Koop AI, Unboxed

We reveal how Koop AI turns 25 years of coaching into an adaptive, terrain-aware training system for trail and ultra runners. We share why we built it, what it does today, where it falls short, and how we’re improving it with real race data.

• why serious runners wanted expert validation without elite pricing
• the three pillars of coaching we modeled: prescribe, analyze, communicate
• onboarding that pulls real history, constraints, and race demands
• dynamic plan changes for travel, missed runs, and time budget shifts
• NGP-based feedback for trail intensity instead of raw pace
• race specificity: course demands, heat, altitude and future prompts
• features live now: chatbot tied to Koop’s content, calendar, swaps, block moves, cross-training, volume trends
• limits we own: early-stage feedback, UI/UX, depth of “why,” and NGP-driven plan adjustments still shipping
• how we validate against coach standards and major races like UTMB
• long-term mission to make top coaching accessible to everyone

Sign up: https://www.koopai-endurance.com/app/

Episode Transcript

Jason Koop (00:12):
Hello members, what is going on?
Welcome to another episode ofthe Koop Cast.
As always, I'm your host, CoachJason Koop, and this episode of
the Public Half is a specialone because it is the moral
public public coaching productthat I have been involved with
and has been under developmentfor over two years, and that is

(00:34):
Koop AI.
So, what is Koop AI?
Simply put, this is my brainand public.
It's an artificial intelligencecoaching product that takes all
of my knowledge and all of myexpertise that I've garnered
over the last 25 years of mycareer and it makes it
accessible to anyone through anapp that we've developed.
Let's face it, artificialintelligence is everywhere.

(00:55):
Even big apps like Strava aretrying to put it into their
product so that athletes canaccess it.
And this is my humble attemptto do so with a very specific
goal of essentially mimickingwhat I do from a coaching
standpoint on a day-to-daybasis.
We've been working on this forover two years and have a core
group of users.
And I have to say that theresults that we've got from

(01:15):
these users are quiteencouraging and something that
I've been extremely impressedwith.
As an example, most recently,nine of our test users
participated in the most recenteditions of all the UTB races.
Seven out of ten peoplefinished their races, and we
actually had one person who waspreparing for the most recent
trail world championships, theworld championships using this

(01:38):
product as the coaching platformthat they were receiving all of
their advice and all of theirtraining community.
This product really excels attraining prescription and
modifying the trainingprescription based on whatever
life throws your way.
It's extremely accurate inrepresenting what I would do
with actual athletes, so much soit's actually kind of scary.

(02:00):
I'm looking at it and I'm like,that's exactly what I would do
in that situation based off ofthis athlete's particular set of
circumstances.
Needless to say, I wouldn't puta product out in the world that
I didn't believe in, and I'mextremely bullish not only on
the technology as a whole, butthis product for its stage of
development, and it's only goingto get better from here.

(02:21):
So on the podcast today, toilluminate this product even
further is the founder of theproduct, Binette Mansia, who is
instrumental in me adopting thetechnology.
One of our Crack CTS coaches,Ryan Anderson, who you guys have
recognized from a number ofpodcasts.
He has been instrumental inhelping bring this product to
life.
We go through all of the bitsand pieces of the actual

(02:43):
product.
I'll let the conversation speakfor itself.
In the show notes is where youcan actually sign up.
And this is available toeverybody right now with the low
price of $12 a month.
You can access my entire brain,which is still, once again, I'm
saying it right now and it'sstill kind of scary.
Coming to the conversation, I'mgonna get running in the way.

(03:04):
Here's my conversation withBenem C and Ryan Anderson all
about the world of my lunch andthe development behind the
scenes of our brand new coachingproduct, Koop A.
We could start the podcast outwith that good big brain.

(03:25):
Okay, so just to set the uh setthe stage for everybody, we're
in Chamony, France again, alltogether.
One big happy family in advanceof uh UTMB 2025.
We have no idea what's going tohappen, and this is not gonna
try to predict what's gonnahappen.
In fact, this podcast is gonnacome out many weeks after UTMB,
so the dust will all be settledthere.

(03:46):
We are not here to talk aboutUTMB, we're here to talk about
something that um I was going tosay that I've been working on,
but I really haven't been doingall that much.
Um, uh, that's been in theworks for a couple of years now.
Uh in the studio with me today,in the studio is a basement of

(04:07):
a chalet in Chemini, France.
I have my good friends Binettand Ryan Anderson.
Ryan's been on the podcastbefore, one of our CTS coaches.
But many of you guys are notfamiliar with Bennett.
And so just to start out with,before we start to talk about
this project that we're gonnakind of unveil to the broader
world, Bennett, who who are you?

Benat Mencia (04:31):
Great question.
Um well, I I met you two yearsago through a common friend.
I am a theoretical physicistthat transitioned to AI uh about
eight years ago.
And um I started off doingresearch in AI and then
transitioned to the industryworld.

(04:52):
Um and yeah, two years ago,yeah, this project arrived to my
life.
It started as a side project,and the whole thing gained lots
of momentum.
So I've been yeah, focused oncoupe AI for the past couple of
years.

Jason Koop (05:10):
So, as the name would indicate, coupe AI is me
as an artificial intelligencebeing, and we're gonna describe
a little bit more about that uhkind of like going forward and
how we're all piecing ittogether with coaching and math
and AI and and what the outcomeof all of this is.
But just to give the listenersa even a even a bigger backdrop

(05:34):
than that, and I don't thinkeither of you know this origin
story, so it'll be kind of likenews to you, and you can laugh
about it or whatever afterwards.
But one of the primary drivingforces of why I started to get
so incredibly interested increating an AI coaching product,

(05:55):
and I didn't know what it wasgoing to be at the time, was I
was actually a featured guest ata business coaching summit.
So I happen to have a mutualcolleague that has a really big
business that's multiple timesthe size of CTS, and they're and
they have a they have acoaching service that's

(06:16):
analogous to what we do inultramarathon coaching and
cycling coaching, but they do onthe business side.
So they work with C-levelemployees across Fortune 500
companies and things like that,and they coach them to be better
C level employees and betterleaders and better managers and
things like that.
And one of the things that cameup in this coaching conference,
it was several years ago now,was how artificial intelligence

(06:40):
is going to present a threat tothe kind of the broader
employment space, but morespecific to them, the sea level
and the management space.
And the overall tone that I gotfrom that group of people was
one of threat and fear.
You know, the big bad machinesare gonna take over and replace
all of our jobs, and they'llnever do as good of a job as we

(07:03):
can as humans and things likethat.
And I've learned that over thecourse of my career that you
can't fight technology.
The technology is gonna come,whether you like it or not, and
it's gonna do some things reallywell, and it's gonna do other
things not so well, and peoplewill adapt to whatever new
iteration of technology thereis.
There that was that way whenthe internet came about.

