April 2, 2025 • 20 mins
What separates AI businesses making millions from those burning out in months? In this episode of the Total Sum Game podcast, Adam Feuerstein reveals the 5 most profitable AI business models in 2025, 3 overhyped traps to avoid, and a powerful framework to help you spot sustainable opportunities in a noisy AI market.
If you're an entrepreneur, founder, or investor trying to navigate the world of AI startups, this episode is a must-listen. Discover why leading with “AI” isn’t enough, what the most successful companies are doing differently, and how to evaluate your own ideas using Adam’s MOATS Framework (Moat, Outcomes, Automation Balance, Training Data, and Scalability).
👉 Subscribe for more no-fluff conversations about AI, entrepreneurship, and building real businesses that last.
🧠 Ready to incorporate AI into your business?
Check out our courses at totalsumgame.com
🔔 Don't forget to like, comment, and subscribe for future episodes.
#AIBusiness #StartupStrategy #TotalSumGame #AI2025 #Entrepreneurship #MOATSFramework #AIStartupTips
Follow Adam Feuerstein:
YouTube: https://www.youtube.com/@adam_feuerstein/?el=youtube_organic
Facebook: https://www.facebook.com/adamfeuersteinofficial/?el=facebook_organic
TikTok: https://www.tiktok.com/@adamfeuerstein/?el=tt_organic
Instagram: https://www.instagram.com/adamfeuerstein_/?el=instagram_organic
Follow Total Sum Game:
Facebook: https://www.facebook.com/totalsumgame/?el=tsg_facebook_organic
Instagram: https://www.instagram.com/totalsumgame/?el=tsg_instagram_organic
Spotify: https://open.spotify.com/show/0hp1IXWWxZYcqr5TprXiNS/?el=spotify_organic
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:00):
The most successful AI businesses today don't
actually lead with AI in theirmarketing.
Instead, they focus on thespecific problem they solve and
the outcomes they deliver.
Welcome to the Total Sum GamePodcast.
(00:20):
I'm Adam Fierstein, and todayI'm tackling one of the most
requested topics from ourlisteners how to identify
genuinely profitable AI businessopportunities.
In a market flooded with hype,it seems like every day there's
a new revolutionary AI businessmodel being promoted as the next
gold rush, but, as we've seenover the past year, not all that
(00:44):
glitters is gold.
I've spent the last six monthsdeeply analyzing dozens of AI
business models, talking withfounders and tracking real world
performance data, and what I'vediscovered is that there are
clear patterns separating the AIbusinesses that are generating
sustainable profits from thosethat quickly fizzle out, despite
(01:04):
the initial excitement.
In today's episode, I'mbreaking down the five most
promising AI business modelsright now the three that are
most overhyped and giving you apractical framework to evaluate
any AI opportunity that comesyour way.
Let's start with a quickreality check on where we
actually are with AI businessmodels in 2025.
The market has definitelymatured compared to 2023's Wild
(01:29):
West phase, when Chad GPT firstexploded into the mainstream and
everyone was scrambling tolaunch something anything with
AI.
We're now firmly in what I'dcall the reality phase, where
the fundamentals of businesshave reasserted themselves.
Revenue unit economics andsustainable competitive
(01:52):
advantage matter again.
The days of raising millions ona deck with the words
AI-powered are largely behind us.
Investors and customers alikehave become much more
sophisticated about separatingactual value from marketing hype
.
One interesting trend I'venoticed is that many of the most
successful AI businesses todaydon't actually lead with AI in
(02:13):
their marketing.
Instead, they focus on thespecific problem they solve and
the outcomes they deliver.
The companies seeing the mosttraction aren't selling AI.
They're selling time savings,cost reduction, creative
breakthroughs or competitiveadvantages.
This is similar to whathappened with cloud companies or
mobile-first businesses inprevious tech waves.
(02:36):
Eventually, the technologybecomes table stakes and the
focus returns to fundamentalbusiness value, and the focus
returns to fundamental businessvalue.
So, with that context in mind,let's dive into the five AI
business models that aregenerating real, sustainable
profits right now.
The first model that's showingstrong performance is what I
call AI-enhanced professionalservices.
(02:57):
These are businesses that taketraditionally high-cost
professional services like legalwork, accounting or design, and
use AI to dramatically improvetheir efficiency.
