The Siren Song of Services

A cautionary guide for Services-as-Software + vertical AI pioneers

Services have taken on a whole new meaning in VC land. As we’ve discussed in recent essays, private capital firms are embracing service-forward models, not merely as an enabling tactic (e.g., white-glove onboarding or integration) but as a primary investment thesis. There are some arguments that modern pure-play services can achieve high multiples at scale, justifying venture (or at least growth & buyout) investments.¹ We see the most compelling varietals for early-stage VC, however, as “Service-as-Software” plays. Such startups attempt to maintain the benefits of SaaS—high margins, low marginal costs of growth—while capturing the potentially larger TAM of a service market (or of a more vertically integrated play which naturally includes services functions). These “Service-as-Software” businesses are multiplying at a rapid pace, in the footsteps of break-outs like Abridge and EvenUp.

As with any venture thesis in vogue, however, an antithesis is looming. Today, we are seeing a broad embrace of startups tackling services, many without clear paths to automation or without a natural venture-scale second act. Some bet that AI will catch up, and margins with it. Others trade away the software business model wholesale in exchange for TAM, taking vertical integration to its extreme. Far too many are forgetting why software is such a hallowed category of investment, making all-to-liberal assumptions around the ease of automating humans out of an industry, the down-market valuation of services revenue, or the right to grow a product suite horizontally. In today’s essay, we will discuss the promise and pitfalls of Service-as-Software businesses, concluding with our framework for vertical software & AI founders considering product strategy in service automation.

The Rationale Behind Services

A common rationale for Service-as-Software is that—due to a number of factors, including the ballooning number of SaaS tools and disincentives to innovate in some industries—the traditional model of selling and monetizing cloud software has hit a wall. In order to achieve further penetration, startups must reposition from SaaS to services, delivering results rather than tools and, in some cases, wholly subsuming the operating businesses they once saw as customers. Armed with the potential of LLMs and agentic platforms to up-level automation and bring expanded gross margins to traditionally manual workflows, they posit, services markets are ripe for the taking.

Proponents might argue this shift mirrors past step-changes in the evolution of software delivery. SaaS 1.0 businesses supplanted traditional expensive on-premise software licensing models by offering a cheaper, more flexible pricing structure. In 2000, Salesforce launched the Eve of SaaS subscription plans: a $25 / month Essentials offering that made competitor Siebel Systems’ upfront licensing investment look usurious.² Thanks in part to the attractive recurring nature of its revenue, Salesforce went public in 2004 and garnered revenue multiples 3x of traditional on-prem peers.³ SFDC was doubly attractive for its ability to grow TAM: subscriptions lowered the barrier to entry, reduced cost of ownership in early years, and could grow with a customer from its earliest days. If we could consider on-premise software a sort of evolutionary bottleneck for software, what followed was a burst of adaptive radiation in SaaS, powered by the possibilities of the cloud and continuing to this day.

The ability to sell SaaS into certain categories and segments, however, may have limits of its own. Some businesses are be hesitant to take on more OpEx or grow IT budgets; some verticals aren’t used to paying on a subscription basis for products; in some cases, incentives to buy are weak for gate-keepers with monopolistic control and / or high complacency. Admittedly, we feel these concerns around the “death of SaaS” are overstated. Public SaaS earnings and net new ARR have reverted to the mean since nadirs in 2022/2023, with a particularly strong recent Q3.⁴ Cyclical fragmentation (best of breed point solutions develop outside platforms) and consolidation of software (many vendors breeds a desire for a single pane of glass) in a given space is a well-known historical phenomenon. And the overall SaaS market continues to grow at a near-20% CAGR.⁵ We shouldn’t derive generational conclusions from a data points gathered over a fairly unique preceding 3 years.

All that said, if startups “sell the work” rather than SaaS—to borrow a term from Sarah Tavel at Benchmark⁶—perhaps they could avoid these concerns altogether, whether cyclical or enduring. Like SaaS’ disruption of on-premise software, Service-as-Software startups reimagine product delivery and monetization to reduce friction of adoption and grow addressable market. If this business model shift could be even a fraction as impactful, the hype we’re seeing today isn’t misplaced.

Benefits: Positioning, Ease of Adoption & TAM

There are several potential benefits to the Service-as-Software model. First, by positioning as a service, a startup can appear to customers as just another vendor in a budget category like professional services rather than IT. If positioned as a natural replacement for an existing vendor, the service may present less friction in switching than replacing an existing service with a software product or tool that requires deeper implementation, education, or training.