(07:26):
That was that was the same waywhen onboard cycling power
meters started to come around,and I saw a lot of flavors of
those technological iterationsand this new AI wave that was
kind of coming through.
And it was punctuated by thiskind of like threatening
environment that this group ofvery successful business coaches
actually had.

(07:46):
And it was that moment in timewhere I said, I'm gonna figure
out a way to play in this pool.
I don't know left from right orup from down in this pool.
I, you know, I've heard aboutit just as much as anybody else.
And so I started to sewtogether the connections to try
to make it work.
I, you know, networked like Ialways do.

(08:07):
I kind of hit the pavement, Italked to people, I found
people, I, you know, gotdifferent directions from
people.
Some of our, uh some of the uhpeople are here in Chamony,
France, with all of our mutualcolleagues and partners over at
Training Peaks.
So I had a lot of counsel in alot of different areas, and that
winding trail eventually led meto you, Binett, through really
serendipitous circumstances, andyou happen to be the right

(08:30):
person in the right circumstancewith the right background and
the right education at the righttime that was sort of
interested in trail running andendurance sports and things like
that.
And now we're kind of finallyturning, you know, kind of like
turning around to it.
But um I guess my point withthat is it's been a really long,
windy, like four or five roads,four or five year road to kind
of like get to this point wherewe've got this product that

(08:52):
actually now has a form and ashape and delivers something
that it was actually that we'vekind of cooked up for to to to
intend to deliver, which isreally cool.
So with that as a backdrop,really long, kind of like
long-winded backdrop, we're nowhere to kind of reveal what

(09:12):
this, what this actually is, andthen Ryan, how you finally get
involved kind of from thecoaching side of things.
So I think the first thing isis like outside of us just like
Binette, you and I just likephysically meeting and like
shaking hands and kind of likefiguring out why don't you take
the listeners through the firstpart of the iteration of what

(09:34):
eventually has been formed?
Like the first few times thatyou and I kind of like sat down
and said, Okay, what can thisactually be?
What did that like look likefor you?
And how did you start to figureout how to like solve
ultimately what is, and you candescribe this however you want
to, but what uh how I'vedescribed it is putting my brain

(09:55):
in a technological box thatother people can access.
And my brain being theknowledge base, the programming,
kind of everything that I'veacquired from a coaching
perspective over the course ofthe last 25 years, putting that
in a technologically accessiblebeing or entity or thing that

(10:16):
people can that kind of likepeople can access.
Take us through like the verybeginning of it.
Like, what did we actually gothrough during the formulation
of this to start to start topiece it together?

Benat Mencia (10:28):
So one of the first things we did was to try
to talk to runners um to try tofigure out whether there is a
need or a pain point out therein the running community uh that
we could help with.
And so we talked to about 50runners back in May 2024, and

(10:53):
what we realized is that thereis a group of runners out there
that take running very seriouslyin the sense that they prepare
hard races, follow structuredtraining, they take the time to
read books and listen topodcasts, etc., to you know, be
in good shape to prepare theirtraining plans.
And these runners wouldcomplain about two things.
One, the time and energy ittakes to do this well, i.e., the

(11:16):
time it takes to read books,keep up with the information
that keeps coming out, listen topodcasts, um, and also they
complain about the fact thatthey tried their best uh trying
to prepare the training plans,but then they would miss the
validation of the expert.
Um so when we asked them, well,why don't you hire a coach to

(11:36):
do this, that's exactly what acoach does for you, they would
say that it's too expensive tobe coached by a top coach.
So that was yeah, that was avery, very clear outcome of of
that first round ofcommunication with with runners,
and so we decided that weshould uh focus on that and and

(11:56):
build a product that would be asolution to that problem, i.e.
give access to everyone to topcoaching.
Um so that's how it allstarted.
Um I guess then we given that'swhat we wanted to achieve, we
had to decide okay, how are wegonna achieve that?
And what this has been aboutfor the past two years mainly is

(12:21):
about building a mathematicaltheory of your methodology um so
that we could then put it in acomputer.
And uh this happens to be apretty challenging task um uh
because well it's it's easy, youknow, taking a specific case,

(12:46):
working with a specific athleteand and figuring out what's
what's best for this athleteaccording to your knowledge and
your methodology, but buildinguh an app that can potentially
be used by millions of usersrequire capturing the the
general the general shape ofyour methodology, i.e.

(13:07):
finding the complete set ofmathematical rules that would
apply to every single case, notonly the mainstream 60-70% of
the cases.
And so that's that's has beenour focus uh for the past for
the past two years.
We are that part of the app isquite mature now.
Um so we will start focusing onother aspects uh that are let's

(13:33):
say less this is the core ofthis product, this is the the
main value of this product basedon what we hear from our users.
But of course, um once the themain the main part of the
product is built, one needs toalso address other things like
user experience, user interface,etc etc.
which which also matters um forfor most users.