What makes this model soeffective is that it combines AI
capabilities with humanexpertise.
The AI handles the routine,time-consuming aspects of the
(03:19):
work, while humans focus onjudgment strategy and client
relationships.
While humans focus on judgmentstrategy and client
relationships.
A great example is CaseText,which uses their AI tool,
cocounsel, to automate contractanalysis, but keeps experienced
lawyers involved for finalreview and strategic advice.
They've been able to reducecosts by 60% while maintaining
(03:40):
quality, which has opened uplegal services to previously
underserved market segments.
Quality, which has opened uplegal services to previously
underserved market segments.
Another excellent example isClarity, which automates
document review for legal andfinance teams.
They claim to reduce documentreview time by 85%, while
improving accuracy by 30%.
What's impressive is thatthey've managed to secure
(04:02):
enterprise clients like Coupaand Gusto.
Is that they've managed tosecure enterprise clients like
Coupa and Gusto?
The second promising model isvertical-specific AI
applications.
These are tools built to solvevery specific problems in
particular industries, ratherthan being general-purpose AI
platforms.
What makes these effective istheir deep domain knowledge.
(04:24):
What makes these effective istheir deep domain knowledge.
They're trained onindustry-specific data and solve
problems that generalist AItools simply can't address
effectively.
For instance, construct AI inthe construction industry uses
AI to analyze building plans andidentify potential code
violations and safety issues.
It's a narrow use case, but itsolves a real pain point that
(04:46):
costs the industry billionsannually.
They've reported a 78%reduction in compliance-related
delays for their clients.
The third model showing strongresults is AI-powered data
analysis and decision support.
These businesses take theoverwhelming amounts of data
companies collect and turn itinto actionable insights and
(05:08):
recommendations.
What's powerful here is thatthese tools don't just provide
analytics.
They actually help decisionmakers understand what actions
to take based on the data.
A standout example is BloomReach, which helps e-commerce
businesses optimize theirinventory and pricing in real
time based on the data.
A standout example isBloomreach, which helps
e-commerce businesses optimizetheir inventory and pricing in
real time based on dozens ofmarket variables.
(05:30):
Their clients report an average22% increase in profit margins
and a 15% increase in conversionrates.
I'm also impressed by Sky,formerly Kenshu, which helps
marketing teams optimize adspending across multiple
platforms.
They've built AI that canpredict campaign performance and
automatically reallocatebudgets to maximize ROI.
(05:53):
Their clients are seeing anaverage 31% improvement in
campaign performance.
The fourth model is AI workflowautomation.
These businesses identifycomplex multi-step processes
within organizations and use AIto streamline or completely
automate them.
The key to success here isfocusing on end-to-end processes
(06:17):
rather than just individualtasks, which creates much higher
value for customers.
Automation Anywhere is doingthis exceptionally well in the
healthcare space.
Automating the entire insuranceverification and billing
process for medical practices,they're saving some clinics over
30 hours of staff time per weekand have reduced claim denial
(06:39):
rates by up to 63%.
Another great example is UiPath, which has evolved from simple
RPA to using AI for complexworkflow automation.
They're helping companiesautomate everything from
customer onboarding to employeeoffboarding, with an average ROI
of 383% according to ForresterResearch.
(07:02):
And the fifth promising model ispersonalization at scale.
These businesses use AI tocreate customized experiences,
products or services for largecustomer bases in ways that
would be impossible manually.
What makes this modelcompelling is that it creates
value that literally couldn'texist without AI.
(07:23):
It's not just making anexisting process more efficient.
Fascinating example is DreamboxLearning, which creates
personalized learning paths forstudents based on their
individual strengths, weaknessesand learning styles.
They're seeing learningoutcomes improve by 40% compared
to standardized approaches andhave now reached over 5 million
(07:45):
students.
Another impressive example isStitch Fix, which uses AI to
personalize clothingrecommendations at scale.
Their success rate inpredicting what customers will
keep has improved by 35%, andthey've been able to reduce
returns by 20% compared toindustry averages.
Now let's talk about the threeAI business models that are
(08:08):
generating a lot of buzz butshowing concerning signs when
you look beneath the surface.
No-transcript.
The first overhyped category iswhat I call AI middlemen
businesses that essentiallyrepackage existing AI APIs from
Google, openai or Anthropic withminimal added value.
