The AI Services Wave: Lessons from Palantir in The New Age of AI

Last month, the first company I co-founded, Palantir, joined the S&P 500. For most of 20 years, the naive mainstream view of Palantir was that it was a “glorified consultancy” – a services firm and not a real tech innovator building SaaS “products” or “platforms”. To dismiss Palantir early on was short-sighted, given they’d hired some of Silicon Valley’s top tech talent, but it was based on a factual observation: unlike most software businesses, many of our engineers spent significant time working alongside our customers. We called this team “Forward Deployed Engineers”, and they obsessed over the intricacies of our customers’ daily work, business models, and pain points. 

We started this way by necessity. While Palantir had a strong product, including an advanced data integration platform, revisioning database, and more, we (unsurprisingly) didn’t have much experience with the complexities of workflows or data sources of our early defense and intelligence customers. That necessity became our unfair advantage. We iterated and extended our platform alongside our customers, improving their mission effectiveness by understanding their needs, and improving our products by abstracting those needs into universal components we could then apply to other problems. We ‘productized’ our services. Year over year, the labor and configuration required to establish value have decreased, while the scope of what Palantir can do has exploded. What’s been consistent throughout is that services have supported product revenue, not vice versa. This remains unintuitive for people who think a business must be one thing or the other.

Our success relied on boldness and talent, and a unique approach to mapping customers’ data and processes and encoding it in our solutions. We call this map an ontology, and it enabled us to serve the world’s largest organizations, where outcomes require complex workflows across numerous functions, systems, and data formats. But it’s only possible through a depth of customer partnership unimaginable to classic “product” companies. A “product” approach works best when solving a common pain point for countless customers (like payment processing). When serving the world’s most complex organizations, a special operations “services” mentality is what actually lets you build complete solutions that, in the end, are worth more than out-of-the-box products.

‍Services are cool now?

With the arrival of LLMs, mere “services firms” have become much more exciting businesses. At 8VC we’re seeing extraordinary results from integrating AI with best-in-class operations to transform the economics and scalability of B2B service industries. AI can automate messy, language-based human workflows, and is often even better than humans at catching mistakes or dealing with edge cases. This is not true universally - there are still lots of places where human judgment or creativity is important (try asking ChatGPT to suggest team-building exercises). And so a hybrid business model often ends up the best way to solve the real problem the end customer has, while letting you capture more of the value you create in the process. 

Software might have eaten the world in the last decade, but it left plenty of legacy services industries on the plate. We estimate that the American services sector spends $5T on wages in AI-exposed job functions every year. In many of these industries, total factor productivity has gone down even as IT spend has grown. It’s not uncommon to find incumbents with largely paper-based or on-premise systems of record. When SaaS has been adopted, it is typically for non-core parts of the business and rarely used to its potential. We see an opportunity not just to sell more software to these businesses, but to compete directly. Over the last few years 8VC has invested in full-stack challengers in healthcare billing (Candid); freight audit and pay (Loop); insurance third party administration (Reserv); and lead generation (Landbase). We founded the tech-enabled architecture firm Pantheon out of 8VC Build, and many more to be announced.

‍

These startups have a similar offering to incumbent competitors, but use AI and a vertically-integrated technology stack to drive much higher margins and deliver unrecognizably better service (faster / cheaper / 24-7). Breaking down the workflows performed in each of these professions, we estimate that there are over a trillion dollars in support, back office, operations and sales wages that companies could recapture as profit by improving margins with automation, doubling or tripling productivity. Billing workflows, accounting, recruiting, legal, insurance, and many more are prime candidates for such disruption. We also see big opportunities in less obvious industries, like higher education – where Campus is providing higher-quality community college at higher margins. These businesses are also more scalable than incumbents, and can grow efficiently with a mix of organic and inorganic strategies.

‍What exactly is an “AI Services” company?

A lot of the recent AI discourse has focused on the concept of “services-as-software” and rightfully so – some of the fastest growing startups in the world right now (including 8VC companies like Fieldguide) are taking advantage of software’s newfound ability to go after labor spend as part of the value prop, instead of just serving as an enabler. When model performance alone is not quite sufficient for the task, these are often sold as “Copilots” that proactively guide end users to superior results. A prime example is Numeric’s Technical Accounting AI, which lets in-house accountants skip to the first draft of memos, reports, and other products. In some cases, existing workflows can be largely replicated by “Agents”, allowing process owners to focus on higher-order priorities – like Tezi, which helps in-house recruiters elevate the candidate experience and scale without new headcount.