Jason Koop (13:56):
So the the way that I've always viewed this is
there's uh three primary thingsthat I do as a coach and that
any coach really does.
We prescribe training, meaning,hey, let's go out and run two
hours on Tuesday.
Maybe there's some intensitystructured around it a couple of
times a week.
Training is also how the macroview of training looks over the

(14:18):
course of nine months or 12months or maybe 18 or even 24
months.
But we're prescribing training.
That's one fundamental thingthing that we do.
Second thing is analyzing thattraining.
So seeing what's coming downthe pipeline.
Okay, are you better, worse, orthe same than you were
yesterday, the day before, themonth before that, and how does
that impact what we have what wetheorized we were going to

(14:41):
prescribe next week and nextmonth and things like that.
So prescribing training,evaluating training.
And the third piece iscommunicating with the athlete.
And there's a lot of differentlayers to that.
Some of that is technicalcommunication.
Hey, your intervals were betterthan they were last week, or
hey, your workload is higherthan it was last month.
Some of it is emotivecommunication, which is what I'm

(15:01):
doing a lot of this week,right?
Hey, your training's goinggreat, you're gonna crush it,
you know, on Friday, like thatkind of, you know, that that
kind of stuff.
But there's a communicationpiece.
I think the way that I've kindof viewed the first part of this
project is mathematicallysolving for those first two
pieces, right?
So looking at all of thedifferent scenarios and kind of
coming up with you you keepsaying a mathematical model,

(15:24):
I'll say an extremely elaboraterule set um uh of which any
runner can kind of like look atit and come out with an outcome
that if it meets its merit, it'swithin 95% of what I would
actually prescribe.
It's within the human error ofwhat I would, what I would
actually uh what I wouldactually prescribe.

(15:46):
And then taking that trainingand looking at it and coming up
with uh coming up with a uh away to mathematically solve for
the second piece of it, which isanalyze the training.
Are you better, worse, or thesame as compared to your last
week in that and last month?
The fair fair assessment of forhow I kind of view it from a
coaching perspective versus uhversus the customer perspective.

Benat Mencia (16:09):
Yeah.
Yeah, I think I would describeit quite similarly.
Actually, I you know, when Ifind myself explaining to others
what Koop AI is, I say it's gotthree parts.
One is the chatbot that isconnected to your content, which
allows runners to kind of quoteunquote talk to your content.
Then there is the training plangenerator, which prescribes

(16:31):
training, um which is fullytailored and is dynamically
dynamically adjusted based onhow training is going.
And then the third bit isfeedback on training, um, which
is something that we recentlydeployed.
We've been working on thisright now, and I have been
working on this for the pastcouple of months.
It's it was very clear, it'sbeen very clear, users have been

(16:54):
very outspoken about the needfor feedback.
And so we finally deployed,let's say, version zero, a
simpler form of feedback, whichum analyzes the NGP of of the
run and gives feedback on NGP,that is whether the intensity of
the run or the intensity of theintervals on hard workout days

(17:15):
was the right intensity, orwhether it should be higher,
lower, etc.
etc.
So, yeah, I I would say thatthat's that's a a fair
description.
I mean, here what we want to dois we want the the user of
Koopai to get as you say 95% ofwhat they would get if they were

(17:35):
coached by you.
Of course, we're missing thehuman component here, and and it
remains to be seen howimportant that is, the absence
of there being a human on theother side, and actually that
that remains an open quest openquestion to me.
But I believe that everythingto do with the exchange of
information between coach andathlete, I believe that we can

(17:58):
capture that with high accuracyin finite time, and and we will
do that with Coupei.

Jason Koop (18:03):
Wow, that's a pretty incredible statement.
I was kind of I was rememberingabout like this first iteration
of like the fear of you knowmachines are gonna take over or
whatever.
I very distinctly remember inone of these meetings, I'm like,
listen, I'm not afraid of beingobsolete.
If I can build something thatmakes myself obsolete, that's a
win for everybody.
That's a win for other people,that's a win for me, like that's

(18:25):
a win for whoever else we cankind of bring under the
umbrella.
So I'm not I'm not scared.
So you keep mentioning we.
So let's kind of go over likewho's a critical part of the
team.
And once again, to the theme ofme being obsolete, I have been
only very kind of likeinterestingly enough, kind of
like loosely involved in thedevelopment.

(18:46):
You can describe like who's allinvolved, and we can bring Ryan
in the conversation becausehe's been one of the key players
in how this whole thingdeveloped and putting some like
coaching, just some real-timelike coaching checks and
balances on things.

Benat Mencia (18:59):
So, you want me to go through who has been
involved in this project?
Yeah, so uh I would say on theon the technical side it's been
mostly me full-time for for thepast couple of years, but there
have been three other friendswho have also developed part of
the app.
Those are Andy Andy Highsmith,Jonas Narkelunas, and Gasper who

(19:24):
is just starting now.
So that's on the technicalside, and on the coaching side,
it's been Ryan Anderson who ishere with us today, and then you
yourself, Jason Koop.
Um so yeah, so but I would sayit's been mostly Ryan and me, me
on the technical side, Ryan uhbuilding theory that then I

(19:44):
would I would put on a computer,and also you you've been also
you've been involved in the inthe theory building, especially
in the first 10 months of thisproject, you kind of build the
backbone, the the base of of thewhole of the whole training
theory.
And then Ryan joined joined usabout a year ago and has been

(20:05):
building on top of what you hadbuilt initially.

Jason Koop (20:08):
So, Ryan, how would you describe like your specific
role?
Because the the way that I,when Bennett and I were
organizing this, one of thethings that I very quickly
recognized is that I was gonnalog jam things.
I I work with you know a lot ofathletes.
I'm somewhat hard to get a holdof and get responses to, and

(20:28):
Bennett is very fast and numbleand needs a response like right
then and there.
And I needed to bring insomebody that I trusted a lot,
that also knew the backbone thatyou just mentioned intimately
well, and you you kind of likefit that profile.
And I kind of went into theproject kind of describing it
just in just in that way.
But now that you've like hadyour hands dirty with it for

(20:50):
over a year, like how would youdescribe to the listeners like
what you've actually been doingkind of behind like underneath
the hood, so to speak?

Ryne Anderson (20:57):
Yeah, it's applying the art of coaching and
answering all that well, itdepends questions, because
that's always the can coachanswer to anything.
Because it it truly does.
It always depends.
But how do you how do you comeup with these rules that can be
applied to uh this format ofdelivering plants to people?

Jason Koop (21:19):
It depends on, which is which I always like force
the coaches to think about.