The fundamental problem withthis model is the lack of
(08:30):
defensibility.
When your core value comes fromanother company's technology
that anyone can access, you'reextremely vulnerable.
We've already seen severalwell-funded startups in this
category implode when theunderlying AI providers change
their pricing or launchedcompeting features.
A notable example is Jasper AI,which initially gained traction
(08:54):
as one of the first open AIwrappers, but has faced
increasing pressure as open AIimproved its own direct
offerings.
Another example is the flood ofchat, gpt plugins and GPT
stores apps that gained initialusers but struggled to build
sustainable businesses becausethey couldn't differentiate
beyond the underlying modelcapabilities.
(09:16):
The second overhyped model is AIcontent farms Businesses built
around generating massiveamounts of content across
thousands of websites orchannels to capture ad revenue.
While this can work in theshort term, it's not sustainable
.
Search engines and platformsare rapidly getting better at
(09:38):
identifying and penalizing aigenerated content that doesn't
provide genuine value.
I've tracked several contentfarms that saw their traffic and
revenue plummet by 80 or moreafter recent algorithm updates
specifically targeting thisapproach.
Companies like red venturesuresand Content Mills in the health
(09:58):
and finance space have beenparticularly hard hit.
Google's helpful contentupdates and EEAT standards have
been particularly effective atidentifying and demoting
AI-generated content that lacksexpertise and authenticity.
The third concerning model is AIfeature companies.
(10:19):
Startups built entirely arounda single AI capability that
should really be a featurewithin a larger product.
These businesses areparticularly vulnerable to being
made irrelevant when largerplatforms simply incorporate
similar functionality as astandard feature simply
incorporate similarfunctionality as a standard
feature.
We've seen this happenrepeatedly with AI summarization
(10:43):
tools like SumEyes, simple AIchatbots like many customer
service bots, and basic imagegeneration apps like early
versions of Lenza that brieflygained traction but couldn't
sustain themselves as standalonebusinesses.
It's the classic feature versusproduct problem, but AI has
(11:03):
accelerated the cycle offeatures being absorbed into
platforms.
Many standalone AI tools thatraised millions in 2023 are now
basically free features inMicrosoft 365, google Workspace
or other major platforms.
Based on my analysis of what'sworking and what isn't, I've
(11:23):
developed a framework to helpentrepreneurs and investors
evaluate AI businessopportunities.
I call it the MOTES framework,which stands for MOTE Outcomes
Automation, balance, training,data and Scalability.
Let's break down each component.
First, moat refers to yoursustainable competitive
(11:44):
advantage In the AI space.
This often comes fromproprietary data, unique domain
expertise or network effects,not just from using AI
technology itself.
The question to ask is if awell-funded competitor had
access to the same AI models,what would still make my
business special Companies likeDatabricks have built strong
(12:08):
moats through their combinedexpertise in data infrastructure
and AI model development.
Next is outcomes the specific,measurable results you deliver
for customers.
The strongest AI businesses canpoint to concrete improvements
in metrics that customers careabout.
For example, we reduce customerservice costs by 35% is much
(12:32):
stronger than we use advanced AIto optimize customer
interactions.
Ai to optimize customerinteractions Gong, the revenue
intelligence platform, does thisexceptionally well by
specifically quantifying howthey improve sales conversions.
The automation balance refersto how effectively you combine
(12:52):
AI automation with humanexpertise.
The most successful modelsaren't trying to eliminate
humans entirely, but ratherredefining how humans and AI
work together.
Companies like Scale AI havemastered this balance.
While they automate significantportions of data labeling, they
maintain human oversight toensure quality, resulting in
(13:13):
data sets that are 99.8%accurate.
Training data is about youraccess to unique, high-quality
data that can train AI models tosolve your specific problem
better than general purposemodels.
Samsara is a great example here.
They've collected billions ofdata points from IoT sensors in
industrial settings, giving themunique training data that
(13:36):
allows their ai to predictmaintenance issues with 92
accuracy far better than genericmodels could achieve.
And finally, scalability refersto how efficiently your
business can grow.
The best ai business modelsactually get stronger as they
scale due to network effects,improving data or decreasing
(13:57):
marginal costs.
Snowflake exemplifies this withtheir Data Cloud platform.
