Startups focused on selling services have historically been viewed as uninvestable by the venture capital community. They typically have a weak margin profile and the rate of revenue growth in white-collar services is often constrained by the need to continually accrue specialized human capital in a manner you don’t see in software businesses. Until recently, during pitch meetings with startups who monetize through a blend of services and software revenue, founders were always quick to highlight how that ratio would tilt heavily towards SaaS as the business scales - with services revenue eventually becoming an afterthought. This bias against service revenue is rapidly fading as advances in artificial intelligence are enabling a new business model - companies that sell professional services via a software interface. In the same way investors flocked towards ‘Software as a Service’ over the last ~15 years, a similar wave of interest has now formed around ‘Services as Software’.
To zoom out for a moment, the existing landscape of AI business applications can be bucketed into three general categories.
- AI copilots that make workers more effective but are still driven by people - some of these are horizontal (i.e., MS Copilot) and some vertical (i.e., Hebbia)
- AI agent frameworks that handle a discrete set of tasks autonomously - typically vertical and often centered on customer communications or back office administrative tasks (i.e., Harmonyze)
- AI services businesses which provide a specific output on demand (i.e., EvenUp)
The first two buckets both look like traditional software - AI copilots are analogous to last-gen internal knowledge management tools and AI agents can be compared to existing security and monitoring tools that offer a ‘set it and forget it’ dynamic for the user. AI services businesses look different from these first two groupings. Their focus is on executing a subset of industry specific tasks on-demand with a higher degree of speed and accuracy than a human would be capable of. At times they can rely on agents to produce their outputs but they will be acting beyond the view of the user.
The AI Services Opportunity
We believe the AI services wave will be analogous to the impact outsourcing had in the 1980s and 1990s, except these software packages will be capable of much more complex work than was typically delegated to the last generation of BPO firms. And the most successful AI services software will be verticalized - these businesses will have a specific category focus and will try to build a data moat by aggregating and training on industry-specific inputs - as they get bigger and accrue more data, this data advantage can become self reinforcing and will offer them an edge in terms of output quality.
Since the advent of LLMs, we have seen a v1.0 wave of AI services businesses - some of the business outputs they are capable of generating include:
- Drafting Legal Documents
- Drafting RFP Responses
- Drafting Industry-Specific Compliance Reports
- Drafting Prior Authorization Requests
- Drafting SOAP/OASIS Submittals
- Drafting Permitting Requests
- Drafting Construction Estimates
- Drafting Title Search Reports
At their core, these ‘Software as Service’ businesses will offer an application where users load in documents or recordings and then receive an output similar to what is produced today by consultants, offshore resources, and in-house staff. Some AI services businesses will complete the entire task, while some get you 80% of the way there and keep a human-in-the-loop to give the user a degree of discretion over particularly complex or revenue generative tasks. As the underlying technology improves, we would expect AI services businesses to move more towards full automation with humans increasingly playing a smaller role largely focused on QA and oversight.
Who Wins in AI Services?
It is yet to be seen how much of the value of AI services will accrue to pure play software companies vs. traditional professional services organizations. Some legacy professional services providers will move fast enough to re-make themselves into AI-enabled v2.0 consultancies that run much leaner and operate with much higher margins. Ongoing advances in AI - namely the decreasing costs of LLM access and the commoditization of their performance - means some service providers that own these functions today could hypothetically innovate around AI to achieve the kind of margins historically associated more with SaaS. However, most of these firms will likely move too slowly towards this new model, making room for software-native challengers. These software-native challenges will also benefit from these same tailwinds around decreasing cost and increasing performance of LLMs and will be able to outcompete on price, being free of any legacy cost structure.
Who wins will also depend on how LLMs continue to advance and how effective they are at performing services vs. how much human touch-up is needed on the outputs - if advances continue apace, this favors software, if it doesn’t, that would be another opening for tech-enabled consultants.
We also see some entrepreneurs buying old line businesses (i.e., a title insurance firm) and then trying to infuse them with AI to undercut the competition. As software investors, we are inclined to focus primarily on software-style bets but we are in the opening chapter of AI’s creep into services and the market is still trying to determine the best way forward.
As things stand, we believe AI services firms that position themselves as a tech vendor to businesses will ultimately win and will have the best unit economics and most of the early breakout successes follow that model. Professional and business services account for ~13% of GDP and today’s AI capabilities are well suited to reproduce much of the services work product that focuses on document and desktop analysis. These AI services businesses may look very different from SaaS and may rely more heavily on consumption based pricing and delivery models. While this raises questions about revenue durability, companies building AI services software will have the opportunity to build wide user suites around their core on-demand services that can drive lock-in and potentially even lead to a subscription component.
The Text to Multimodal Shift
Most of the AI services businesses that have launched in the last 18 months operate on a text-in, text-out framework. As analysis of multimodal inputs (i.e., video, images, etc.) improve, that should broaden the range of services businesses at threat by AI and expand the kinds of challenges they can tackle. This has the potential to move AI services capabilities from white collar to blue collar with many new use cases around things like claims adjustment, site monitoring, maintenance checks, and more. We are excited about the AI services opportunity and will be digging further into relevant use cases over the coming months.
If you are building AI-native services software, I would love to hear from you - patrick.mcgovern@bowerycap.com.
If you liked “AI & The Services as Software Era” and want to read more content from the Bowery Capital Team, check out other relevant posts from the Bowery Capital Blog.