Ryne Anderson (21:23):
It depends on you can't just say it depends and
then pull something out ofnowhere or leave it at that.
Like, no, based on yourexperience and expertise, it
depends on, and then there areanswers, models that then define
what is the next best thing todo.
Um, and so if I was to explainvery simply what I've been

(21:45):
involved with is that coming upwith like the it depends.
And so you you created the thetheory of it all of like, okay,
this is how the different typesof intensities are gonna flow.
Um, this is how volume is gonnabe modulated based on
intensities, and then coming upwith all the contingencies that

(22:08):
come into play of like, well, ifwe have eight weeks and it's a
50k race, we're gonna do thisblock to this block and and all
these different rules.
Um and probably the the partI'm most proud of of working
with this is like making itspecific to the athlete.
It's not like somebody goes inand says, Um I'm a beginner

(22:33):
based on this rubric I see.
I'm doing a 50K in eight weeksplug and play.
It is not that it's they theyhave to input or we have a way
to pull out their past trainingdata and find their volume, find
consistency or lack thereof,and then meet the athlete where
they're at because that's what ahuman does.

(22:53):
So a good human coach is gonnado that always.
Meet the athlete where they areat first.
And one of the first things youtold me, Koop, is answer and
solve all these as you would asa coach.
It was like, don't think toohard of like, well, this is how
um it needs to be programmed orthink of this contingency.
It's like, no, just first andforemost, you have a problem.

(23:16):
How do you solve it as a coach?
And then we'll massage that andwork through it and figure out
how to apply it from the onesand zeros side of things.

Jason Koop (23:25):
Um that's that's the way that I've always like tried
to communicate with you,Bennett, about the project.
Is this like I don't know shitabout shit in your world?
This is how I would alwaysstart out the my answer with
this.
This is how I would do this asa coach, and then kind of like
walk through, walk you throughthe scenario, and then let you

(23:45):
figure out how to take thatstrategy and operations and put
it into put it into a electronicand mathematical package, so to
speak, right?
And I I think one thing that'sthat that I think the listeners
should probably come toappreciate is this isn't just go

(24:08):
forward programming that we'recoming up with.
So to your point, like itactually takes an athlete's
history and analyzes it like wewould as a coach.
How much volume were you doing,how much intensity were you
doing, when were you training?
Exactly.
Yeah, like all those thingsthat we would go through as a
coach gets taken into accountbefore day one, workout one

(24:31):
actually gets programmed.
So I'm wondering if you canlike uh briefly synopsize the
run of show as the user wouldexperience it.
New person signs up, what arethey gonna what are they gonna
go through, and then whatultimately is the thing that
they get to interact with?

Benat Mencia (24:52):
So when user logs in for the first time, the first
thing is to go throughonboarding.
And the onboarding formcontains all the relevant
questions, i.e., it's the job ofthe onboarding to gather all
the relevant variables to getthe athlete started.
Um and so this is a long listof questions, and uh these

(25:15):
questions are to do with thevolume they've been doing in the
past two months, with theconsistency uh they've had in
the previous two months, withthe intensity, where they've
been doing intensityconsistently or not, whether
this led to fatigue or injury,um it's to do with Well, this is

(25:36):
okay, it's a with the past,that's that's the most important
thing, what they've been doingin terms of running.
And then there is everythingelse to do with the feature,
like um when do you like to doyour longer runs?
When do you like to take yourrest days?
What's your daily time budgetfor training?
Um what's what are the racesyou have in mind for for the

(25:56):
coming for A race?

Ryne Anderson (25:58):
Is it an A race?
Yes.

Benat Mencia (26:01):
Um Yeah, so these are these are the key elements.
And again, the the way, as withany other part of this app, the
way we designed this onboardingform was by asking you and
Ryan, all right, so what do youtalk about on your first day
when you first meet an athlete?
What are the relevant variablesyou need to gather to get the
athlete started with an initialtraining plan that will then of

(26:23):
course be adjusted dynamicallybased on everything that happens
throughout the journey?
Um okay, so that that's thestarting point, and after that,
the first, let's say the firstversion of the calendar is
generated, what we've beencalling the static plan.
Um and yeah, the athlete istold what to do every day.
We've we've put all theinformation that athletes have

(26:46):
requested.
There to be no ambiguities interms of what needs to be done
every day.
Um and then the athlete we askrunners to provide feedback on
training, uh both objective andsubjective.
There is integration with theStrava.
Down the road, there will beintegration with all the
watches, of course, but at themoment we found a quick hack
that did the job to gather theobjective numbers.

(27:09):
And there is a whole set ofrules, what we've been calling
the dynamic theory that adjuststhe training plan based on how
training is going.
Actually, you, Koop, you werevery involved in the dynamic
theory, so it's been mostlydefined by you actually.

Jason Koop (27:24):
Um I've got a question on that.
How often do you find thedynamic theory actually kicking
in and like supplanting whateverthe static theory had
initially?
Like how much is it actuallyadjusting?
I wonder if you can kind oflike encapsulate that at all.

Benat Mencia (27:41):
Um the answer is I don't know.
I I wouldn't be able to saywhether it happens a lot or not.
Um what I what I will say isthat based on conversations with
with runners, uh there seems tobe the perception that the

(28:01):
dynamic theory is not reallythere.
So I'm guessing that it doesn'thappen too often.
So so every time there is uhsomeone reaches out asking what
is the dynamic theory, I I needto remind them what the actual
set of rules is and say, well,in your case, there hasn't been
none of these rules have beentriggered because your training

(28:21):
pattern has been this or that.
Um we have a pretty compliantinitial group.

Jason Koop (28:26):
Let's just put it that way.
I mean I think it's worth sayingthat like the initial people
that have come on board, whichwhat's the number on that right
now, Bennett?

Benat Mencia (28:37):
So at the moment we have eighty paying customers,
out of which about fifty-fiveare active users.
That is, users that are usingthe app on a daily basis,
checking the calendar, givingfeedback on training.
Um yeah, those are the numbersnow.

Jason Koop (28:49):
And those were I'm gonna use the word curated.
But they were curated to theextent to where like we didn't
blast this out to everybody.
We kind of used some selectivelike it's not like we were hand
choosing them, but we used someselective channels to acquire
those people.