Their AI capabilities improveas more customers join and share
data, creating a powerfulflywheel effect where each new
customer makes the platform morevaluable for everyone.
Using this framework, I found Ican quickly assess whether an AI
(14:18):
business concept has genuinepotential or is likely to
struggle as the market matures.
To make this more concrete,let's apply the MOTES framework
to a few real-world examples.
First, let's look at an AIbusiness in the marketing space
that I've been impressed by.
Persado helps businesses createand optimize multi-channel
(14:40):
marketing campaigns usingartificial intelligence.
Promote they have proprietarydata from thousands of campaigns
across different industries,giving them insights no
competitor can match.
Their message machine hasanalyzed over 100 million
marketing messages For outcomes.
They can point to an average32% improvement in campaign ROI
(15:05):
for their customers, a metricthat directly impacts bottom
line For clients like Chase.
They've increased credit cardapplications by 47%.
Their automation balance isstrong.
They use AI for creativegeneration and optimization, but
keep experienced marketersinvolved in strategy and brand
(15:27):
alignment.
Their platform suggestsmultiple options that human
marketers can choose from andrefine their training data
advantage comes from theperformance feedback loop of all
those campaigns, whichcontinuously improves their
recommendations.
They've built language modelsspecifically trained on
(15:49):
marketing, language and consumerresponses.
And for scalability, theirtechnology platform can support
enterprises 10 times larger thantheir current client base with
minimal additional costs.
They've successfully scaledfrom working with small
businesses to global enterpriseslike Dell and Vodafone.
Now let's contrast that with anAI writing tool that recently
(16:12):
shut down.
Despite initially strong usergrowth, their moat was
essentially non-existent.
They were using OpenAI's modelswith a nice interface, but
dozens of competitors couldeasily offer the same thing.
When OpenAI improved ChatGPT'sinterface, many users simply
switched to the source.
Their outcomes weren't clearlydifferentiated.
(16:35):
They promised better contentbut couldn't quantify the
improvement over other methods.
They couldn't demonstrate aclear ROI to justify their
subscription prices.
Their automation balance wasskewed too heavily toward
replacing human writers entirely, which created quality issues
for anything beyond basiccontent.
(16:57):
Without human editorialoversight, the quality varied
wildly.
Their training data was generic, with no specialized datasets
or feedback loops to improveperformance for specific use
cases.
They were essentially passingthrough OpenAI's general models
without any domain-specificimprovements, and their
(17:18):
scalability was compromised byhigh API costs that actually
increased linearly with usage,creating margin pressure as they
grew.
They were paying almost 70% oftheir revenue to OpenAI for API
access.
When you compare these twobusinesses through the moats
framework, it becomes muchclearer why one succeeded while
the other struggled.
(17:39):
Looking ahead, I see severalemerging opportunities in the AI
business landscape that aren'tyet saturated but show strong
potential.
One area I'm particularlyexcited about is what I call AI
orchestration tools thatcoordinate multiple specialized
AI models to solve complexproblems that no single model
can handle effectively.
Companies like Langchain andFixie are pioneering this space,
(18:02):
creating platforms that cancombine the strengths of
different models and tools intocohesive workflows.
This AI of AI's approach solvesthe limitations of any single
model.
Another promising direction isembedded AI financial services,
using AI to fundamentallyreimagine lending, insurance and
(18:24):
investment products, based onmuch richer data analysis than
traditional approaches.
Upstart and lending andlemonade and insurance are early
examples, but I see much morepotential.
Upstart has shown they canreduce default rates by 75%
while increasing approval ratesfor underserved populations,
demonstrating the power of AI infinancial risk assessment.
I'm also seeing interestingdevelopments in AI for physical
(18:48):
operations, bridging the gapbetween digital intelligence and
real-world processes inmanufacturing, logistics and
physical retail.
Companies like Covariant inrobotics and Standard Cognition
in retail are showing how AI cantransform physical operations.
Covariant's robots can nowhandle over 10,000 different
(19:09):
items in warehouses with 99%accuracy tasks that were
impossible for automation just afew years ago.
For entrepreneurs listening, thekey insight is that the most
compelling opportunities oftenlie at the intersection of AI
capabilities and deep domainexpertise in a specific industry
(19:30):
or problem space.
The days of success coming fromsimply applying generic AI to
generic problems are largelybehind us.
