The worldwide SaaS market is about $300 billion per year.
The labor market in the U.S. alone is $13 trillion.
What software is now going after, the prize that it's going after, is the labor market.
Almost every software company has basically taken a filing cabinet and turned it into a database.
But what's happening now is the whole thing is effectively done end to end.
Software ate the world.
Now it's coming for labor.
At A16Z's LP Summit, general partner Alex Rampell, who leads the apps fund, took to the stage to discuss why the real market opportunity isn't the 300 billion SaaS industry.
It's a $13 trillion U.S. labor spend.
From filing cabinets turned databases to AI that actually does the job.
Alex breaks down outcome-based software, new pricing models and what happens when agents sell, support and collect on their own.
Let's get into it.
I'm going to talk about how software eats labor.
First, Mark wrote an essay for the Wall Street Journal a long time ago about 10 years ago about how software eats the world.
I guess the labor force falls within the world, so this is a natural follow-up.
But if there's only one takeaway that I can leave with you today, it's that the labor market this is almost obvious is so much bigger than the software market.
So the worldwide SaaS market is about $300 billion. per year.
Worldwide market cap is about $2.2 trillion.
The labor market in the US alone is $13 trillion, so much, much bigger.
And it doesn't mean that the software market will suddenly be worth 13 trillion a year.
But what software is now going after?
The prize that it's going after is the labor market.
And I figured there's nothing better than starting a venture capital conference with a picture of the world's most famous communist.
That's Karl Marx right there.
And if you've read Das Kapital, which I read back in college, if you remember, the central tenet is that there's capital and labor, and capital exploits labor.
They're two distinct things.
But what's really exciting this is almost like a chemical equation is that the capital that you give us, we give to companies.
What do the companies do with the capital?
They buy GPUs or rent them.
They hire engineers.
They buy coffee.
They give coffee and GPUs to the engineers.
And then out pops software that does the job of labor.
It's like the new E equals MC squared.
And we're seeing this happen at so many companies.
And these companies are scaling so quickly because they really are selling into clients or end users and saying hey, we'll do this job for you.
We're not giving you software.
We're going to do a job for you.
And that's the sales pitch.
And obviously, like, this concept is not new per se.
I mean, automation has been around for a long time.
So you know seamstresses.
The loom replaced those jobs, but you still had somebody that had to operate the loom.
Or steamships.
Sailboats and steamships obviously made that form of labor much more efficient, but you still need somebody to operate and shovel coal into the steamship and put the sails up on the mast.
And then of course, the printing press invented by Gutenberg, that was a big innovation.
And the steam powered press allowed the LA Times to print out hundreds of thousands of copies of that newspaper, versus somebody manually cranking one each time.
And lastly, like the assembly line was a massive innovation in terms of replacing people, that hand assembled every gizmo.
And now you have an assembly line, but you know, a lot of people worked on the assembly line.
So it Again, this concept is not new.
You could always take capital, make a machine, and then have more efficient labor on the other side.
But what's happening now is the whole thing is effectively done end-to-end.
And I want to take you through a journey of what the software market has always been, because I think understanding the past is the best window of actually understanding the future and what's possible.
So my thesis is that almost every software company has basically taken a filing cabinet and turned it into a database.
And this is where the $2.2 trillion of market cap has come from.
This is where the $300 billion in annual software revenue has come from.
And the first example that I'd like to point to is a company called Saber Systems.
This is with two A's, if you've seen it.
It's a company in Texas, but this was a joint project of American Airlines and IBM back when IBM was the most powerful computer company in the world, technology company in the world because, at least according to chat GPT.
What would it have looked like to book an airline ticket in 1959?
How did the American Airlines handle this?
They probably had like lots of filing cabinets with like sheets like this.
So Betty Owens calls up and says, I want to sit in seat 4A, talks to somebody on the phone.
Oh, wait, actually, I want to cancel my flight.
They have to erase it.
No, actually, I want to sit in seat 2C, erase that again.
Filing cabinets basically kept track of everything.
