Inverting SaaS: Services as Software
LLMs represent a phase change in software that could remake the SaaS business model. What will software companies look like a decade from now? And who stands to benefit?
LLM advancements and the inversion of the SaaS business model
Last year saw massive advancements in GenAI, and soon, we’ll finally see LLMs and large multi-modal models become capable of consistently producing outputs on par with skilled, white-collar workers.
When tech advances rapidly, business model innovation follows. LLMs represent a phase change in software that could usher in a new form factor for software products and remake the software business model. In many pockets of the market, LLMs meaningfully reduce the need for software companies to sell workflow tools to customers: the traditional SaaS model. Instead, software companies will “sell outcomes” to organizations, functioning much more like today’s services businesses and BPOs. New “services as software companies” will offer services in the front powered by AI-enabled software in the back.
Step-change model improvements and their ability to make programmatic use of unstructured data will power the shift, and customers won’t need to learn a new procurement motion: IT services spend already outpaces software spend. But the business model transformation won’t happen overnight. Big changes take time, models must become more capable of producing consistently intelligent outputs, better hardware and model architectures must drive costs down, and AI vendors must invest in better enterprise readiness around data security and compliance. Still, expect many SaaS companies to look more like services ones a decade from today. What will this mean for the software business model?
Changing the entire business model: pricing, GTM, cost structures, and more
Monetize outputs, not seats
As SaaS looks more like AI-enabled services, software companies will shift away from seat and usage-based pricing models. Measuring the impact of workflow tools on fundamental business outcomes (revenue expansion, cost reduction) is challenging today. By delivering outcomes, LLMs enable pricing models that much more closely align revenue generation with end-customer value created, rather than models that charge by the individual user or amount of customer resource consumption. This doesn’t necessarily mean software companies will capture more of the value they create – Microsoft, arguably the software GOAT, has captured only a small fraction of the value creation that Office enables. But it does mean that products will generate much clearer customer ROI.
Some companies can price outputs directly, particularly in spaces where their software replaces employees whose outputs are clear, discrete, and enumerable: generating sales leads, driving downloads via marketing campaigns, closing customer support tickets, sourcing candidates, etc. For instance, customer support software might price per ticket successfully resolved.
Outcomes-based pricing will be less straightforward for other companies, often those replacing G&A employees who complete processes rather than delivering discrete outputs: administering payroll and benefits, closing the books, managing internal FP&A, providing advice on basic contract negotiation and drafting, etc. Identifying a price-able atomic unit of output might be too challenging, so these companies will iterate with customers over time and monetize their offerings as a function of expenses saved (likely a small %) via automation.
In the future, expect software companies to de-emphasize seats and usage, instead prioritizing outcomes – and value – delivered.
Less PLG, more enterprise sales
Bottoms-up GTM (particularly PLG) was last cycle’s darling: it enabled rapid and viral growth, high gross margins, and attractive CAC payback dynamics. Some of the early prosumer “GPT wrappers” adopted PLG because they demonstrated lightning-fast time-to-value. But as companies re-focus on long-term durability and ultimately sell outcomes, we’ll see a shift back towards top-down enterprise sales.
PLG is not a viable model for vendors selling organizational outcomes to executives rather than tools to rank-and-file workers. This is in part due to their sophisticated implementation motions: integrating with key data sources, ensuring data privacy, and configuring products for specific customer needs. It doesn’t help that many employees might view these products as a threat to their jobs, making selling to them a political non-starter.
I don’t believe PLG will die. The strategy will work for some companies, potentially those who (1) sell to SMBs or (2) facilitate interactions between ecosystem participants, creating a natural product-led flywheel. But PLG companies were already encountering headwinds due to a challenging macroenvironment: vendor consolidation, CFOs clamping down on self-procurement, a “frictionless in, frictionless out” dynamic, etc. We’re already seeing fast-growth companies like Harvey and Hebbia scale rapidly by focusing on enterprise sales, and I expect selling outcomes will only accelerate the shift away from PLG towards a top-down, land-and-expand motion:
Sell to senior executives excited about the product’s concrete business impact
Lead a narrowly-scoped product rollout to demonstrate the value prop
Expand into the rest of the organization and iterate on pricing (based on actual outcomes) to more closely align with value-add
“Zero marginal cost of distribution” no more
SaaS companies make large up-front investments and have high fixed costs that afford them significant operating leverage: nearly every dollar of incremental revenue hits their bottom line.