The next wave of successful AIbusinesses will come from
founders who deeply understandboth the technology and the
domain they're working in.
To wrap up today's episode, letme summarize the key insights
(19:52):
for identifying promising AIbusiness opportunities in
today's market.
First, focus on solvingspecific high-value problems
rather than showcasing AItechnology for its own sake.
Second, build real moats beyondjust using AI.
Proprietary data, domainexpertise and network effects
are crucial.
Third, think carefully aboutthe right balance between
(20:15):
automation and human expertise,rather than trying to automate
everything.
And finally, ensure you have apath to sustainable unit
economics as you scale,particularly if you're relying
on third-party artificialintelligence APIs.
So, if you're ready to level upyour business with AI, check
out our Total Sum Game courses,where we teach you how to
incorporate AI into yourbusiness.
(20:36):
Thank you for joining me on theTotal Sum Game podcast.
Until next time, keepinnovating and building.
The most successful AI businesses today don't
actually lead with AI in theirmarketing.
Instead, they focus on thespecific problem they solve and
the outcomes they deliver.
Welcome to the Total Sum GamePodcast.
(00:20):
I'm Adam Fierstein, and todayI'm tackling one of the most
requested topics from ourlisteners how to identify
genuinely profitable AI businessopportunities.
In a market flooded with hype,it seems like every day there's
a new revolutionary AI businessmodel being promoted as the next
gold rush, but, as we've seenover the past year, not all that
(00:44):
glitters is gold.
I've spent the last six monthsdeeply analyzing dozens of AI
business models, talking withfounders and tracking real world
performance data, and what I'vediscovered is that there are
clear patterns separating the AIbusinesses that are generating
sustainable profits from thosethat quickly fizzle out, despite
(01:04):
the initial excitement.
In today's episode, I'mbreaking down the five most
promising AI business modelsright now the three that are
most overhyped and giving you apractical framework to evaluate
any AI opportunity that comesyour way.
Let's start with a quickreality check on where we
actually are with AI businessmodels in 2025.
The market has definitelymatured compared to 2023's Wild
(01:29):
West phase, when Chad GPT firstexploded into the mainstream and
everyone was scrambling tolaunch something anything with
AI.
We're now firmly in what I'dcall the reality phase, where
the fundamentals of businesshave reasserted themselves.
Revenue unit economics andsustainable competitive
(01:52):
advantage matter again.
The days of raising millions ona deck with the words
AI-powered are largely behind us.
Investors and customers alikehave become much more
sophisticated about separatingactual value from marketing hype
.
One interesting trend I'venoticed is that many of the most
successful AI businesses todaydon't actually lead with AI in
(02:13):
their marketing.
Instead, they focus on thespecific problem they solve and
the outcomes they deliver.
The companies seeing the mosttraction aren't selling AI.
They're selling time savings,cost reduction, creative
breakthroughs or competitiveadvantages.
This is similar to whathappened with cloud companies or
mobile-first businesses inprevious tech waves.
(02:36):
Eventually, the technologybecomes table stakes and the
focus returns to fundamentalbusiness value, and the focus
returns to fundamental businessvalue.
So, with that context in mind,let's dive into the five AI
business models that aregenerating real, sustainable
profits right now.
The first model that's showingstrong performance is what I
call AI-enhanced professionalservices.
(02:57):
These are businesses that taketraditionally high-cost
professional services like legalwork, accounting or design, and
use AI to dramatically improvetheir efficiency.
What makes this model soeffective is that it combines AI
capabilities with humanexpertise.
The AI handles the routine,time-consuming aspects of the
(03:19):
work, while humans focus onjudgment strategy and client
relationships.
While humans focus on judgmentstrategy and client
relationships.
A great example is CaseText,which uses their AI tool,
cocounsel, to automate contractanalysis, but keeps experienced
lawyers involved for finalreview and strategic advice.
They've been able to reducecosts by 60% while maintaining
(03:40):
quality, which has opened uplegal services to previously
underserved market segments.
Quality, which has opened uplegal services to previously
underserved market segments.
Another excellent example isClarity, which automates
document review for legal andfinance teams.
They claim to reduce documentreview time by 85%, while
improving accuracy by 30%.
What's impressive is thatthey've managed to secure
(04:02):
enterprise clients like Coupaand Gusto.
Is that they've managed tosecure enterprise clients like
Coupa and Gusto?