It was pretty inefficient.
You had a lot of people that worked the filing cabinets.
You couldn't share information from one office to another, because everything was domiciled in one filing cabinet.
And Sabre changed the game by putting it all on an IBM mainframe and then having lots of kind of thin client terminals that were used by travel agents around the world that could access that mainframe.
And this is how travel was revolutionized.
Galileo did this for hotels.
Amadeus did this for Europe.
I mean, like big companies came out of this.
And of course, the same process happened for sales.
So I think somebody yesterday mentioned Glengarry Glen Ross, great movie.
But if you remember, there were lots of business cards.
You wanted to get the Glengarry leads.
Those were the good leads.
Those were pieces of paper.
And for those of you that are old like me, you might remember a company called Axe Systems in the 1980s.
This was one of the big CRM companies of the day.
Goldmine followed that in 1990.
And then Tom Siebel started Siebel Systems in 1993.
But all of these took the filing cabinets of sales, put them originally in kind of mainframes, and then Salesforce came in 1999 and put that in the cloud.
But you know again, the same salesperson that accessed the physical file in a fictional movie set in the 1950s would now access a Salesforce record in 2010.
Same process, just different medium.
Manufacturing and inventory, this is another big one.
So imagine that you're a product manufacturer.
How many widgets do I own?
What's my inventory?
What's my sales?
IBM, again at the forefront of this, but other companies, some that are around today, like SAP, started in 1972.
Bond, JD Edwards, Epicor, Sage, i2.
These are all big companies that basically digitized old-fashioned records.
My favorite one, just to show how pervasive this idea is, is there is actually a big business in library card catalogs.
So libraries were, have been around for a very, very long time.
Library of Alexandria, right?
Like long, long time.
The Dewey decibel system comes out.
And when I was growing up I'd go to like the card catalog and figure out okay, they're all alphabetically sorted.
And then eventually somebody created, started with this company called OCLC, built a reasonably sized business and sort of innovative Circe Dynics business of digitizing those card catalogs where now you enter into a terminal at the computer in the library and then you figure out where your book is.
Is it in stock?
Legal case files.
Like every time I went to a law firm in the 1980s, it's like most of the square footage was filing cabinets.
And companies like PC Law, like a lot of LexisNexis and Reuters, revenue comes from selling to law firms digitizing things that would otherwise occupy so much square footage in the nice Fifth Avenue offices of law firms.
My parents were accountants growing up and I remember going to their office and again there was no room for a little five-year-old to run around because all filing cabinets Intuit comes along with QuickBooks, digitized financial statements.
Peachtree this is a company from the 1970s MYOB, like lots of companies again filing cabinets, filing cabinets.
You get the drill here.
My favorite name in the history of software is filing cabinets.
The first electronic health records company was a company called Mumps, which arguably is the worst name in the history of software.
It somehow lost out to malaria or measles or something.
But this was Mass General Hospital and they wanted to replace their filing cabinets and they came up with this programming language and database system called Mumps.
But Epic and Cerner, Epic is the biggest electronic health records company in the world.
It was started in 1979.
So again, and what they did was just digitize these massive number of files that every hospital system and doctor would have in their office.
And kind of lastly here, HR and payroll.
So actually, even before Sabre, automated data processing, ADP, was started in 1949.
How do you keep track of time and attendance?
You had your time slip, your time cards.
How do I figure out tax withholding?
All of these companies came out.
And again the progression was I take the file, I put it in a mainframe and then companies like Workday was effectively PeopleSoft.
It was the same team that started PeopleSoft.
They put it in the cloud. but the same process was there.
The people that looked at your time and attendance slip in 1940 were the same people that looked at it in 2015, but now the medium was not paper, it was not mainframe, it was cloud.
And the reason why I mention all of this is because nothing has actually gotten that much more efficient.
Because the filing cabinets were read by humans.
The digital records are read by humans.
It's like this woman here, whatever she's doing, looking at helping a customer with customer support needs or something once upon a time would have looked at a piece of paper, now is looking at a computer.