As the SaaS model inverts, software companies will have lower gross margins and less operating leverage. Delivering and monetizing outputs will incur both traditional cloud infra costs and new inference costs as well. Many companies – particularly those with a services-first posture and significant human review – will incur wages too, especially for the foreseeable future, before LLMs are capable of full task automation.
Even if vendors ultimately pass through costs (like usage-based pricing), services gross margins will lag those of SaaS. And because these COGS scale linearly with outcomes-based revenue, the new software cost structure will be much more variable (with a smaller share of up-front and fixed costs), and the business model will have less operating leverage.
Expect these changes to have impacts on software balance sheets and market valuations.
Capital structures and valuations in the services era
New pricing models may make software revenue less predictable than the contracted, recurring, seat-based revenues that define SaaS. But outcomes-based revenue might not necessarily be less sticky. If the new software form factor approximates labor – with “churn” analogous to “layoffs” – products might be quite sticky after all. This is especially true given their complex implementation cycle and how deeply integrated vendors will be in organizational systems and data sources. AI-powered services won’t be easily ripped out.
Sticky revenue and a cost structure with fewer fixed costs will enable AI-enabled services companies to employ more debt than SaaS ones. High operating leverage in the SaaS model makes financial leverage very risky: in contractionary periods, revenue falls, costs don’t, and the ability to service debt collapses. Expect the SaaS inversion to favor companies that responsibly use leverage to boost ROE.
All things considered, it’s likely that these changes – less topline predictability, lower gross margins, and a less viral (and more expensive) GTM motion – will weigh on valuations. Software investors should behave accordingly.
Business model innovation and value accrual
As an investor, I’ve been considering which businesses will benefit from the SaaS inversion using a three-pronged framework:
Will building valuable services “apps” require (1) weaving intelligence into existing products or (2) that products and user experiences be wholesale rebuilt?
If the latter, can startups building services with outcomes-based pricing achieve meaningful distribution before incumbents innovate? If not, despite disrupting incumbent business models, new tech will ultimately sustain their positioning.
If so, can these startups build competitive advantages vis-à-vis other startups? If not, startups may not become durable winners despite out-competing incumbents.
Inverting SaaS and selling outcomes certainly changes software pricing and packaging: it’s disruptive. Disruption usually favors startups who counter-position against incumbents suffering from a classic innovators’ dilemma. But incumbents have moved unusually quickly and made GenAI a priority; most also successfully navigated the last major transformation in software business models. There’s a reason markets have largely coalesced around the idea that GenAI will ultimately be a sustaining innovation in software.
Some incumbents will inevitably stumble during the during the transition and open the door for startups. Even still, startups face an ultra-competitive battle against each other. Dev tools, cloud infra, and widespread playbooks drove down the cost of producing software in the 2010s, and new software copilots only accelerate the decline. Startups today can build better, faster, and with fewer resources. While “moats” are often the product of compounded daily execution rather than grand strategy, neither a head start nor excellent engineering execution will be as strong a competitive advantage as in the past. And while the SaaS inversion is a source of disruption, it’s a double-edged sword: workflows create process, process creates lock-in, and shifting away from selling workflows reduces one source of switching costs and long-term defensibility.
In the last cycle, the costs of building workflow SaaS created barriers to entry, providing monopoly profits for scaled, durable incumbents – they captured value by enabling people who delivered outputs. As SaaS inverts, LLMs that produce outputs directly upend the value capture paradigm. Moving forward, value will accrue to those who own end-customer relationships, which are difficult to dislodge and enable them to capture a greater share of underlying economy activity.
A golden opportunity for incumbent service providers
Investing in legacy service providers isn’t my domain. But incumbent BPOs are well positioned for a world where SaaS and services converge. They already sell outcomes and don’t face an innovators’ dilemma. Although services are lower margin (wages), less scalable (worker recruitment and retention), and more operationally complex than SaaS, AI-enabled products meaningfully reduce these challenges. VCs are plowing money into vertical AI SaaS solutions, so incumbents will have many cheap options at their disposal. Some might even build solutions in-house as the cost of developing software declines, especially given they won’t face the expensive R&D and S&M pressures of the venture model. They have rich, proprietary datasets and access to employees (a source of RLHF), both of which will power valuable AI offerings. Importantly, unlike many of the software companies selling to them, they directly own end-customer relationships.
Legacy players will face a “thrive or die” moment. The train will soon leave the station: embrace these tectonic advancements and reap the rewards or get left behind. Those who reinvent themselves have a golden opportunity. With AI, they can offer better products at faster speeds, improve their scalability, and boost margins. And with incremental profits, they’ll return capital to shareholders, reinvest in organic growth, or grow inorganically via acquisition.