The second promising model isvertical-specific AI
applications.
These are tools built to solvevery specific problems in
particular industries, ratherthan being general-purpose AI
platforms.
What makes these effective istheir deep domain knowledge.
(04:24):
What makes these effective istheir deep domain knowledge.
They're trained onindustry-specific data and solve
problems that generalist AItools simply can't address
effectively.
For instance, construct AI inthe construction industry uses
AI to analyze building plans andidentify potential code
violations and safety issues.
It's a narrow use case, but itsolves a real pain point that
(04:46):
costs the industry billionsannually.
They've reported a 78%reduction in compliance-related
delays for their clients.
The third model showing strongresults is AI-powered data
analysis and decision support.
These businesses take theoverwhelming amounts of data
companies collect and turn itinto actionable insights and
(05:08):
recommendations.
What's powerful here is thatthese tools don't just provide
analytics.
They actually help decisionmakers understand what actions
to take based on the data.
A standout example is BloomReach, which helps e-commerce
businesses optimize theirinventory and pricing in real
time based on the data.
A standout example isBloomreach, which helps
e-commerce businesses optimizetheir inventory and pricing in
real time based on dozens ofmarket variables.
(05:30):
Their clients report an average22% increase in profit margins
and a 15% increase in conversionrates.
I'm also impressed by Sky,formerly Kenshu, which helps
marketing teams optimize adspending across multiple
platforms.
They've built AI that canpredict campaign performance and
automatically reallocatebudgets to maximize ROI.
(05:53):
Their clients are seeing anaverage 31% improvement in
campaign performance.
The fourth model is AI workflowautomation.
These businesses identifycomplex multi-step processes
within organizations and use AIto streamline or completely
automate them.
The key to success here isfocusing on end-to-end processes
(06:17):
rather than just individualtasks, which creates much higher
value for customers.
Automation Anywhere is doingthis exceptionally well in the
healthcare space.
Automating the entire insuranceverification and billing
process for medical practices,they're saving some clinics over
30 hours of staff time per weekand have reduced claim denial
(06:39):
rates by up to 63%.
Another great example is UiPath, which has evolved from simple
RPA to using AI for complexworkflow automation.
They're helping companiesautomate everything from
customer onboarding to employeeoffboarding, with an average ROI
of 383% according to ForresterResearch.
(07:02):
And the fifth promising model ispersonalization at scale.
These businesses use AI tocreate customized experiences,
products or services for largecustomer bases in ways that
would be impossible manually.
What makes this modelcompelling is that it creates
value that literally couldn'texist without AI.
(07:23):
It's not just making anexisting process more efficient.
Fascinating example is DreamboxLearning, which creates
personalized learning paths forstudents based on their
individual strengths, weaknessesand learning styles.
They're seeing learningoutcomes improve by 40% compared
to standardized approaches andhave now reached over 5 million
(07:45):
students.
Another impressive example isStitch Fix, which uses AI to
personalize clothingrecommendations at scale.
Their success rate inpredicting what customers will
keep has improved by 35%, andthey've been able to reduce
returns by 20% compared toindustry averages.
Now let's talk about the threeAI business models that are
(08:08):
generating a lot of buzz butshowing concerning signs when
you look beneath the surface.
No-transcript.
The first overhyped category iswhat I call AI middlemen
businesses that essentiallyrepackage existing AI APIs from
Google, openai or Anthropic withminimal added value.
The fundamental problem withthis model is the lack of
(08:30):
defensibility.
When your core value comes fromanother company's technology
that anyone can access, you'reextremely vulnerable.
We've already seen severalwell-funded startups in this
category implode when theunderlying AI providers change
their pricing or launchedcompeting features.
A notable example is Jasper AI,which initially gained traction
(08:54):
as one of the first open AIwrappers, but has faced
increasing pressure as open AIimproved its own direct
offerings.
Another example is the flood ofchat, gpt plugins and GPT
stores apps that gained initialusers but struggled to build
sustainable businesses becausethey couldn't differentiate
beyond the underlying modelcapabilities.
(09:16):
The second overhyped model is AIcontent farms Businesses built
around generating massiveamounts of content across
thousands of websites orchannels to capture ad revenue.
While this can work in theshort term, it's not sustainable
.
