was aggressively conservative, a strategy clearly designed to minimise injury risk and maximise sustainability — not to push the boundaries of human potential.

Its five-kilometre running target, for instance, was under 35 minutes. Its half-marathon goal was two hours and 20 minutes.

The AI model was risk-averse to a fault. It was a handy piece of kit but, at least for me, lacked the crucial element of human competitive drive. My goal ahead of Sydney was to run sub three-and-a-half hours.

To inject more personalisation and aggressive guidance, I turned to a more advanced Whoop device, a screenless, biometric sensor that samples physiological data 100 times a second. This device – engineered by Harvard alumni – is also integrated with OpenAI’s ChatGPT. This allowed me to interrogate my data — strain, recovery, and sleep — in natural language via a chat function, creating the illusion of a true digital coach. I was able to ask it how my training on a particular day affected my marathon preparation. It even suggested weekly programs based on my performance and goals.

Then I met Lydia O’Donnell. Nike’s Pacific run coach and elite runner. ‘Lyds’ finished 20th in the New York Marathon in 2018 and second in the 2015 Melbourne Marathon. She has also helped scores of women achieve their fitness and athletic goals via Femmi, an app she developed that offers personalised training programs in sync with a user’s menstrual cycle, guiding runners to better understand hormonal and energy fluctuations, and the link to performance.

There is no AI with Lyds. Just boundless enthusiasm and belief that people – if they put in the work – are more capable of achieving what they think they can.

In the final lead-up to the Sydney marathon, I used this biometric data in tandem with the Nike Run Club app and Lyds’ guidance. While Nike’s app didn’t offer the same degree of personalisation as the Samsung AI, it provided a simple, tried-and-tested plan, complete with guided runs from its head coach, Chris Bennett — a human, not a bot.

I welcomed Coach Bennett’s recorded voice encouraging me to “dig a little deeper.” His weekly program involved a long run of up to 32 kilometres, a couple of fast tempo runs, with the remainder dedicated to slow recovery runs. In total, I was running 70-90km a week.

And to ensure I stayed on track, Lyds was only a video call or WhatsApp message away.

The supremacy of recovery: biometric data as an investment

The most profound lesson the technology delivered, repeatedly and with undeniable data points, was the quantifiable importance of recovery.

The slow runs were the biggest behavioural change. I had never valued the recovery runs, which help repair muscles and consolidate performance from faster runs. And Lyds stressed that ‘slow means slow’. She told me she often completes recovery runs at around a pace of five minutes, 45 seconds per kilometre, ensuring the body is primed and ready to really go fast in the faster runs.

Yet, the most potent recovery metric was sleep. The Whoop provided two daily summaries, one in the morning suggesting what training I should do, and one in the evening, rating my day and telling me when I should “wind down” for bed. It’s hard to believe that one of the biggest secrets to bettering your athletic performance is doing nothing, or more specifically sleeping.

“I’m really adamant that my athletes get eight plus hours sleep a night, and then if they’re training for, say, a marathon or an ultra, adding in naps throughout the day where possible, even if it’s a 20 or 15 minute nap,” Lyds says.

“Sleep is like one of our biggest performance enhances that we have access to, and it is so overlooked by a lot of people.“

The data-driven tracking also serves as a crucial injury and management tool for elite athletes. Darren Burgess, high performance manager for the Adelaide Crows, explained how technology like the Australian-developed app Eclipse Yourself paired with an Apple Watch Ultra allows his athletes to strike the right balance between activity, recovery and readiness.

The data provides a non-confrontational prompt, offering an opportunity for coaches to check in on their players’ welfare. “It allows you to go to the players and say, ‘OK, is there anything going on I’m not aware of’,” Burgess told me. The technology identifies the problem; the human coach addresses the context.

And working with Lyds I began to realise how critical this nexus is in training for a marathon.

The race day discrepancy: GPS vs. official clock

Before completing Sydney, I also got the chance to enter the Nike Melbourne Marathon Festival. I needed to step up my efforts while keeping my body injury free. There was only a six week gap between the Sydney and Melbourne marathons, so I also consulted exercise physiologist Aaron Jenkins for more strengthening and fine tuning.

Lyds advised me to treat Sydney like a “long hard training run”, aiming to complete it in three hours and 40-50 minutes.

But on race day, I was surprised.

My finishing time was 3:17:21 — 20-minutes better than my previous personal best, and achieved despite a significant pre-race setback involving a border collie biting my leg.

