Is AI Killing The $1 Trillion SaaS Industry?

The Trillion-Dollar Question

MARKETS
Is SaaS Dead?

Today, we are trying to answer a trillion-dollar question that has investors panicking: Is SaaS Dead?

A few weeks ago, an announcement sent shockwaves throughout an industry that, for two decades, felt untouchable and printed over a trillion dollars. Klarna, the “buy now, pay later” company, intends to substitute its Salesforce and Workday systems with in-house OpenAI developments.

What was thought impossible is now a reality; SaaS companies, enterprises with some of the highest customer retention rates worldwide, are at risk of becoming disposable because of AI, possibly becoming one of the first industries to succumb to its wrath.

But what are the signals this is happening? Are markets backing this idea? What will determine the survival or death of a given company? And do these threats also affect you as an employee or founder?

Today, we are learning about the recent AI breakthroughs that have spooked SaaS investors and are punishing hundred-billion-dollar companies, the accurate state of the market and its sinister future and, crucially, a simple guide for you to evaluate which companies will survive and which most probably won’t.

Concerning Signals

Few industries today are thought to be more exposed to AI than SaaS, companies that offer Software-as-a-Service through the cloud (the Internet), despite being some of the most successful companies in the past two decades.

But why have they been so successful? At first, it may seem hard to believe; SaaS’ appeal is obvious.

They offer industry-tailored workflows that allow you to streamline your internal processes. For instance, Salesforce will enable you to guide the entire sales lifecycle, from sourcing to account management, helping your sales reps focus on selling instead of bureaucracy.

They are also very convenient, as they usually offer a fixed price on a seat-based pricing system, billed monthly or annually. In simple terms, if you deem it necessary to provide access to ten sales reps in Salesforce, you’ll need to pay for ten licenses. Importantly, this means that the costs are predictable and, crucially, are Opex-based, preventing risky big purchases on long-term outcomes (capital expenditures or Capex).

However, despite all these advantages, many are proclaiming the industry's unavoidable death, as AI’s deflationary effect, combined with enterprises' tendency to leverage their data to train in-house AI models, might render their offerings useless.

And while this may seem like a multi-year prediction, this is taking place already. But how?

Klarna Leading the Way

As mentioned, Klarna has taken a step that has spooked many in the industry: it will completely dispose of its big OPEX SaaS costs and substitute them with OpenAI-based in-house developments.

Simply put, they are replacing third-party software partners they’ve had for years, like Salesforce or Workday, with Large Language Models (LLMs). With both companies' combined market values of $330 billion, this is an extreme sign of weakness regarding an industry that felt untouchable just a few years ago.

And while cases like Klarna are undoubtedly rare for now, is the writing on the wall by now? Well, maybe:

This Twilio demo provides insight into the types of use cases OpenAI’s real-time API unlocks, and also shows how to implement it with Twilio for instant calls.

  • Applications like Cursor, Claude’s Artifacts, or OpenAI’s brand-new Canvas already allow people with basic programming skills to build incredible stuff.

While no enterprise wants to endure the pain of imitating the features that products like Salesforce or Workday offer if AI reduces complexity and speeds up the development process, the reasons for continuing to pay the expensive licenses these SaaS companies demand will be hard to justify for the CIO once the CFO or the CEO starts hearing the ‘horn of war.’

Nonetheless, this horn is already sending warning calls, as CEOs are already asking their CIOs to rationalize and integrate their SaaS point solutions (SaaS products specialized in one single point of action, the exact opposite of what was promoted a few years ago).

And how is all this revolution coming to fruition? To start, all SaaS companies face one undebatable fact: their current business model is dying.

Increasing Pressures Toward Usage-Based Pricing

A month ago, Canva made headlines—but not in the way they desired. They introduced AI features to the license without adding any additional costs.

Although not officially acknowledged, the cost of running the AI models (most presumably OpenAIs) was eating them alive. And what is the best solution Canva leadership proposed?

A 300% price increase, or three times the previous value, leading to a hugely negative reception.