Search engines and platformsare rapidly getting better at
(09:38):
identifying and penalizing aigenerated content that doesn't
provide genuine value.
I've tracked several contentfarms that saw their traffic and
revenue plummet by 80 or moreafter recent algorithm updates
specifically targeting thisapproach.
Companies like red venturesuresand Content Mills in the health
(09:58):
and finance space have beenparticularly hard hit.
Google's helpful contentupdates and EEAT standards have
been particularly effective atidentifying and demoting
AI-generated content that lacksexpertise and authenticity.
The third concerning model is AIfeature companies.
(10:19):
Startups built entirely arounda single AI capability that
should really be a featurewithin a larger product.
These businesses areparticularly vulnerable to being
made irrelevant when largerplatforms simply incorporate
similar functionality as astandard feature simply
incorporate similarfunctionality as a standard
feature.
We've seen this happenrepeatedly with AI summarization
(10:43):
tools like SumEyes, simple AIchatbots like many customer
service bots, and basic imagegeneration apps like early
versions of Lenza that brieflygained traction but couldn't
sustain themselves as standalonebusinesses.
It's the classic feature versusproduct problem, but AI has
(11:03):
accelerated the cycle offeatures being absorbed into
platforms.
Many standalone AI tools thatraised millions in 2023 are now
basically free features inMicrosoft 365, google Workspace
or other major platforms.
Based on my analysis of what'sworking and what isn't, I've
(11:23):
developed a framework to helpentrepreneurs and investors
evaluate AI businessopportunities.
I call it the MOTES framework,which stands for MOTE Outcomes
Automation, balance, training,data and Scalability.
Let's break down each component.
First, moat refers to yoursustainable competitive
(11:44):
advantage In the AI space.
This often comes fromproprietary data, unique domain
expertise or network effects,not just from using AI
technology itself.
The question to ask is if awell-funded competitor had
access to the same AI models,what would still make my
business special Companies likeDatabricks have built strong
(12:08):
moats through their combinedexpertise in data infrastructure
and AI model development.
Next is outcomes the specific,measurable results you deliver
for customers.
The strongest AI businesses canpoint to concrete improvements
in metrics that customers careabout.
For example, we reduce customerservice costs by 35% is much
(12:32):
stronger than we use advanced AIto optimize customer
interactions.
Ai to optimize customerinteractions Gong, the revenue
intelligence platform, does thisexceptionally well by
specifically quantifying howthey improve sales conversions.
The automation balance refersto how effectively you combine
(12:52):
AI automation with humanexpertise.
The most successful modelsaren't trying to eliminate
humans entirely, but ratherredefining how humans and AI
work together.
Companies like Scale AI havemastered this balance.
While they automate significantportions of data labeling, they
maintain human oversight toensure quality, resulting in
(13:13):
data sets that are 99.8%accurate.
Training data is about youraccess to unique, high-quality
data that can train AI models tosolve your specific problem
better than general purposemodels.
Samsara is a great example here.
They've collected billions ofdata points from IoT sensors in
industrial settings, giving themunique training data that
(13:36):
allows their ai to predictmaintenance issues with 92
accuracy far better than genericmodels could achieve.
And finally, scalability refersto how efficiently your
business can grow.
The best ai business modelsactually get stronger as they
scale due to network effects,improving data or decreasing
(13:57):
marginal costs.
Snowflake exemplifies this withtheir Data Cloud platform.
Their AI capabilities improveas more customers join and share
data, creating a powerfulflywheel effect where each new
customer makes the platform morevaluable for everyone.
Using this framework, I found Ican quickly assess whether an AI
(14:18):
business concept has genuinepotential or is likely to
struggle as the market matures.
To make this more concrete,let's apply the MOTES framework
to a few real-world examples.
First, let's look at an AIbusiness in the marketing space
that I've been impressed by.
Persado helps businesses createand optimize multi-channel
(14:40):
marketing campaigns usingartificial intelligence.
Promote they have proprietarydata from thousands of campaigns
across different industries,giving them insights no
competitor can match.
Their message machine hasanalyzed over 100 million
marketing messages For outcomes.
They can point to an average32% improvement in campaign ROI
(15:05):
for their customers, a metricthat directly impacts bottom
line For clients like Chase.
They've increased credit cardapplications by 47%.
Their automation balance isstrong.