And the reason why this is so important to understand is because the whole business model of software has to change.
It has to change.
This is what I call, as homage to Starbucks here, the tall, grande, venti model of SaaS.
If you go to any SaaS company in the world, go to their landing page, it probably looks like this.
This is a company called Zendesk.
It's now a private company.
It was taken private by Premier and Hellman Friedman a few years ago.
And this is a $2 billion ARR company that sells seats.
So the most popular, like the Venti package is you get the sweet professional $115 a month.
But we've talked about in the last couple days how now AI does a really good job answering customer support queries.
So how many seats do you need if every one of your agents is 9,000 times more productive?
Imagine that you have 1,000 seats.
Imagine that I've got 1,000 people working in my customer support call center.
I pay each one fully loaded $75,000 a year.
So that's $75 million a year in cost for people.
Well, what's my software cost?
Well, it's 1,000 times 115 times 12.
It's about $1.4 million a year for the software.
So the people cost is so much higher than the software cost.
And this could go one of two directions for a company like Zendesk.
If it turns out that AI can answer all the questions, how many seats do you need?
Zero.
You don't need a single seat left.
The AI answers everything, and then Zendesk is charging per seat.
So they would go from $1.4 million to $0 in revenue.
That would be very bad.
On the other hand, I mean, look at the math here.
If each human is answering 2000 questions a year right now, your fully loaded cost as the company that uses Zendesk as your system of record for answering customer support, it's about 37 of human cost and 69 cents of software costs.
The cost per answer is $38.
I mean, this is rough company example.
Maybe Zendesk could charge $5 million a year.
Like, hey, don't spend $75 million a year on support anymore.
Spend $5 million a year on support.
Just pay it all to us.
Don't pay us 1.4 million.
Pay us 5 million, save 75.
So Zendesk is really at the precipice of it's like their revenue could go to zero or their revenue could 3X.
And like, where is it going to go?
I don't know the answer.
They don't know the answer.
I've been talking to their CEO, they're piloting outcome based pricing in New Zealand right now.
So New Zealand holds the answer to all of our questions.
So stay tuned.
And just to give again another example of just how much bigger these labor or quasi labor markets are, like the 13 trillion of wages that we talked about.
You know, software revenue is very small if you just take one category, like just one particular profession, and take the example of nurses.
Just because we have a portfolio company in this space, nurses in the United States of America earn about 650 billion a year.
There are about four and a half million registered nurses.
That's bigger than the entire worldwide software market.
And it doesn't mean that the nursing software market will be this big, but it means like, this is the pool that you're really playing against.
And the reason why I wanted to start with the filing cabinets is like this is going to start moving to outcomes.
Like the software is going to go from being the filing cabinet to effectively operating on the filing cabinet.
And what does that mean?
Well, you know, take the first example.
If I have the filing cabinet for travel, you know, maybe the software can rebook flights.
Or I want to book a trip for 75 kids at my son's high school and need to talk to an agent.
I don't need to talk to an agent anymore.
I just talk to the software company.
I talk directly to United Airlines, you know, AI, and they do the entire thing for me.
Sales, I mean, this is kind of obvious, you know, Salesforce charges per seat per month.
Salesforce should just sell for me.
I don't want to pay for a thousand seats.
I want to pay for customers like hey, go get me customers, because you're the backend for that anyway.
Or survey all of my customers.
Do a 30-minute phone call with every single one of them.
See how they're doing.
Are they happy?
Are they going to renew with me?
Yes or no?
You know, What about manufacturing?
Well, imagine that I make widgets and there are these things called tariffs that are happening.
I'd love to figure out what my tariff exposure is.
Let me just ask my ERP system, research that for me.
Or I want you to do an audit or call my suppliers and see if they're going to still be able to ship me their things on time.
A library card catalog is as crazy as this one is.
You know, what if my book is overdue?
The librarian shouldn't be calling me.
The library card software company should be calling me saying hey buddy, return the book or order more books, because this one's very popular.
You know, Ben's book is selling out.