To use a well-worn example: a SaaS company sells a solution to help dental practices streamline insurance billing. The challenge may be that only large practices and DSOs would use it because the manual lift of self-managing billing was too much, incentivizing outsourcing. A Service-as-Software model, however, might simply sell the work, competing as would any other 3rd party billing service and monetizing on a percentage of billings. Clients may perceive this latter model as an easier buy because they are familiar with the concept of 3rd party billers, and they don’t have to train staff on something new. Not to mention SMB customers might now be back on the table.

For a Service-as-Software startup, it’s easy to see how deeper vertical integration could offer greater TAM expansion. Instead of creating software that serves as an operating system for accounting firms, for instance, some startups are now positioning themselves as de novo accounting firms. The difference with the new model—let’s call it NewCo Accounting—being higher-than-typical target margins thanks to planned automation of human labor. A SaaS 1.0 startup might capture only 1-10% of a business's revenue, companies being unlikely to spend more on an OpEx cost center. NewCo, of course, would capture 100% of the client’s accounting spend. The trade-off is that while the accounting SaaS startup would likely have margins between 80-95%, our software-powered accounting firm might achieve margins more like 70% (assuming our bet that AI can automate significantly is good).

To illustrate: consider an accounting firm generating $10M in revenue per year. It might spend, at most, $1M a year on software—this is the opportunity for a SaaS startup. Alternatively, the software-powered operating startup could capture the full $10 million. The SaaS startup would make $800-900k in gross margin, while the vertically integrated services startup might see $5M in gross profit, at typical accounting industry margins. Factoring in the potential of AI-powered software to contribute to margin expansion makes the picture even rosier. Let’s imagine the ability, over time, to automate from 50% to 70% gross margins. Its now $7M in gross profit would looks quite attractive compared to the SaaS company’s per-firm gross profit of less than $1M. Of course, the ultimate question is the ratio of “market share” between the two models—SaaS may be able to sell into firms touching many more paying end users than any one accountancy brand, AI-powered or not, which influences TAM massively. But we can see how, in verticals challenging for traditional SaaS, some believe Services-as-Software expands TAMs.

An AI maximalist might put this all more simply: the replacement of human labor comprises a $4.6B revenue opportunity.⁷ It’s easy to see why services markets present such an alluring target—and should their pitfalls be ignored, perhaps a siren song.

Challenges: Automation Risk, Competition & Revenue Quality

This tradeoff of TAM for margin risk, however, is not as clear a bet as it might seem. First, we should recall that despite all the excitement around LLMs, the extent to which they can replace human labor (outside of few, albeit impressive use cases), is very much up in the air. In an ideal world, as Dan at Deibel outlined his recent post, “Managed Service-as-Software,” humans should be able to bootstrap AI startups while they fine-tune automation (assuming labor costs outpace GPU costs for inference).⁸ Timelines are idiosyncratic and unclear. Bandied comps still feel like marketing and overstatement (e.g. Klarna’s claims of eliminating 2/3rds of its customer support costs in partnership with Open AI).⁹ In our past essay, The AI-First Roll-Up, we touched on this point:

[It] remains unclear what the margin lift from AI will be—and opportunity across industries is unlikely to be universal… [Service-as-Software companies are] selling an outcome and internalizing the margin risk. I.e., they are making a bet that they can automate manual an increasing share of work over time. These have been famous last words for many startups whose margin uplift materialized too slowly because it was (i) too hard technically, or (ii) too hard to escape reliance on services once addicted to the high growth rates their venture backers demanded.

Second, we also need to consider realistic penetration rates at scale. Horizontal SaaS companies often slow down around 10-20% market share. Vertical SaaS, especially in less competitive markets, may achieve much higher penetration, sometimes over 80% within certain (albeit often PE-consolidated) focused markets. Few operating businesses (and hence fully vertically integrated startups), however, will ever reach these heights. Despite their incredible growth, none of Headway, Alma, Rula, or Grow Therapy are estimated to have low single-digit shares of the total US therapy market.¹⁰ Even after its roll-up of the largest legacy player in the market, vertically-integrated home health business Honor likely has 2-4%.¹¹

Beyond the uncertain lift of automation, the price a fully vertically integrated Services-as-Software startup pays to expand its TAM can be a significant increase in its competition. And while a 5% share of a massive market is nothing to shake a stick at, the math doesn’t always pencil out against SaaS. Let’s take a $50B market with 50k firms as an example. A vertical SaaS company might achieve 20% market share at a $100k ACV, resulting in $1B in revenue on a 90% gross margin. By contrast, an operating startup with a 2% market share would see $1B in revenue at 50-70% margins. Obviously, these dynamics are highly market and product dependent, but the point stands that a bigger TAM isn’t always the answer.