The moral of the story? With AI features, SaaS companies accustomed to billing users a fixed price will have to resort to usage-based pricing. In other words, customers will have to be billed based on their actual usage of the products.

This has very good things and very bad ones.

On the latter, it makes projecting your future costs unpredictable for both parties. This hurts a SaaS company’s capability to confidently project run rates (how much they will earn a given year), making it harder for investors to evaluate the Enterprise Value. This will also make Monthly Recurrent Revenue (MRR), a key metric, a vibe session more than a metric.

Undeniably, this will hurt the company's ‘spreadsheet appeal’ in the eyes of potential investors.

On the former, customers' views of the value provided to them become totally tangible, as they are billed based on usage. Also, this allows SaaS companies to fix the margins (what they earn on top of the customer’s unit of usage), allowing them to project their profitability much better. Still, if their product isn’t needed, revenues go to zero.

As always, there’s a trade-off, but the point is that this is an inevitable transition.

In seat-based pricing, users feel like they are overpaying, and with AI features that imply unpredictable costs due to token-based API pricing, SaaS companies can’t predict per-customer profitability because AI costs can be higher than the actual license cost.

Long story short, usage-based pricing is the future… for companies that survive (more on that later). But making unsubstantiated predictions is one thing; seeing what the market is really doing is another.

Thus, let’s look at the true current state of SaaS markets.

The State of SaaS

SaaS is a decent-sized industry valued at well over $1 trillion in the public markets. Considering multiple hundred billion-alone private companies like Stripe, we can argue that the total SaaS industry is over $2 trillion (valuation-wise, not revenues).

It may seem small, considering that many Big Tech companies are larger than the entire market, but that’s simply because these companies are abnormally huge.

Market-wise, according to Morgan Stanley, it had more than $500 billion in revenues in 2023 alone (considering both public and private markets).

But how did public SaaS companies perform this last quarter? Simply put, not great.

Happy Days are… Over?

For this, I will primarily use Jamin Ball’s analysis, partner at Altimeter Capital. I’ll provide a summary, and you can read the full report for free here.

The rough numbers are as follows:

  • SaaS markets are having trouble increasing revenues, having sold more than the previous quarter but less than last year (measured by new Annual Recurrent Revenue (ARR), or revenues they expect to collect that year). Performance-wise, they barely beat analysis estimates by 1.6%, although only 10% of them did not.

  • As for guidance (what these companies predict will be their near-term performance), only 55% of companies are confident they will beat what analysts expect (consensus). This means that SaaS companies are being very conservative on their future projections.

  • Growth-wise, this is one of the most concerning metrics in the industry, as growth estimates have collapsed considerably from a traditional 30% overall to approximately half of that. The fear of SaaS companies cooling down their growth estimates has caused several debacles. For instance, Salesforce dropped a whopping 20% in one day last May after they announced single-digit growth for the first time.

  • In terms of retention, the number of customers a company retains year over year, a signal of how much its products are liked, dropped from historical averages of 115-120% to 110% this last quarter (if you had ten customers one year ago who were paying you $1M in aggregate annual recurring revenue, and today they are paying you $1.1M, your net revenue retention would be 110%).

All things considered, SaaS’ elevator pitch looks more like a Requiem since its heydays are long gone (or, as investors like to put it, the industry is maturing), with sales and growth becoming the leading lagging indicators.

Consequently, moving forward investors will look closely at the operational numbers (costs, margins, etc.) to determine their investments, instead of pure growth.

That said, most public SaaS companies have grown a median of 7% over the last 12 months, but their mean revenue multiple, how much they are valued with respect to their Next Twelve Months (NTM) revenue, is just x6.7 over that revenue (i.e., if a company has projected 100 million in ARR, their valuation is ‘just’ $670 million).

But how does that multiple fair with other sectors?

  • The S&P 500’s average forward p/e is x23.2, almost five times larger than the SaaS median,

  • while AI public stocks (Big Tech + Semiconductor companies) trade at an average of x38.51 and x14.83, respectively.

In layman’s terms, SaaS companies are valued at way less than the overall US market. And if we look at AI’s private markets, an excellent projection of investors’ future bets, things get even more concerning.