They use AI for creativegeneration and optimization, but
keep experienced marketersinvolved in strategy and brand
(15:27):
alignment.
Their platform suggestsmultiple options that human
marketers can choose from andrefine their training data
advantage comes from theperformance feedback loop of all
those campaigns, whichcontinuously improves their
recommendations.
They've built language modelsspecifically trained on
(15:49):
marketing, language and consumerresponses.
And for scalability, theirtechnology platform can support
enterprises 10 times larger thantheir current client base with
minimal additional costs.
They've successfully scaledfrom working with small
businesses to global enterpriseslike Dell and Vodafone.
Now let's contrast that with anAI writing tool that recently
(16:12):
shut down.
Despite initially strong usergrowth, their moat was
essentially non-existent.
They were using OpenAI's modelswith a nice interface, but
dozens of competitors couldeasily offer the same thing.
When OpenAI improved ChatGPT'sinterface, many users simply
switched to the source.
Their outcomes weren't clearlydifferentiated.
(16:35):
They promised better contentbut couldn't quantify the
improvement over other methods.
They couldn't demonstrate aclear ROI to justify their
subscription prices.
Their automation balance wasskewed too heavily toward
replacing human writers entirely, which created quality issues
for anything beyond basiccontent.
(16:57):
Without human editorialoversight, the quality varied
wildly.
Their training data was generic, with no specialized datasets
or feedback loops to improveperformance for specific use
cases.
They were essentially passingthrough OpenAI's general models
without any domain-specificimprovements, and their
(17:18):
scalability was compromised byhigh API costs that actually
increased linearly with usage,creating margin pressure as they
grew.
They were paying almost 70% oftheir revenue to OpenAI for API
access.
When you compare these twobusinesses through the moats
framework, it becomes muchclearer why one succeeded while
the other struggled.
(17:39):
Looking ahead, I see severalemerging opportunities in the AI
business landscape that aren'tyet saturated but show strong
potential.
One area I'm particularlyexcited about is what I call AI
orchestration tools thatcoordinate multiple specialized
AI models to solve complexproblems that no single model
can handle effectively.
Companies like Langchain andFixie are pioneering this space,
(18:02):
creating platforms that cancombine the strengths of
different models and tools intocohesive workflows.
This AI of AI's approach solvesthe limitations of any single
model.
Another promising direction isembedded AI financial services,
using AI to fundamentallyreimagine lending, insurance and
(18:24):
investment products, based onmuch richer data analysis than
traditional approaches.
Upstart and lending andlemonade and insurance are early
examples, but I see much morepotential.
Upstart has shown they canreduce default rates by 75%
while increasing approval ratesfor underserved populations,
demonstrating the power of AI infinancial risk assessment.
I'm also seeing interestingdevelopments in AI for physical
(18:48):
operations, bridging the gapbetween digital intelligence and
real-world processes inmanufacturing, logistics and
physical retail.
Companies like Covariant inrobotics and Standard Cognition
in retail are showing how AI cantransform physical operations.
Covariant's robots can nowhandle over 10,000 different
(19:09):
items in warehouses with 99%accuracy tasks that were
impossible for automation just afew years ago.
For entrepreneurs listening, thekey insight is that the most
compelling opportunities oftenlie at the intersection of AI
capabilities and deep domainexpertise in a specific industry
(19:30):
or problem space.
The days of success coming fromsimply applying generic AI to
generic problems are largelybehind us.
The next wave of successful AIbusinesses will come from
founders who deeply understandboth the technology and the
domain they're working in.
To wrap up today's episode, letme summarize the key insights
(19:52):
for identifying promising AIbusiness opportunities in
today's market.
First, focus on solvingspecific high-value problems
rather than showcasing AItechnology for its own sake.
Second, build real moats beyondjust using AI.
Proprietary data, domainexpertise and network effects
are crucial.
Third, think carefully aboutthe right balance between
(20:15):
automation and human expertise,rather than trying to automate
everything.
And finally, ensure you have apath to sustainable unit
economics as you scale,particularly if you're relying
on third-party artificialintelligence APIs.
So, if you're ready to level upyour business with AI, check
out our Total Sum Game courses,where we teach you how to
incorporate AI into yourbusiness.
(20:36):
Thank you for joining me on theTotal Sum Game podcast.
Until next time, keepinnovating and building.
Popular Podcasts
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.