There are plenty of industries where approaches like this can work if you build a great product and capture market share quickly. It does come with two big risks. First, because anyone can leverage the models, your product is typically quickly under threat from competitors — including incumbent SaaS winners your potential customers are already using. Second, you’re at risk from the foundation models themselves, which can make parts of your product redundant with emerging capabilities natively encoded into the next generation of models. 

The models, however, are unlikely to achieve the end-to-end automation of many complex processes that are either already outsourced to service firms, or done internally by employees. This would be God-like AGI! The key to going after these workflows (and the spend associated with them) is combining technology with human expertise. This approach doesn't just create a lot of value for customers – it allows companies to capture a larger share of that value and become more resilient in the face of future AI advancements. As the customer, you are not going to care how good the models are at doing something that’s already fully solved for you!

The best tech-services companies use technology to drive operating leverage in two ways:

1. Create a unique value proposition with faster, cheaper, and better service quality than legacy providers.
2. Improve unit economics by removing big chunks of labor from COGS; both altering the margin structure and making the business easier to scale relative to incumbent competitors.

Take freight audit and payment: billions of dollars are lost every year in incorrect freight audits, and tens of billions are spent on outsourced service firms (of mixed quality). A point solution that automates invoice checking is useful, but is not as valuable or sticky as a company like Loop that handles everything from billing decisions to payment processing. Their full-stack approach allowed them to redesign the entire process and create efficiencies throughout, not just in isolated pockets. From their customers’ perspective, this is transformative. Instead of juggling multiple vendors or struggling to integrate a patchwork of software tools, they can outsource the entire function to a single, AI-enabled provider. In turn, Loop can optimize across the entire workflow, finding efficiencies and cost savings invisible to both off-the-shelf software and legacy service providers. There is, of course, a tradeoff: Loop has to manage the complexity of combining software and AI capabilities with an operations team that picks up the tasks that can’t yet be fully automated.

This isn’t the first time people have tried to build vertically integrated “tech-services” companies. Many great companies were built through tight integration of in-house software and best-in-class operating practices. But there are also many recent examples of VC-backed tech-services startups that raised hundreds of millions of dollars and didn’t work out. These companies often spent huge amounts on R&D without getting any real P&L benefit from that spend. At its peak, WeWork had almost 500 software engineers. It turns out having lots of engineers doesn’t make you a tech company!

Having worked with our top-performing tech-service companies over the last few years, four core principles stand out as major determinants of success:

1. Map the whole business ontology to prioritize R&D focus
2. Obsess over metrics
3. Combine organic and inorganic growth
4. Build a team fit for the task
   

1. Map the whole business ontology to prioritize R&D focus

At Palantir, we developed the concept of a business ‘ontology’, which defines the data schema and workflows that underpin every operation in a company. Understanding the end user’s business ontology is essential for building great SaaS products, but it’s even more important when building great tech-enabled service businesses –– where the ultimate value is created by the symbiosis of software and operations. An ontology is made up of data, logic, and actions. It acts as a business process map (BPM) that defines any given workflow. A simplified version of an ontology can look something like this for an airline (h/t PLTR):

An ontology is the syntax and grammar of a company. It consists of different objects (plane, flight, airline, airport, or delay). It maps data about those objects –– their states, relations, and operations –– to concepts, like “owned by”, or “operated by”. We can then translate those states and concepts to the workflows, gaining insight into the business which lets us act to automate or augment our processes.

In tech-enabled services, the natural starting place is to map the end-to-end ontology as it exists in the status quo: How does the work get done today? What objects, data, and actions are involved at each step? What set of states can each object be in? What set of relations can they have to one another? This kind of holistic mapping gives you a granular understanding of the jobs to be done, their atomic units, and how they are currently split between people and software systems. Building an ontology is also the key enabler for data integration – which in turn enables automation. Most people would assume that you have to integrate data first, then derive the ontology, but the reverse is true. In any massive organization, data is scattered across many formats and places. An ontology creates the structure to address the full scope of legacy and future data challenges. Palantir became the best in the world at data integration, and a leader in enterprise AI, because they made ontology a first principle early (and spent 20 years deploying top talent towards creating state of the art solutions for their customers).

This is the exact playbook Reserv followed as they were starting out – by deeply understanding the claims adjuster workflow, they were able to make the right decisions on what to build in-house vs licensing, which ultimately let them scale extremely quickly when they won a big customer contract. A detailed business ontology lets you prioritize R&D investments in margin improvement and customer-facing features. Combining the two is what creates enduring value.

