There are roughly 50 million people in the United States alone who regularly run. Globally, the number is somewhere in the hundreds of millions - it’s hard to land on an exact figure because most runners never enter a race, never join a club, and never appear in any database. They just lace up and go.

Of those hundreds of millions, a meaningful subset are serious about it. Not professional. Not sponsored. Not trying to qualify for the Olympics. But serious in the way that matters: they have a goal, they follow a structure, they think about their training between sessions, and they care about getting better. They're the runner chasing a sub-4 marathon, chipping away at a half marathon PB, or building toward their first ultra.

These athletes, call them serious recreational, or competitive amateur, or whatever label you prefer, and they represent the fastest-growing segment of endurance sport. Marathon participation in the US hit 432,000 finishers in 2024, up 5% year-over-year and climbing back toward the all-time high. Half marathon finishers globally topped 1.5 million across 190 tracked events, up 21% on the previous year. And that's just the people who race. The training-but-not-racing population is vastly larger.

Here's the thing about this group: almost none of them have a coach.

The 95% problem

The statistic gets quoted so often it's become a cliché, but it persists because it's true. Roughly 95% of endurance athletes train without any form of individualised coaching. They use static plans downloaded from the internet, follow programmes from books, copy workouts from Instagram, or just run by feel with no structure at all.

This isn't because they don't want coaching. Ask any serious recreational runner whether they'd benefit from a coach and the answer is almost always yes. They know they would. They've read the articles about periodisation and progressive overload and the 80/20 rule. They understand, at least conceptually, that someone who knows what they're doing could help them train smarter, avoid injury, and race faster.

They don't have a coach because they can't get one. And the reasons are entirely structural.

The cost barrier

The most obvious barrier is money. A qualified running coach in the UK or US typically charges between $100 and $300 per month for online coaching, including a personalised plan, weekly check-ins, session adjustments, and some form of communication channel. At the premium end, with experienced coaches who work with smaller rosters and offer more intensive support, you're looking at $250 to $500 per month.

For a 16-week marathon block, that's somewhere between $400 and $2,000. For year-round coaching, which is how serious athletes actually improve, it's $1,200 to $6,000 annually.

This is a meaningful expense for most people. It's not prohibitive in the way that, say, a personal chef is prohibitive, but it's enough that most runners can't justify it. Especially when they're already paying for race entries, shoes that last 500 kilometres, a GPS watch, and a gym membership. Coaching becomes the thing that would be nice to have but isn't strictly necessary. So they download an off-the-shelf plan and get on with it.

The irony is that coaching would often save them money. An injured runner who misses their goal race has wasted every penny they spent on entry fees, travel, and accommodation. A coach who prevents that injury - by catching a training load spike early, or recognising the signs of accumulated fatigue - pays for themselves several times over. But the saving is invisible and hypothetical, while the monthly fee is concrete and immediate. So the fee wins, and the runner gets injured.

The quality problem

Cost isn't the only barrier. Even runners who can afford coaching face a quality problem that nobody in the industry talks about honestly.

The coaching market is unregulated. Anyone can call themselves a running coach. Some coaches are exceptional, deeply knowledgeable, experienced, attentive, and genuinely invested in their athletes. Some are mediocre. Some are actively harmful, prescribing cookie-cutter plans with their name on top and calling it personalised coaching.

The athlete has almost no way to distinguish between these categories before committing. Certifications exist but they're inconsistent. A weekend certification course and twenty years of coaching experience carry the same title. Testimonials are curated. And the quality of coaching is hard to evaluate from the outside because the outcomes are confounded by everything else in the athlete's life.

Even good coaches have structural limitations. A coach managing thirty athletes, a typical roster for a full-time online coach, has limited bandwidth for each one. Check-ins happen weekly, not daily. The coach sees your TrainingPeaks upload but doesn't see that you were up three times with a sick child, that your work stress has been weighing you down the past week, or that the niggle in your Achilles has shifted from occasional to persistent. They adjust the plan based on the data they have, which is always a fraction of the data that matters.

This isn't a criticism of coaches. It's a description of a model that has inherent constraints. A human being can only process so much information about so many athletes. The economics of one-to-many coaching mean that the coach's attention is always divided, and the athlete who needs the most attention right now is competing with twenty-nine others who also need it.

Why generic plans fail

So if most athletes can't access coaching, what happens? They use plans. And plans, as we've discussed before, have a fundamental problem: they don't know you.

The training science behind good plans is sound. Pfitzinger's marathon schedules, Daniels' VDOT tables, Friel's periodisation frameworks; these represent decades of accumulated coaching wisdom. The problem isn't the science. It's the application.

A static plan assumes the athlete exists within a vacuum. It assumes you'll sleep eight hours, eat well, manage your stress, stay healthy, and absorb each training stimulus exactly as the plan designer intended. It assumes your Tuesday tempo run will be run on a Tuesday, at the prescribed pace, in reasonable conditions, after adequate recovery from Sunday's long run.

In practice, life intervenes. The research is unambiguous about what happens next: training errors - too much load, increased too quickly, without adequate recovery - account for the vast majority of running injuries. A study published in the British Journal of Sports Medicine, the Garmin-Runsafe Running Health Study, which tracked 5,205 runners across 87 countries over 18 months, found that when a single run exceeded the longest run of the previous 30 days by more than 10%, injury risk increased by 64%. Not a gradual accumulation over weeks. A single session spike.