Order more copies of that legal case files.
Like a lot of software companies that are popping up in this space, it's no longer the system of record for like time and attendance.
Like draft me a contract, like do that work for me.
You could start billing out again, unclear how the pricing model of this is going to work, but start billing for that.
Um accounting and bookkeeping um, you know, AR aging summary.
Like this concept has been around for a long time.
I have a lot of companies that owe me money, a lot of clients that owe me money.
Um, In 1940 I'd look at the printout and say I'm going to call these customers and hopefully collect for them or send the guy with the crowbar or do something.
In 2000, I'd look at the QuickBooks thing and do the same thing.
Now the software company can go call, like QuickBooks is going to start calling customers, customers of their clients and say hey, you owe me money, please pay me back, and I can accept a payment on the phone right now.
Health records.
So if you saw me limp off stage, I sadly had Achilles repair surgery about three months ago.
I don't recommend it.
The day after my surgery I got a call from Stanford Hospital saying you know Alex, on a scale of one to 10, what's your pain level?
And I said, 11.
And they said, very funny.
And I said, no, it really is 11.
And an AI nurse can't do CPR, right?
An AI nurse can't attend to a gunshot wound victim.
But an AI nurse can totally call a mid 40s patient like me and say, how are you doing?
Is there anything we can help with?
You know, do you have a fever?
Oh, you do, you should go to the hospital right now and stuff like that.
And again, these are the outcomes.
These are the operations that can be performed on.
If you have my medical record, that's what you do with it.
And charge for that, charge $20 for that outbound call.
HR and payroll, how do I do a reference check?
How do I make sure that on your resume, you said you worked at these three companies?
You know what?
Workday should call those three companies and say, did Alex really work there?
Explain benefits, help with enrollment.
Workday could probably triple their revenue if they start doing this, because they're already the system of record.
So if you know the story of Airbnb, Airbnb famously started by somewhat scraping Craigslist.
Craigslist is this horrible site from the mid-90s that has not changed since the mid-90s.
And they have lots of listings for apartments.
And you go look at an apartment and like half the time it's a scam, half the time it's already been booked or it's been relisted every single day.
And they basically put them in a better interface and called it Airbnb.
And that's how they got started.
And this is one of the most exciting things that we're seeing, which is here's a real ad on Craigslist.
Because, since I have so much copious time, due to my said Achilles injury, I used to run every day.
I bike every day.
I can't do that anymore.
So I hang out on Craigslist looking for jobs.
Not for myself, of course.
And here's a job for Plaza Lane Optometry.
They're hiring a front desk receptionist.
And they've had this job opening for six months.
It's been hanging out there for six months.
You have to now say in California how much you're going to pay.
It's $45,000 a year.
So supply and demand.
If they probably said we're going to pay 100000 a year, they would have filled this job.
They need to pay $45,000 a year to make this work within their cost model or something.
If you look at the job responsibilities, if you can read this like the first one is like open and close the shop, lock the door.
AI can't do that.
But a lot of the other job responsibilities AI can completely do.
It's you know.
Argue with insurance company.
Call patient the day before their appointment to prevent no-shows.
If somebody doesn't show up, that's a big opportunity cost for the optometrist.
And if you were to look at the optometry market and say, is that a good software market?
You would say categorically no, because Plaza Lame Optometry probably spends 500 a year on software.
They probably have a Microsoft Office license.
They probably have a website with Squarespace or Wix, and that's it.
So they probably spend $500 a year.
So in this new world and again we're seeing a lot of companies doing this they peruse Craigslist, they look for a job listing and they're like hey, Plaza Lane Optometry, I want to apply for the job.
And, you know, the optometrist is like, okay, that sounds great.
Tell me about your qualifications.
Where have you worked before?
It's like I know this sounds crazy, but I'm a software company and I can't close the doors and do this.
But I can do these other eight things.
Can I give you a demo?
And at first, the optometrist is like, no, I want to, oh, okay, I'll try it out.