But the race highlighted a persistent technological frustration: the discrepancy between official course time and GPS watch data. My smart watch logged my time as 3:12:42. This was in part to starting in the four-hour wave, which meant I wove through other runners the entire race, resulting in running more than the course’s 42.2km. Then there was the high level of interference with thousands of other runners also using GPS watches and signals dropping in built up areas like Sydney’s CBD. It is a technical reality every runner using a device must acknowledge. The race clock rules.

Despite this digital noise, I was still ecstatic with my time. The run had felt comfortable the entire distance, with my pacing consistent.

Lyds immediately recalibrated the goal for Melbourne. She said I could do a sub-three. I thought she was joking. ‘Yes, you can Jared. You just need to believe that you can do it.”

And my training began to shift from AI bots to the invaluable psychology of human coaching.

One Saturday morning, I programmed into my Apple Watch Ultra one of Lyds’ plans, running 35km with three five kilometre intervals at my target marathon pace, 4:15/km.

I felt ordinary when I started that run at 7am. I didn’t want to do it. Yet I completed the first interval after a 10km warm up above target at an average pace of 4:08. It felt good. I completed the second and third intervals in the same time. It was then I began to believe that a sub-three hour marathon, while still ambitious, may not be beyond me.

But on the morning of the Melbourne race, the Whoop delivered a stark warning that tested my resolve. My recovery score was in the red, for the first time in weeks, at 43 per cent. My resting heart rate had risen from below 40 beats a minute to 46bp, while my skin temperature was “elevated” at 34.7 degrees Celsius. “This is clear signs that your body is under acute strain. This may reflect race-week load, nerves or mild inflammation,” Whoop’s summary read.

The body was signalling a system breakdown, detected instantly by the technology. I ignored the warning, a clear case of human aspiration overriding AI caution. I had this. I also had a pair of fast Nike Alphafly race shoes, which former Olympian Jeff Riseley – who completed last year’s Melbourne marathon in 2:38:17 – could shave off up to eight seconds a kilometre from my time.

But marathon champion Steve Moneghetti later told Matt McKenzie from the Good Good Running Podcast that while ‘supershoes’ were a welcome innovation, you still needed to put in the work. You still have to run.

On the course, I relied on Apple Watch Ultra’s custom pace function, setting the first half at 4:20/km and the second at 4:10/km. This digital guardrail proved vital.

I started too quickly, averaging 4:08/km for the first five kilometres, but the watch’s alerts forced me to slow down and settle into a 4:16 groove until the 30km mark.

As the race entered its crucial, final stage — things got tough — Lyds warned me the day before I needed to expect pain. This wouldn’t be like Sydney. I’d be pushing myself the entire distance.

It was in the final 10km that the technology retreated, and human coaching took over. At 30-35km, my average pace jumped to 4:25/km. I then remembered Coach Bennett’s guided run advice on how to complete a “systems check,” forcing myself to focus on form. I made sure my shoulders were relaxed, fists were unclenched and began to focus on breathing. In through my nose, out through my mouth. Repeat.

I managed to bring my average pace back down to 4:10 in the final two kilometres. My Apple Watch clocked me finishing the 42.2km well within my goal at 2:57:40. Accounting for the digital noise, I thought it was enough of a buffer to scape just under three hours.

I was wrong.

The official clock delivered the final, objective result. My chip time was 3:00:17 — a mere 18 seconds behind my goal. It was a failure of the sub-three. If only I had dug that little bit deeper. A split second per kilometre. Marathons are humbling.

The Australian’s technology editor Jared Lynch crosses the Melbourne Marathon’s finish line. Pic: supplied

A tool, not a replacement

The pursuit of the three-hour marathon proved that while technology is a powerful accelerant, it is not a panacea. The machines can optimise recovery, quantify strain, and enforce pace with relentless objectivity. They democratise access to high-level training data and promote consistency, which Lyds calls the true key to progress.

Ultimately, the most successful training model is a hybrid one, leveraging the power of data for self-awareness and the empathy of a human coach for decision-making. As Lyds advised, runners must find a “healthy balance”. “The thing about the data is it should allow you to make good decisions to get that consistency in your training. And that’s where we get better as runners.”

While a sub-three hour marathon eluded me, it was a feat I thought impossible only a couple of months ago.

The 18-second gap was a powerful reminder that while data can push the human body to the brink of a new frontier, the final margin of victory still rests with the runner’s will — an element that remains outside the algorithm’s code. The machines are making the sport smarter, but the heart of the marathon remains decidedly, and necessarily, human.

This article first appeared in The Australian as The biometric edge: can algorithms help you crack the three-hour marathon?