Ontologies, in concept and practical application, are well understood by most great SaaS businesses, and only become more important in the world of tech-enabled services. When building a SaaS product you are mapping a customer’s workflow to a piece of software. When you’re building tech-enabled service companies, you have to create an ontology that captures the three-way relationship between your customers, your employees, and your software system. This is a big reason building in tech-enabled services is so challenging – you have to combine cultures of technical and operational excellence to create a positive feedback loop between the software and the labor.

Many smart young entrepreneurs think doing this type of work upfront and distilling it into something as simple as a deck is a bad use of their time. In reality, mapping the ontology and building a set of materials around the work serves multiple purposes: nailing the vision, staking out specific goals, and aligning the team. It also enables the people around the business, whether investors or advisors, to provide much more substantive feedback based on their prior work in the field. This type of alignment is more complicated than ever in tech-services companies, since there are many tradeoffs to think about on day one and the cost of choosing the wrong direction can be severe. 

2. Obsess over metrics 

Once you have an ontology, it becomes easier to see what to measure to read the vital signs of your business. SaaS metrics have become standardized over the last decade, and today, between knowing the right KPIs and the surplus of products designed to help you track and analyze them, it’s pretty easy to have high-fidelity instrumentation for your operations. And SaaS companies’ high gross margins give them a bit of breathing room to be less disciplined about their operational metrics, without affecting their P&L (as long as sales is working!).

Service businesses are the exact opposite. It is especially important to identify and track core metrics, because value creation in services is often more complex and intangible than in product-centric companies. It’s also a lot less obvious which metrics really matter –– a SaaS company selling to debt collectors has to measure roughly the same things as one that sells to architecture firms, but there are few KPIs common to both debt collectors and architects. Not doing this work, or (worse) choosing the wrong metrics can quickly lead to painful P&L problems. 

A classic metric for a wealth management firm might be assets under management. But an ontology reveals how client satisfaction, portfolio performance, and advisor efficiency are connected over the long term. This lets you create stronger metrics and direct your actions towards improving them. When done right, choosing and tracking the right metrics gives you a far more sophisticated grasp on your business, which lets you consistently make better decisions. 

Ontology-driven metrics analysis can reveal unexpected leverage points as well. In customer support, you might find that accurate problem categorization matters more than initial response speed for overall resolution time and satisfaction. When Reserv automates part of the claims process, it can trace the impact not just on processing speed, but on downstream factors such as customer satisfaction and cash flow. This panoramic view of the metrics lets them identify and head off issues before they happen. 

The premise behind all tech-enabled service businesses is that getting people and software working in tandem will improve margins and service delivery. To achieve this, metrics can't just arrive quarterly in a backwards-facing report that executives glance at occasionally. They need to be the focus of the entire business. The obsession with metrics is universal among 8VC’s best-performing service companies. They don't just measure – they live and breathe their metrics. They use them to drive decision-making, prioritize investments, and align their teams. In these companies, you'll find that everyone, regardless of their role, can tell you their key metrics and how their work contributes to moving them in the right direction.

3. Combine organic and inorganic growth

During the 2010s many venture investors rightfully viewed inorganic growth (ie. acquiring customers via buying another business) in the operating plan as a bit of a red flag. More often than not, M&A was used as a bandaid over product gaps or poor sales execution instead of addressing the root cause – lack of product-market fit! Today’s paradigm for tech services companies is entirely different, and entrepreneurs and investors alike need to update their frameworks accordingly.

Buying an incumbent can help solve the “cold start” problem for companies operating in markets with high regulatory barriers or switching costs, and a relative lack of differentiation between service vendors. Reducing hiring and sales pressures can allow startups to focus on validating tech-driven margin improvements in the early phase of company building. M&A is especially effective, and sometimes necessary, in industries with sticky contracts or risk-averse customers. 

If done well, inorganic growth can be a powerful accelerant for companies that have proven stronger margins in industries with a long tail of service vendors. A legacy service business with 15% margins might trade at around 6-8X cash flow, or about 1X revenue; and even less if it’s losing business to your upstart. An AI service company that’s been built well over a few years might already have 60% margins and be growing faster: it will trade at a sometimes significantly higher multiple. Integrating acquired legacy companies into the platform fundamentally increases the enterprise value of the existing revenue pool by improving the underlying economics, as well as, in some cases, re-accelerating the growth. Buying competitors for 1X their revenue, the AI company can invest $100 million in M&A to create ~$60M in cash flow, or $600M in equity value at a 10X multiple. 