Benat Mencia (29:06):
Yes, that's right.
It's true for about half of ofthese people.
Um so we first gave the app to10 runners about a year ago, and
that was a hand-picked group.
Uh, we then extended the groupin October, so we were working
with 20 runners uh at thatpoint.
And then for the first timeever, at the end of January, we

(29:30):
well, you posted on Instagram,and that brought a new wave of
users.
So the current group is a mixof this initially handpicked
group of runners and people thatjoin afterwards.
Something I wanted to add onthe dynamic theory side though
is that well it remains to youknow, a lot of work remains to
be done in this front, andspecifically one we've been

(29:52):
talking about in it's becomepart of the existing theory.
It's just that it's notimplemented yet, is the
adjustments.
coming from NGP.
So the initial dynamic theorycontained well uh a set of rules
then or so um but one that uhwe've defined recently with Ryan
well it's been defined by Ryanactually is how does your plan

(30:15):
change based on NGP?

Ryne Anderson (30:19):
It's based on like NGP normalized great pace
um that's from training peaks ifyou just use Strava that'd be
gap great adjusted pace umbecause there's other AI
coaching programs out there forroad and it would just use pace.
But for trails, you go uphill,you go downhill if you take raw
pace, you're gonna get reallywhack feedback and adjustments

(30:44):
because you go do a six hourrun, you have a 15 minute raw
pace, but that could have been anormalized grade pace of 1130
and you're in in zone as we'vedetermined it so it wouldn't
adjust or whatever.
So when he's saying NGPnormalize grade pace that's
essential for our user basetrail and ultra runners because
we have to take into account theelevation change.

(31:05):
Yeah.

Benat Mencia (31:06):
Yeah so what I wanted to say is that we
deployed the first version offeedback on training which is
feedback on NGP about a monthago but we have not implemented
yet the rules that will adjustyour plan based on NGP that is
say you're on your first weekwithin your block and it's
expected for you to be at thetop of your zone on your hard

(31:27):
workout days on the intervalsfor example so if you are not
there then the theory wouldprescribe your intensity
throughout the block to be to bechanged, to be lowered one
notch let's say etc etc so thisis to say this could let
everyone know that the dynamictheory also the static part but
then now we're talking now thatwe're talking about the dynamic

(31:49):
theory um this will keepincreasing and will keep adding
nuance and and so the perceptionof it being fairly static will
will change.
I think it will change a lotspecifically with with this
feature because it's verymanifest as in you when you get
feedback on NGP it's verymanifest whether you were in the

(32:09):
right zone or you were not inthe right zone and it's and it
will be very clear to see howyour plan changes according to
this.

Ryne Anderson (32:16):
And it's going to integrate the subjective
feedback as well somehow.
Because again it all comes backto how do you solve this
problem as a coach.
And it's not just the data it'sthe subjective feedback.
And in terms of that thedynamic part with the static
plan, they can go in there andchange their time budget or
whatever and then boom boom boomboom boom it backfills and

(32:37):
corrects it.
So that's a very dynamic partof it all that's a critical part
of a coach of oh I've got thiswork weekend I wasn't planning.
How do I adjust?
Like that's the dynamic theoryyou're talking about now, which
is like you're very much in thecave trying to figure out is
utilizing their uh data and howthey're performing on the runs.

(33:00):
But a key part that's alreadypretty damn good is they can go
in and edit their time budgettravel or whatever and boom it
fixes it.
They don't have to be left inthe dark of I'm not gonna get
that run in, even though it's myfour hour day, blah, blah,
blah.

Jason Koop (33:18):
And all of that's based off of just just to
reiterate, all that's based offof what we would actually do, me
being Ryan and I, like lookingat that situation, really common
one, which is just one ofseveral dozen, I'm gonna miss
this weekend due to my kids'soccer game, or I got sick, or I
ran out of time, or I didn'twant to I wanted to watch

(33:39):
football instead of instead ofgo out and run.
Like those types of trainingadjustments happen all the time
as a real coach.
And seemingly you have infinitenumbers of ways that you can
solve for those.
You can simply let the time go,you can add it to the next
week, you can add it tosomething four weeks down the

(34:00):
road.
The adjustment piece of it hastried to encapsulate and I agree
with you Ryan does a reallygood job of encapsulating what
we would actually do based onwhat is going on with the
athlete at the time.
It's taking our coachingworkflow and then putting it
into a system that would thatwould essentially automate it
and does it with kind ofremarkable accuracy.

(34:22):
Okay, so what is the what doesit look like?
So describe the user interfacetoday.

Benat Mencia (34:34):
So you've got the chatbot that you can talk to and
you've got the trainingcalendar where you see what
what's prescribed for for thecoming 14 days actually that's
that's what we have at themoment and um then there are all
these features that revolvearound the training plan like

(34:58):
well we just mentioned one theability to change your daily
time budgets but there is a longlist of features there that
that revolve around the plan andwhich is by the way I would
like to say based on ourconversations with runners this
seems to be the most importantpart of of the app you know
runners care about being given atraining plan that will extract

(35:21):
the best of their fitnesspotential so so yeah all these
other features revolve aroundthe training plan itself and and
these features are the theability to to swap days, the
ability to move your recoveryblocks say you're traveling and
you won't be able to to run asmuch volume as you were

(35:42):
prescribed so so you you're ableto move your blocks so that
there is no volume wasted.
But then we have um you can addcross-training that's a very
another recent feature we'vedeployed.
You can see the evolution ofyour volume throughout the

(36:04):
months and see why your volumeis being modulated the the way
it is based on where you'recoming from and based on what
your targets are for for for thecoming foreseeable feature.

Jason Koop (36:31):
Based on the race that they're eventually
choosing.

Ryne Anderson (36:33):
That's another thing it's again you don't just
put in a 50k a hundred milerit's the hundred miler this
hundred miler it's this hundredmiler and we'll probably get to
a point of like these hundredmilers you must do heat
acclimation for sort of likealtitude considerations where it
gets flagged and then chatbotgoes in there and say if you're

(36:56):
gonna brace this you either needto do the day before seven days
out blah blah blah again it'sit's very individualized and
it's not just very basic thingslike I've already said if it's
it's just this race.
It's this race and what are thedemands of it.
Because again, how do we solvethis problem as a coach?
We're digging in to find thesethings to serve the athlete best

(37:18):
and be specific to the race andthen with their training
history and how much they cantrain.