It’s also crucial to consider revenue quality differences between the two models. SaaS revenue is considered high-quality because it’s recurring and has limited marginal cost to growth. Even assuming Service-as-Software businesses automate their way to SaaS-like margins and maintain high retention, there are many more potential vectors of cost to its expansion trajectory. As discussed above regarding automation risk, unscalable elements exist in almost every industry—most especially those that are highly fragmented, regionalized, or dependent on trust. Trusted agent dynamics are common in high-stakes advisory services (e.g., RIAs, M&A advisory, legal, consulting, accounting, and RE brokers). Beyond brand, the human relationship is the moat. These intangible limits to automation can present the most intractable challenges for fully vertically integrated Service-as-Software plays.

Moreover, startups that overextend on scope and fail to achieve software-like automation will receive a significant discount on enterprise value. SaaS businesses receive a substantial premium in public markets due to high revenue quality and future profit potential. Investors also tolerate prolonged unprofitability in SaaS because a recurring model has high lifetime values relative to customer acquisition costs, assuming retention holds and fixed costs eventually shrink on a relative basis. Currently, the median SaaS company trades at 7x ARR, while service businesses average around 3x revenue—meaning a service business would need twice the revenue of a SaaS company to achieve the same valuation. What multiples can Service-as-Software models achieve? Naturally, it must also depend on their margins, retention, and cost of growth. These, in turn, will be determined by the selection of scope in vertical integration—choosing to subsume those business functions that truly can be automated while steering clear of those that cannot.

Success is in the Scope

Service-as-Software players that do not choose their vertical integration scope and wedge carefully are bound to follow the path of most venture-backed service startups before them. Success—at venture-scale anyway—will come down to a careful balance of two key factors:

1. Ability to Automate: The label “services business” first brings to mind a product dependent on human labor. Think McKinsey, Kirkland & Ellis, KPMG, WPP, or Capgemini. The additional implication here is higher marginal costs due to personnel as a linear input. Compare, for example, the operating margins of the examples above (WPP: 13%, Capgemini: 10%) vs. mature SaaS companies (Veeva Systems: 31%, Salesforce: 24%, Autodesk: 22%). Low-margin services are not inherently bad—even for core SaaS, they play important roles in sales enablement, customer success, and as supplementary revenue drivers at scale. Perpetually low-GM core businesses, however, are generally a poor use of venture dollars. Compelling Service-as-Software plays should seek to capture the revenue of old services while leveraging technology to achieve software-like margins. As we will discuss below, the automation of manual workflows can be more complex than expected or even difficult to envision in the near term.

2. Scope of Vertical Integration: Whereas traditional SaaS provides users with a tool to perform work, a service business delivers the output (or “sells the work”) without the need for clients to adopt a new tool. Of course, there are many non-services businesses that sell an output and monetize accordingly, from Twilio to Eleven Labs. In this context, however, we are more so talking about vertical integration. A services business absorbs more of the value chain by providing more than just a tool to do the work. This new delivery model may enable them to capture a higher share of industry dollars but it may also require business functions that are less automated or profitably scalable than pure software. Selecting how much of / which elements of a stack to absorb, therefore, is an extremely important and nuanced consideration. Not only to control automation & margin risk, but also to ensure a wedge isn’t where the opportunity ends. We have seen how a vertical software “operating system”—centered around, say, an CRM or ERP—can expand its suite to peripheral SaaS modules or even new revenue streams (e.g. payments, procurement). Systems of record leverage their data access and stickiness with users to pave the way for additional features. “Selling the work” means your platform may not be a “daily driver” for users, or may not even have a customer-facing UI.