A Hot Bubble, But Hot Nonetheless

If we look at the private markets, the AI hype is real. OpenAI just raised money at a $150.4 billion valuation ($157 billion post-valuation), a multiple of x42 over projected revenues (assuming Reuters published revenue run rate of $3.6 billion is accurate).

And if you thought OpenAI’s multiple is crazy, it’s actually one of the healthiest, kid you not. As reported by The Information a few months ago, some of AI’s most prominent companies have little to no revenues while raising money at staggeringly high multiples.

While comparing public valuations to private ones is an apples-to-oranges comparison, the differences are quite telling, especially regarding the considerable gap in excitement and hope for the future between AI and SaaS.

Luckily, not all SaaS companies are sitting on the sidelines waiting for the day of reckoning. Some are making moves, and pretty big ones, as we’ll see later.

But before, what has changed in AI that suddenly it represents such a risk for entire industries?

Agents, The New Step in AI’s Journey

It doesn’t take much genius to acknowledge the appeal of AI agents. Instead of humans actively interacting with a software interface, agents interact with the different workflows (such as logging a new customer lead for the sales team or onboarding a new employee) on your behalf.

But what recent developments make this paradigm closer than ever? Well, three: output consistency, real-time interactions, and raw intelligence.

For starters, it’s appropriate to clarify what an agent looks like. In essence, it’s an LLM or LRM with many tools available. These tools are in two forms:

  1. Code functions. For instance, if you want the model to perform calculations using a calculator, LLMs will make simple mistakes even in elementary calculations, like addition, due to their stochastic nature. Thus, the model calls an ‘addition function’ that returns the correct result always, and we simply have to append the result to the model instead of making the prediction.

  2. APIs. In some cases, you might want your LLM to interact with third-party software. This software provides an interface with which LLMs can interact through code without actively using the website interface (the one humans use). This unlocks much more powerful features, but APIs require an extremely rigid structure that LLMs must always follow.

And what type of use cases can we build with APIs?

For instance, workflows such as the model listening to a customer complaint through the phone, extracting the key insights (customer name, ID, reason for complaint, etc.), building the API payload (the information you want to send to the dispute management software), and actively sending the data to the software so its logged into the system with no human intervention is the kind of use cases that robust agents unlock.

While the first set of functions has been widely available to LLMs for over a year, APIs, the true unlocker of agentic behavior, have been out of reach due to LLMs’ incapacity to generate consistent outputs.

However, that has changed recently.

Moving on, models are also much brighter thanks to the transition to inference time-compute, with models like OpenAI’s o1 models. I'm not going into much detail because I already did, but LRM-based agents are smarter and, thus, can perform a broader range of actions.

The third catalyst for agents is real-time APIs we mentioned earlier like the one OpenAI just launched in DevDay, which I covered last Thursday. Having almost no latency in speech-to-speech interactions with AI unlocks several agentic cases, especially disrupting entire industries like customer support.

All this leads us to the great question: Is SaaS Dead? And which companies are in the greatest peril?

Vertical LLM Agents, The Industry’s Future

The appetite for a new set of agent tools that combine effortless interactions (I declare, AI executes) coupled with the uber-comfortable usage-based pricing, where customers only pay for what they consume, has many investors crazy to find the first company that offers vertical LLM agents.

And in this new era of software, we already have several remarkable usurpers to the throne. In the era of declarative software, companies are building agents that specialize in the workflows and processes of a given sector or industry, turning employees from doers to thinkers.

And while we are focusing the discussion on SaaS companies, they aren’t the only ones affected by this change; customers will also need fewer sales reps if each sales rep now outputs 10x. This means everyone will need fewer employees, which also affects you.

Subscribe to Full Premium package to read the rest.

Become a paying subscriber of Full Premium package to get access to this post and other subscriber-only content.

Already a paying subscriber? Sign In.

A subscription gets you:

  • • NO ADS
  • • An additional insights email on Tuesdays
  • • Gain access to TheWhiteBox's knowledge base to access four times more content than the free version on markets, cutting-edge research, company deep dives, AI engineering tips, & more