Benat Mencia (37:24):
Yeah.
Something I would like to addis that today we are where we
are and there is a very longlist of limitations with this
app not only on the userexperience side but also on the
nuance to do with trainingtheory.
But this product will never endits development process as in

(37:46):
we have a system to prioritizewhat to build first.
So there is a roadmap builtwhich gets gets adjusted uh
based on the feedback we obtainfrom runners but it's a never
ending project and so there willbe the day we will arrive today

(38:06):
where when I'm completelyobsolete the app is absolutely
perfect.
There's no such thing as okayyou know we've got distant tasks
and after we've done thisproduct is finished and now we
can all chill.
No like best products bestproducts never end uh developing

(38:26):
and this will be such a productthat for as long as we are
alive we'll keep evolving.
So we are where we are todayit's still an early stage even
though it's been a couple ofyears but we do what we can with
resources we have we move asfast as we can given the
resources we have but yeah likeif I had to put numbers in terms
of how much we've done out ofeverything that needs to be done

(38:49):
for it to be the perfect appfor runners I would be under
five percent of that it has beendone um so so I just want to be
very clear that this we won'tstop working until it's not and
it's not close.
Exactly because even even if wecould define now what the

(39:11):
perfect running app looks likenew technology keeps coming out
new information keeps coming outnew science keeps coming out
and we want to be connected toto how the times change and and
so we will always be able toimprove the product for our
runners and that's what we careabout.
It's our mission to to make topcoaching accessible to everyone

(39:37):
and so yeah this is an infinitetask in that sense.

Jason Koop (39:50):
Mainly from the perspective of who are the
really good candidates for thisin terms of their like athletic
profile and what they'retraining for and then who who
and what athletic profiles maybeare not good candidates for it,
that there are too manyexceptions or they wouldn't come
through the initial filter orwhatever.
But before I want to get intothat I just thought of a really

(40:11):
great thought experiment.
So one of the primary ways thatI let me back up but just a
little bit what one of one of myfundamental roles at CTS is not
only coaching uh coachingathletes but also recommending
our coaches to athletes that arecoming in we have an athlete

(40:31):
services team that's kind ofdedicated to doing that but I
also help that athlete servicesteam out as well as people just
reach out to me directly and sayhey Koop what would you kind of
recommend for this I'vedeveloped that recommendation
kind of knowledge base throughour educ through our continuum
education system, our internalcontinuum education and open

(40:53):
office system and primarilythrough our schedule review
process, which we've talkedabout many times on this
podcast.
A coach throws up a trainingplan that they're going to it's
kind of like feared and you knowloathed and loved and
everything kind of like all atonce.
But uh it's it's it's itdefinitely has evolved to this

(41:14):
thing that we do routinely inthe office where our coaching
director Cliff Pittman will pickout one of our coaches and say
hey I want you to do a schedulereview based on whatever's
coming up sometimes it's basedon a race that just happened
sometimes it's based on whateveris kind of like going on in the
social media space.
Ryan I want you to find anathlete that's kind of like this
throw up a schedule and we'llkind of talk about it.

(41:36):
And it's when I get to viewthose, it's through that lens
that I can very keen I'm verykeenly aware of where that coach
is really strong, where theyneed to develop a little bit
more, how they communicate, whatthey think about load
management and stuff like that.
And it kind of gives me a youknow a scoring system
essentially based on you knowhow they're kind of how they're

(41:57):
doing with it.
But I guess my point with thatis is the I've been able to
recommend coaches to specificathletes as well as recommend
how they need to develop throughthis schedule review mechanism.
And you've been through a lotof that Ryan you actually do a
great job with that.
That's one of the reasonsyou're involved with this

(42:17):
project how would coopai stackup in that environment like if
we went through that processwith the schedule that Koop AI
generated and we know theathlete profile coupe AI
develops the schedule what likewhat do you think that would be
what would the banter be likebased off of what we normally do

(42:38):
in our coaching environment whowould be the first one to say I
think you're doing everythingwrong which I've done with a
couple of our coaches over thecourse of the year.

Ryne Anderson (42:52):
But I guess what I'm trying to say is it's like
how like is it how re howrealistic of it is approximation
for what we would actually putthrough a lot of scrutiny I
think yeah if if you were to seethe plan all of our coaches
would recognize the flow of theintensities how the volume is
modulated and everything.
The workout vocabulary yes theworkout vocabulary all that like

(43:16):
if you're if you're taking anathlete that is straightforward
so to speak and that we've got12 weeks we're not coming back
from injury we're not dealingwith anything uh crazy travel
wise or like if it's 12 weeks tothe race it it will input based

(43:43):
on the things specific to theathlete and the set of rules we
have for the distance boomhonestly like I don't know you
have to really split hairs tolook at it and say like hmm like
why did you do that?
Why did you do this?
Maybe you could nitpick thevolume modulation and like oh
this person can handle more orthey're they're gonna be a 30

(44:04):
hour finisher do you reallythink they should be doing
steady state intervals they justneed to do more volume like you
can nitpick things like thatbut in terms of how it flows and
the periodizing the intensityit's what we would do and what
we should do.
Which is that's not easy to donot saying that's always right.

(44:26):
Yeah yeah yeah but it'sfollowing sound principles that
we know work.

Jason Koop (44:32):
That's not easy to do because I've always felt that
like from a coach developmentperspective it takes like 12 or
24 months just to get there.
Like when I work with coachesand I start to kind of develop
them through our educationalsystem and they have different
touch points within our companyand things like that.
It's really 24 months beforethey go into that schedule

(44:55):
review environment and I'm likeokay that's that's really
reasonable I'm only gonnanitpick 10% of this the rest of
the 90% is good.
But for the first two yearsit's like okay we need we really
like we okay let's like figurethis like not that it's like
it's really bad in my eyesbecause I have like I have a
very sharp eye for that stuff.

(45:16):
Most people wouldn't evennotice but I guess what I'm
trying to say is is is like I'veI've always observed that
whenever I've gone into theschedules and you guys have
forwarded them over to me andthings like that I look at it
through that lens of that wouldpass the muster in a schedule
review that any one of ourcoaches could pick up and say

(45:37):
yeah I I don't you know Ryanbuilt this Adam Ferdinand built
it cliff built it Darcy built itNicole built it it would kind
of have like the fingerprints ofall of that methodology kind of
alchemized into one thing youcould you could see that come
through which is exactly how itis intended to is intended to

(45:58):
flow through so I want to goback to like the limitations
piece now Bennett so this islike specifically for you
because you've been so hands-onwith the users where have you
seen some of the biggerpitfalls?
I mean we can kind of be openand honest here like this is
going to go out to a pretty bigaudience like where are some of

(46:18):
the bigger pain points in termsof like athletic profiles that
you've that that we've eitheryet to solve for or you kind of
anticipate is going to be a painpoint in the future.