This is exactly the kind of error a coach prevents and a plan can't. The coach sees that your long run last Sunday was hard, your sleep has been poor, and today's scheduled 20-miler is a recipe for disaster. The coach says: do 14 miles easy, and we'll build back up next week. The plan says: 20 miles. The plan is wrong, but the plan doesn't know it's wrong, because the plan doesn't know you and can’t adjust.

The human cost of this is significant. Injured runners don't just miss training. They miss the race they've been building toward for months. They lose fitness that took weeks to build. And a meaningful percentage of runners who get injured never come back to the sport at all. The plan-to-injury pipeline isn't just an inconvenience. It's the single biggest source of attrition in endurance sport.

The economics of AI coaching

This is the gap. Hundreds of millions of runners. A tiny fraction with coaching. The rest relying on generic plans that break the moment life gets complicated. The coaching wisdom exists. The training science is well understood. The problem has always been delivery.

Human coaching doesn't scale because humans don't scale. A great coach can work with twenty or thirty athletes. They can't work with twenty thousand. And the economics of one-to-many coaching create a structural ceiling: the only way to serve more athletes is to give each athlete less attention, which eventually degrades the quality below the point where it's meaningfully better than a good static plan.

So if human coaching doesn't scale, what does?

AI changes this equation in a way that nothing else has. This is also where we need to be careful, because "AI coaching" has become one of those phrases that means everything and nothing. Most products that call themselves AI coaching are doing one of two things: either generating a static plan using an algorithm (which is just a more sophisticated PDF), or wrapping a chatbot around a generic programme and hoping the conversational interface makes it feel personal.

Neither of these is coaching in any meaningful sense. Coaching isn't plan generation. Coaching is the ongoing, adaptive relationship between the plan and the athlete's reality. It's the daily judgment calls. It's knowing when to push and when to pull back. It's reading the signals, physiological, psychological, contextual, and making intelligent decisions about what this athlete needs right now.

When it's done properly, the early evidence is compelling. Research reviewing data from over 11,000 athletes found that AI-adapted training plans improved workout consistency by roughly 16% and reduced injury rates by up to 23% compared to static plans. Not because the underlying training science was different, but because the plan actually responded to what was happening.

What world-class coaching actually looks like

If you've ever worked with a great coach, you know the experience isn't primarily about the workouts. The workouts are the output. The input is the relationship: the coach's understanding of who you are, how you respond to training, what your patterns and tendencies are, and what's happening in your life that might affect your training.

A great coach knows that you go too hard on easy days. They know that you get anxious before races and tend to over taper. They know that when you say your calf is "a bit tight," it's actually been bothering you for two weeks and you're hoping it'll go away. They know that you train best when you have variety, or that you need consistency, or that you respond well to threshold work but fall apart in long VO2max intervals. They know you.

This is the standard that AI coaching needs to meet. Not "better than a PDF," that's a low bar. The standard is: genuinely comparable to a good human coach, for the aspects of coaching that matter most.

What makes this newly possible is the convergence of three things.

First, wearable technology has matured to the point where continuous physiological data - heart rate, HRV, sleep quality, training load - is available from devices that most serious athletes already own. The data that used to require a sports science lab is now on your wrist.

Second, large language models can carry context across weeks and months of interaction. They can remember that you mentioned a dodgy left ankle six weeks ago. They can notice that your tone in check-ins has shifted from enthusiastic to flat. They can have a conversation about whether you're dreading your long run, not just crunch your TSS numbers.

Third, it's affordable. Because an AI coach isn't limited to thirty athletes, the cost per runner drops to a fraction of what human coaching costs. Coaching that used to require $200 a month becomes accessible to anyone with a smartphone and a running watch.

The comparison that matters

The question isn't whether AI coaching is as good as the best human coach in the world. It isn't; not yet, and perhaps not ever for certain dimensions of the relationship. The question is what the athlete's realistic alternative is.

For the 95% of endurance athletes without a coach, the alternative isn't a world-class human coach. It's a PDF. A static plan that doesn't know them, doesn't adapt to their life, and can't prevent the training errors that cause injuries and derail goals.

Against that baseline, AI coaching that genuinely adapts, that reads your data, understands your context, adjusts your training in real time, and maintains a relationship with you across months of preparation, isn't a marginal improvement. It's a category shift. It's the difference between navigating with a paper map and navigating with GPS. The paper map isn't wrong. It just can't see the traffic. 

The coaching gap has persisted for decades because the solution required something that didn't exist: the ability to deliver expert-level, individualised judgment at scale, continuously, at a cost that ordinary athletes can afford. The static plan was the best approximation we had. It was always a compromise.

The technology to close that gap is now real. Not theoretical, not five years away, not dependent on hardware that doesn't exist yet. Real, and improving rapidly. For the first time, the 95% can access something that genuinely resembles coaching: attentive, responsive, individual, and always in service of the athlete.

The era of the PDF is ending. What comes next looks a lot more like what coaching was always supposed to be.

Zepho is adaptive coaching for serious runners. It's live and available now on the App Store (desktop) & (mobile).