This balance is inherently different for each permutation of service offering and industry. Selection should be driven by two core factors: (1) the ability to automate and (2) the vertical-specific need, pairing high market pull with minimal competition or substitutes. The below graphic illustrates how we think about the decision framework for vertical integrators and those selecting Service-as-Software wedges:

The reason we emphasize vertical differentiation is to highlight the importance of moat in AI-driven service displacement. Pricing is a powerful initial differentiator, especially in the case of what we call Democratized Services: products that can not only undercut competitors but expand TAM by capturing priced-out demand. Historically, the closest analogs to democratized services have been mostly-integrated consumer or prosumer plays, leveraging a marketplace or contractor network to keep labor outsourced (e.g. Uber, LegalZoom, Fiverr, UserTesting). Before them, we saw effect in retail companies like Amazon, Walmart, and Ford (with the Model T), which achieved pricing disparity through mechanization, supply chains, and massive scale. All these businesses bring major price reductions to markets with significant priced-out demand, opening up a massive “blue sea” within a larger red ocean where legacy players can’t compete.

Unlike those historical examples, higher-margin, faster-build, less capital intensive opportunities may be unlocked by AI. The problem is that ease of automation is nice until others follow suit and do the same—which they will, whether legacy or startup. Walmart and Amazon built physical networks that remain near-impossible to replicate. Uber built a digital and human network that is, as Lyft showed, replicable but at great difficulty (until humans are replaced anyway, but Waymo is a story for another day).

AI-driven democratized services, so far, have demonstrated an incredible ability to grow quickly, from ChatGPT growing to 100m users in 2 months to Abridge reaching an anecdotal $100m in revenue in 2 years. Thus far, consumer-facing retention has been poor, with mobile AI-first solutions (e.g. ChatGPT, Lensa, Remini, etc.) reporting median retention about a third lower than top B2C apps.¹² Break-out B2B democratizers may well be able to parlay incredible traction (and therefore lots of capital) into more defensible integration paths. Most Service-as-Software entrepreneurs, however, should take note to understand the source of their differentiation, from wedge to Act Two. As discussed in our essay, The Future of AI is Vertical, we believe the indisputable pillars of AI defensibility are software (workflow) and (proprietary) data.

Winning Models

Winning Service-as-Software models, therefore, should be those that offer the flexibility to scale into vertical integration where and when it results in a stronger business. Taking on services just because SaaS is hard and LLMs exist will be a fast track to a low-margin, low-moat, poorly valued enterprise, even in cases where early traction is impressive.

As discussed in past essays, the Synthetic Roll-Up encapsulates archetypes we believe work best for vertical B2B plays. While not fully vertically integrated—nor even necessarily sold exclusively as a service—it retains the key benefits software, while expanding addressable markets and reducing adoption friction. And while they may increase vertical integration over time, some business functions—those lower margin and more difficult to automate—can be left to industry partners, where their focus on the human relationship can thrive. Those partners generally require a (likely SaaS) UI in addition to LLM-driven features, creating some level of workflow and data gravity—all contributing to proven forms of defensibility. Using this malleable approach, we envision Service-as-Software platforms making the most of a significant business model shift while avoiding the siren song of services.

Thanks for sticking with Euclid Insights. We’d love to hear your feedback and additions in the comments below. If you or anyone in your network is thinking through a vertical service-as-software business, please drop us a line directly. We’re happy to serve as a sounding board and be helpful as early as idea and formation stages.

1. Firms like Insight and BVP are on the forefront of this resurgence, spinning up dedicated teams. Informed by conversations with investors at those firms.

2. Though, of course, Salesforce’s Total Cost of Ownership (TCO) would eventually catch up with and exceed any one-time license alternative, eventually.

3. As of 1/1/2005, Salesforce traded at 15.8x EV / LTM Revenue vs. 5-6x for SAP & Oracle and 2-3x for Siebel and PeopleSoft. Much faster growth factored in as well. Per Pitchbook (2024).

4. Clouded Judgement 11.8.24 - Q3 Earnings + Fed Update

5. Lihn, Skriver (2024). Software-as-a-Service Sector Report Spring 2024: Will the Early Signs of Tailwinds Continue?. HC Andersen Capital.

6. A few "Sell Work, Not Software" updated thoughts

7. Chen, Gupta (2024). AI Service as Software. Foundation Capital.

8. Nguyen-Huu (2024). Introducing the Managed Service-as-a-Product Framework. Founder Catalyst.

9. Oroscz (2024). Klarna’s AI Chatbot. The Pragmatic Engineer.

10. Single Aim Health (n.d.). Growth of the Healthcare ‘Business in a Box’ Model for Therapists. Single Aim Health.

11. Donlan (2021). Honor to Acquire Home Instead, Creating $2 Billion Home Care Services Company. Home Health Care News.

12. Huang, Grady (2023). Generative AI: Act Two. Sequoia Capital.