This is not your father’s inorganic growth strategy. This approach to scaling tech-enabled services may end up being integral to the next phase of American industry, a vector by which best-in-class technology gets integrated into the economy as a whole (arriving just as many owners of these service businesses start thinking about retiring). From the outside, someone might worry that VCs are suddenly trying to get into PE, and foretell disastrous results. That’s not the point here: these are technology companies. It takes a venture and startup mindset, and a deep understanding of what is technically possible, to create and invest in these companies in the first place. But when technology can affect such a meaningful productivity improvement in a service workflow, inorganic growth becomes an obvious way to quickly create tens of billions of dollars of value.

Obviously M&A is not the appropriate strategy for all such companies, especially if customers are cheap to acquire and integrating acquisitions particularly cumbersome. Having an ontology will help you deliberate these trade offs and articulate your specific goals as you wrestle with questions like the right size for your first acquisition, or whether you should pursue one in a service vertical you’re already in versus an adjacent one. While there’s still justifiably some hangover from poor tech M&A examples over the past decade, this new paradigm unlocks a new strategy for companies to predictably deploy large amounts of capital with great ROI.

4. Build a team fit for the task

The most common characteristics across the winners of the SaaS wave is that they have strong technical cultures and empower engineering talent to iterate quickly. As software proliferated into more niche market areas, less known to Silicon Valley native talent, companies learned that domain expertise and speaking the customer’s language are also critical. This required building out a strong set of industry advisors or even at times hiring industry veterans with little tech experience. 

The tech-enabled services wave will sharpen this paradigm. Creating successful businesses will require hiring top technical and operational talent: people who combine IQ with EQ. It will also require combining fast-paced innovation culture with the client-focused, process-oriented approach of top services businesses. It’s not enough to have both types of people on the team; they must collaborate with, learn from, and actually like each other, since many engineering priorities are derived from ops teams that handle customers day to day. Technical talent, beyond having skill in AI and software development, will need to be curious and respectful towards the nuances of service operations. Operational experts need to be open to new technologies and willing to rethink established processes –– up to and including replacing themselves! 

Palantir’s “Forward Deployed Engineers” became a central part of our own internal ontology. Embedding directly with customers to configure the platform to their unique needs was how we brought this domain-specific mentality into our culture. At the time, deploying engineers to work alongside customers was totally alien to top Silicon Valley tech companies, who preferred to leave client interactions to sales and customer success teams. It was (and is) very hard to hire people who combine the technical, operational, and communication skills to work closely with non-technical customers. But this hiring mentality is essential in tech-enabled services, and it’s perhaps not surprising that many founders and early employees of our tech-services portfolio companies are Palantir alumni. 

Building the right team is about creating a culture that values both technological innovation and service excellence. If you’re pursuing an inorganic growth strategy, you also need DNA from the PE world who has the wisdom and experience to manage acquisitions and align incentives. It's this combination that allows tech-enabled service businesses to not only replicate existing services more efficiently, but transform entire industries.

If you’re working on an AI-services company or have ideas around other best practices shoot us a note to jack@8vc.com

–––

It’s impossible to predict where the ladder of rapid AI advancement leads. Implicitly, we’re betting that AGI is not simply going to “figure everything out.” If it does, we’ll all get rich (or go extinct). But even if the evolution of AI plateaued today, there would still be enormous value to unlock! The financial and societal implications of bringing radical efficiency to $2T in service wages are inspiring. Beyond additional GDP growth, more efficient labor markets, and doubled or tripled productivity in some sectors, workers will be free to practice their true skills, and learn new ones, as automation eliminates more rote tasks and processes.

We’re in the early innings of the AI Services Wave, and its implications and possibilities are nowhere near finished emerging. Early in the SaaS wave, our Smart Enterprise thesis introduced some frameworks for platform-driven industry transformation, which have since become commonplace. As the tech services wave gathers force, the principles of ontology mapping and the other strategies we’ve discussed represent a similar conceptual scaffolding, which we are using to explore opportunities to invest and to found companies with 8VC Build. Making any of this work in practice will require exceptional leaders, teams, and persistence, as it always has. With Palantir as an inspiring example, achieving many worthwhile missions––and briefly touching $100 Billion market cap this month––we are confident that we’ll see dozens more generational companies built in this productivity wave. We are excited to partner with them and learn and win together.