And, by the way, it's 20000 a year, which is much less than the 45000 a year that you were going to pay for a human that you can't even hire.
So this is happening right now.
And it's incredible.
And again, like the software spend small for these people, the effective labor spend much, much higher.
And this is what's massively growing the market for a lot of these obscure industries.
So I just thought rather than me talk, I want to give you an example.
This is one of our portfolio companies, Happy Robot, which serves the freight and trucking space.
So listen, this is a negotiation on the phone conducted with a prospective client and AI.
All right.
So this is Joliet, Illinois.
Today between 6 a.m. and 2 p.m. delivers Monday between 6 a.m. and 4 p.m.
I have 700 on this one.
Would you like to book this load?
I'm going to need 800 for this one.
I can check on that.
So we can't do 8 at this point.
Any chance you could come a bit closer to the load board rate?
I could do with 775.
That's really my lowest.
I understand.
So no chance you could do any better than 775.
How about 750?
Let me see.
OK, I was able to get you 735 here.
735, OK.
So that's a closed transaction.
Who's the robot and who's the human?
This is like the new Turing test.
I'll let you guys quiz me afterwards on that with your own theories.
Here's another example.
This is a company called Salient that does collections.
So if you're a lender, if you're an auto lender...
You have to collect repayments on your loans.
And Salient serves a lot of auto lenders in this country.
And here's another call from them.
Currently, your account is 51 days past due for $825.35.
Would you be able to make a payment today?
¿Puedes hablar en español?
Como mencioné, su cuenta está 51 días vencida con un monto atrasado de $825.35.
¿Le gustaría realizar un pago hoy para ponerse al día con su cuenta?
So the cool thing there is that Salient speaks dozens of languages.
So it speaks Tagalog, it speaks Vietnamese, Mandarin.
And again, this is a really important point.
It's not just about cost.
This is where people are getting this wrong.
It's not like oh, AI is going to take all the jobs because humans are too expensive and AI is cheaper.
AI does a bunch of things like intermittent demand is a big one.
Imagine that I'm Black Friday retail.
I'm a retailer and people know that, you know, sales are much higher around Black Friday.
So I would have to hire tons of cashiers or, if I'm an online retailer, hire lots of customer support people to go answer queries and whatnot.
And then what do I do on January 1st?
Do I fire them all and then rehire them in November?
But actually I should probably hire them in September because I have to train them. kind of tricky.
And there are other countries that have intermittent demand issues.
I mean United Airlines.
If they have bad weather over Chicago, which of course never happens, they can't just hire and train 10000 people overnight.
But again, AI very, very good at that.
The other example is there are a lot of demoralizing jobs out there.
What is a demoralizing job?
Well, collections is kind of a demoralizing job, because you go call people and say hey, you're overdue.
About half the time and I've listened to some of these calls.
There are lots of expletives on the other side right.
So it's like, hey, you owe me money.
It's like, F, you never call me again.
It's like, call back again.
Hey, you owe me money.
F, you never call me again.
A human would get kind of tired of that.
It's not a fun job.
AI doesn't get bothered.
So again, demoralizing job is very, very good regulatory certainty.
So go back to this um.
You know i co-founded a company called the firm um.
We got trained every quarter on udap unfair defect sorry unfair, deceptive and abusive practices.
So there are lots of laws that mandate what you can and can't say to a customer.
So imagine you're calling a customer, say hey, you owe me money.
The customer says f you and then you have somebody on the other end who's had a kind of a bad day and you can't blame them.
It's like f you back customer you, they're going to get in trouble for that right.
And you have more certainty when you can program effectively a robot to conduct the entire call end-to-end much more so than people.
And my favorite example here is somebody who studied you know, I speak Russian, Japanese and a smattering of Spanish and spent way too much time learning that stuff.
Language is languages.
I mean, it's just... fact that a nurse, an AI nurse, like what if I only spoke Farsi?
Does Stanford have any Farsi speaking nurses that can call me about my pain level?
What if I only spoke Mongolian?
Like now they do.