Benat Mencia (46:27):
You can describe it however you want to so based
on conversations with runnersthe the most important elements
that they were missing um let'ssay three or four months ago we
sent a survey in in May andbased on this survey we learned
that runners care about not onlybeing told what to do to

(46:52):
maximize their their fitness butalso knowing why they are being
prescribed what they are beingprescribed.
They want to have this um thisconfidence that okay it's not
just a random number generator.
Exactly and also I I guessactually you're touching
something interesting here.
I guess it's also the wholeprocess of building trust

(47:13):
between the AI and the athleteso so it's important to them to
know why they are beingprescribed what they are being
prescribed so we've kind ofdeployed a version zero of that
and we're now obtaining feedbackand we'll keep iterating that
until everyone is happy andthere's enough information for
for everyone to be happy withthat.
It was important for them to beable to add cross-training many

(47:38):
of them from day one.

Ryne Anderson (47:40):
Yes many of them when we went through that
filtering process we filteredout a lot of people who like
cross training is my number onepriority I want to know how to
fit that in because we weren'tthere yet we it's like we have
to figure out how to get it tothat point just to just program
the running.
Yeah exactly um and then likethat's a big thing for people
they want to know how to fit itin we're like we've kind of
deployed first theory of thatbut yeah that's an example of a

(48:04):
big request for people.

Benat Mencia (48:06):
Yeah so this sort of big request on the most
important one I think isfeedback on training.
Runners really want to getfeedback on training because
they don't know whether whatthey are doing they are told
what to do they go do it butthen they are not sure whether
they did it correctly or not.
And so we've been working onthat for the last couple months
we've also deployed version zeroof that but still you know very

(48:27):
simple form.
Feedback on NGP and nodynamical adjustments and no
judgment on more nuanced thingsas you know whether you are at
the top or the bottom of thezone you're supposed to be in
etc etc for now we were just youknow testing the waters we've
also deployed the first versionof our NGP model in-house Coup

(48:49):
AI NGP model before runnerswould need to have a premium
training PIX account to get LAPNGP they could get in average
NGP but not LAP NGP now we'vedeployed the first version first
version of the Coup AI NGPmodel which is working well
we'll keep we'll keep monitoringand we'll keep um assessing

(49:10):
whether the behavior is theexpected behavior um but that
was a nice piece that wasmissing as well.
I would say yeah these are themost important elements based on
the last survey version zero ofthese elements have been
deployed but you know this allthese features need to be
brought to the right level suchthat everyone is happy with

(49:32):
this.
And it will be interesting tosee once we are done with these
key things that we're missingand once everyone is happy with
how this is behaving it'll beinteresting to see what's the
next thing that people careabout.
And for people always careabout what they don't have.
Right.
Scarcity.
Yeah but it's also a matter ofpriority right you know if if I

(49:54):
ask a runner how can we improveKoop AI for me probably they
won't go through everything butthey will want to focus on the
most important things they aremissing and they will tell me
all that but once we fix thatproblem then we go to the next
level of of polishing and nuanceand then and and as I said
before you know this is anever-ending process and so we'll

(50:14):
keep talking to runners on adaily basis to understand how we
can improve Koop AI for them.
So how is the feature lookinguh short term improving these
elements I just mentionedimproving feedback improving the
cross-training featureimproving the part of the app

(50:35):
that explains why you're beingprescribed what you're being
prescribed at the moment it'sfocused on volume modulation and
feed per mile modulation um butthere are other elements that
runners will also want to knowabout but this seem to be very
important.
Then there is everything to dowith UIUX.
Gasper is working on it nowUIUX being user experience yeah

(50:58):
user interference userexperience that's everything to
do with how pleasant andunpleasant it is to use the app.
So far we've been focused onwhat runners seem to care most
about as I said already 10 timeswhich is being told what to do
such that you know they will gocrush the the race.
But next level is okay not onlyyou need to be given the right

(51:18):
information but it should alsobe fun and pleasant to look at
the app and you know move aroundthe app.
So and that's clearlyimportant.
Perhaps it's not important forthe power user that you know the
power user cares so much aboutbeing prescribed training
following Koop's methodologythat you know it doesn't matter

(51:39):
whether it's ugly and painfulbecause they care about the the
core of the product but noteveryone is a power user and not
everyone is a an early adopterand so and of course we want to
serve everyone and for that weneed to also make it not only
good but pretty as well.

Jason Koop (51:57):
We're we're gonna have a somewhat of a proof point
in the next few days with theraces out here.
I mean there's how many peopleracing all the races again
Banette?

Benat Mencia (52:06):
Eight Coupé A runners will be competing in
Chamonix this week.

Jason Koop (52:09):
Yeah and that's across UTMB TDS OCC and CCC are
also UTMB yeah for for yeah sosmattering of everything and
that's not it's not like this islike the first foray anybody's
had into races we've had itbefore but that's a pretty
decent gaggle I mean I haveeight athletes across UTMB and
CCC like just me personally andvery much so my my personal

(52:35):
coaching product has a littlebit of a litmus test.
Every big race we do this afterWestern states how did our you
know success rate compare to thesuccess rate of the race and we
beat it every single year withour coaching staff we better be
doing that compared to comparedto the rest of the field.
It'll kind of be the same withthis we'll look at it how did
all eight people do how doesthat compare against the norms

(52:56):
or the averages and we can evenuse our coaching as a cross
section of that as well as acomparative group across that
and then how do we iterate fromthere and I you know I guess to
be transparent I can post theresults of that in the outro of
this podcast since we don't knowas we're recording it and then
we'll just go from there but Ido I do think that like it's
important to know like peopleare actually using this for a

(53:19):
what is for many people a oncein a lifetime race and that's I
think a really cool thing tokind of just just yeah just to
trust you're absolutely rightabsolutely that's a great way to
put it Ryan yeah yeah yeah soyou're gonna be out there
cheering these people on so thatyou have 100% finished right
yes yes 100% you were out herelast year and got your first

(53:42):
taste in like five differentpositions I remember um last
year one of the first users dida TDS I think Yonas not kill
Yunas finished on a Wednesdaywhat go see this is uh things

(54:05):
that we'll figure out um okaywe're we're gonna wrap it up
here um first off thank you guysI know it's been a lot of hard
work over the last few years andit's not like there's gonna be
you know a $20 million marketingbudget that all of a sudden
like rolls out and launches thisthing it's gonna be slow you
know slow iterations that youknow kind of creep along here

(54:26):
and there this will be anotheriteration where we'll where you
guys will onboard some peoplewith this I will have links in
the show notes for this butwhere do you want to direct
people to find out moreinformation about Koop AI to
potentially sign up for Koopaiwhere can we direct them?