I’m bullish on incumbents who (1) take a forward-thinking posture and aggressively adopt GenAI and (2) sell to end-customers who are already accustomed to outsourcing entire business functions to service providers (HR, recruiting, bookkeeping, marketing collateral) rather than solely purchasing software to manage functions in-house. I’m not alone – there’s a reason TriNet stock is up nearly 75% since the end of 2022.1
Another transformation opportunity for software incumbents
Shifting from on-prem to SaaS didn’t kill software incumbents, and neither will the coming inversion of the SaaS model. While some will unsuccessfully navigate the business model transformation, I expect best-in-class software platforms to thrive.
They’re not asleep at the wheel. They already have huge customer bases, as well as proprietary data that will power the most effective LLMs and valuable end-customer experiences. And, after using the last 18 months to rationalize their cost structures, they have ample resources to invest in product development just as funding has dried up for potential competitors without clear long-term durability.
Specifically, I’m bullish on durable companies who (1) offer scaled, defensible platforms (2) have sustainable cost structures (3) can successfully navigate complex, top-down sales and implementation cycles (4) moved quickly to adopt LLMs over the last year (5) show willingness around experimenting with outcomes-based pricing as LLM capabilities advance and (6) currently serve end users in functions that aren’t usually outsourced (sales, strategic finance, high-value software development). These companies effectively “own the end customer” and are less likely to lose business to incumbent BPOs.
Where should SaaS VCs focus?
Despite structural challenges facing new startups, there’s always reason for optimism in the venture world. Every single important tech platform shift – semis, PC, Internet, cloud, mobile, blockchain – berthed many valuable, venture-scale startups.
I’ve written before about my belief that most new winners will emerge in vertical markets, which are usually (1) less crowded by incumbents with existing products, distribution, and proprietary data and (2) structurally much more monopolistic than horizontal ones. In these markets, new startups can win if they have strong domain expertise and demonstrate a credible moat trajectory: compounding data advantage, complex integrations, regulatory advantages, proprietary GTM, and other sources of durability. These characteristics lend themselves to winner-take-most (or all) dynamics, which the venture model requires.
Some folks were early to the vertical AI opportunity, which quickly became a consensus bet last year, but I’m excited about three categories of startups.
Vertical AI for the real economy
Taking “services as software” to the extreme, software companies will sell only finished work. That’s impossible in the real economy, where serving end customers requires real-world coordination: LLMs can’t automate the atomic unit of labor output. Home services and its subcategories is a great example. AI-enabled vertical SaaS startups can win by (1) counter-positioning against incumbents and adopting outcomes-based pricing (2) offering the most valuable models by aggregating proprietary data and (3) becoming sticky systems of record that customers rely on.
Vertical AI for the relationship and judgement economy
In some industries, fully “selling the outcome” won’t be possible, either because human relationships are so critical to success or extreme levels of human judgement are required, making the cost of displacing employees unacceptably high. I don’t expect to see full AI-enabled services provision in spaces like law, executive recruiting, or private investing (self-serving, perhaps). Like their counterparts selling into the real economy, vertical AI startups will win by quickly adopting innovative pricing models, building valuable industry datasets, and owning end-to-end customer processes.
Tech-enabled services have their moment
Vertical copilots are already a consensus opportunity. Many smart investors I speak to are excited about “selling the outcome,” which is quickly becoming consensus too, if it isn’t already. Still, “services” remains a dirty word in Silicon Valley – and investing in it a source of alpha, for now.
Tech-enabled service startups had a rough go in the last cycle. But LLMs have changed the paradigm, at least in the white-collar services world: AI-enabled service companies can augment (and increasingly replace) machines with humans, solving the margin and scalability challenges that doomed so many venture-backed startups over the last decade. Some winners will look like traditional startups: raise VC to fund expensive R&D and S&M investments in proprietary tech and long-term customers. I’m also excited about companies that take a slightly different approach:
Raise capital to fund R&D
Purchase a small existing service provider, providing an immediate source of data flows, RLHF, and customer ownership
Leverage those resources to meaningfully improve the product offering and start investing in traditional customer acquisition
If you’re building in these spaces, I’d love to hear from you!
Thank you to everyone who weighed in on this piece. If you have questions, comments, or feedback, please reach out: andrewziperski [at] gmail [dot] com.
The views expressed herein are the author’s own and are not the views of Craft Ventures Management, LP or its affiliates.
Data as of 1/10/2024