An AI nurse, an AI collections agent, an AI negotiator, like all of these things can be done in dozens of different languages instantly.
And you just wouldn't hire a human for that.
You can't get somebody in Iowa that speaks Serbian on demand intermittently, with a demoralizing job, and so on and so forth.
You know, AI does great.
And it's also very important that AI expands the market.
And this is why I wanted to start with the story of filing cabinets, because there was no software company for compliance.
Because this is actually a fact from the Bureau of Labor Statistics the fastest growing job in America is manicurist pedicurist.
AI can't do that.
The second fastest is compliance officer.
So compliance officer, you don't need software for compliance officers.
If you're a city bank, You need more people.
And no company has popped up effectively doing software because it's like the software market's kind of small.
The people market's very large.
Now you can roll into Citi and say hey, I will do compliance for you end to end as a software solution.
Pay me 10 million a year and on your software product that kind of tracks everything, because before they would just they would just get more microsoft office licenses or collections.
Like there is no software company for collections, there are collections people at collections firms.
Now you can actually enter potentially with the wedge of voice, and this is going to get commoditized quickly and we realize this but you can backfill into a software company with real software revenue, real software margins, real software retention.
The other thing that you're going to start seeing us do is there are a lot of non-AI companies that now work because of AI.
So because of my said very sad injury, I'm a big cyclist, as I mentioned.
My garage looks like this with a lot of bikes that aren't being used.
I hope to use them again in the future.
Why isn't there an Airbnb for bicycles?
Why hasn't somebody built that?
Well, I'll tell you why.
It's a very bad idea.
That's why nobody's done it.
And why is it a bad idea?
Because the central tenant, AI, not AI, it could be in 2000 BC, it doesn't really matter.
If your customer acquisition cost plus your cost of goods sold is greater than your lifetime value, do not proceed.
That's not a business.
But now with AI, imagine that I wanted to start Airbnb for bicycles.
Well, how am I going to go get the people that have spare bicycles in their garage?
Am I going to hire a bunch of expensive Stanford kids in Palo Alto and have to give them 20 different flavors of coconut water and cater to their millennial needs in order for them to work in the sales operation in Palo Alto?
I'm going to have an AI sales rep call everybody.
And the cost per AI sales rep per year is a few hundred bucks, not 100000.
No coconut water necessary.
Well, what if there's an emergency?
Well, you know, now there's a 1-800 number that you call.
That's an AI rep that can do everything call the police, call whatever.
And, you know, lastly, like, how do I screen this person, do a background check?
Is the bicycle good?
Is it stalled?
Like whatever you would need to do this, this silly business that we're not going to fund.
AI can do this as well.
So you actually have a whole category of business which was like it would have worked, except for this pesky customer acquisition cost or pesky cost of goods sold.
And of course, AI, you can vibe code your way into one of these businesses anyway.
So, you know, massively expands the size of the market.
The non-AI market, given the AI infrastructure.
Just allowing you know CAC to come down, COGS to come down forever.
I mean, every company is going to start using these tools.
So you know it will become a version of the Yogi Berra quote.
It's so crowded, nobody goes here anymore.
But now a lot of companies kind of prosecuting sometimes older new ideas that just wouldn't have worked five years ago.
And this is a global opportunity, right?
I mean, like the U.S. labor market, big.
It's $13 trillion a year.
The worldwide labor market is so much bigger.
And our job on behalf of your capital here, our labor on behalf of your capital, is to find the best companies that will make software look small.
So thank you very much.
Thanks for listening to this episode of the A16Z podcast.
If you liked this episode, be sure to like comment subscribe, leave us a rating or review and share it with your friends and family.
For more episodes, go to YouTube, Apple Podcasts and Spotify.
Follow us on X and A16Z and subscribe to our sub stack at a16z.substack.com.
Thanks again for listening and I'll see you in the next episode.
As a reminder, the content here is for informational purposes only, should not be taken as legal business tax or investment advice or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund.
Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast.
For more details, including a link to our investments, please see a16zcom forward slash disclosures.