Benat Mencia (54:43):
The website coupeaindurance.com and so they
can find some information thereand they can sign up there.
And of course I'm happy to haveconversations with whoever
would like to discuss this orwould like to know more about
the project so they can email meand arrange a call if they

(55:04):
would like to know more aboutthe project.

Jason Koop (55:05):
Man you're taking that on I've had other guests do
that and that's about like halfhalf and half half an hour so
let's see.

Benat Mencia (55:12):
Actually fun fun fact I realized that it helps a
lot if there is an intro callwith runners at this stage at
least where you know we don'thave a world class self-service
onboarding so um for every newrunner I always send them an
email and tell them offer themto have an intro call and this

(55:35):
is usually very useful for manyrunners.
Not everyone wants to do it butyeah happy to talk to anyone
that would like to talk to me.

Jason Koop (55:43):
When pe when you've offered that to people I don't
think I've told you this beforewhen you've offered that to
people I've had maybe four orsix people who I who I know or
I've met say is this real or isthis just some scammer like the
kid like the king of Syria likeyou know wants to wire you a
bunch of money give them yourbank account numbers like those

(56:04):
kind of old school scams theyactually who's this who's this
Binette guy is he real and areyou associated it's like he's
very he's very much real.

Benat Mencia (56:12):
Okay so we'll put it all in the show notes man uh
I'll turn the floor over to bothof you guys to wrap it up what
are you like what are youlooking forward to the mission
of of Coupe A is to build therunning app that will serve
hopefully all all runners downthe road and that yeah as I said

(56:33):
we want to give access to topcoaching to everyone without
money or accessibility being alimitation so so we'll be
working hard towards thatobjective and I have the
conviction we'll get there.
It's just a matter of sittingdown and doing it.
The technology is ripe we havethe best coaches here we just
need to sit down and execute andwe're doing that as I said I

(56:56):
think that at least 95% of thework remains to be done but
we'll keep working until we getthere.

Ryne Anderson (57:04):
I'm gonna say raise the standard raise the bar
of coaching I'm glad you talkedfor a while because I was gonna
say a much more harsh way butpeople don't know what they
don't know and they they get apre-made plan they get their
buddy to make the plan and ohthis is cool I have organization

(57:26):
but I think I think athletescan be served much much better
than they realize with either anexperience they're currently
having or they aren't aware ofwhat coaching looks like.
So I think I'm looking forwardto it raising the bar and

(57:48):
challenging other coaches tolike care about their athletes
care about their coachingpractice and deliver a service
and product that these athletesare like oh I got to this race
and I'm confident in it.
I understand why I'm doingthese things.
Yeah.

Jason Koop (58:07):
I couldn't have said it better myself.
I'm grateful to both of youguys can't wait to see how this
weekend unfolds first offbecause I know we got a lot of a
lot of really importantathletes that are racing across
all the distances but moreimportantly I can't wait to see
how this unfolds after UTMB anduh maybe we'll make a little bit
more progress on that 95%progress bar.

(58:28):
Slowly but surely all rightthank you thanks my friends
there we have a much week to theminute and Ryan you guys have

(58:49):
been amazing at pulling thistogether and the job's not done
yet and we keep integrating thisthing every single week and it
keeps getting better and betterand better and better and I
remember the first couple ofrevisions of the actual product
UI UX and left a lot to bedesired and still there's a lot
of work to do there but itreally left and left a lot to be
desired when we first wanted toplace where we can look at it

(59:11):
and go hey this does a reallygood job and cannot remark
enough description itself whichis what most people find the
most value and is remarkablyaccurate in representing what I
would do with any athlete andthat's how it was always
intended to be designed.
So once again channel mainly toBanette and Ryan for getting

(59:34):
all of this together just done aremarkable job remarkable job
with the product if you guyscheck it out links are in the
links in the show notes to it.
I'd also be remiss to not tomention that it's been a hot
minute since I've produced thepodcast and what I'm starting to
find is that as my time getsmore limited this becomes one of

(59:56):
the things that in additionInto that, let's face it, we get
our information differentlythese days, and that value
proposition is going to continueto change as technology rapidly
evolves.
And what I'm finding is tryingto squirrel away an hour or two
of my time to produce a podcast,and squirrel away an hour or

(01:00:19):
two of guest time to produce apodcast is just getting more and
more challenging, particularlybecause this is an
information-based podcast, it'snot an entertainment podcast.
I don't make any bones aboutthat.
But what I will say is I'vealways been grateful for the

(01:00:56):
opportunity to get in front ofyou guys on a somewhat frequent
basis and hopefully deliver someknowledge to everybody for the
betterment of athletes andcoaches out there.
That's not something that Itake for granted.
Um, I've always consideredmyself very fortunate to have
that type of influence and to beable to uh produce content kind

(01:01:17):
of for the betterment of theentire community.
Uh it's not lost on me.
So thank you guys.
Uh thank you, thank you, thankyou for all that and all of your
attention and all of yoursupport over the years.
I'm gonna continue to bring itin other forms of content, and
this app is just uh one uh onepiece of that.
So I'm not going anywhere, it'sjust gonna be in a much

(01:01:38):
different format.
Alright, folks, that is it fortoday.
Go check the links in the shownotes.
Hope you guys go check the appout.
Super psyched about it.
And as always, we will see youout on the trails, maybe perhaps
coached by Coupe AI.