AI-enabled services is the next great venture category

By Michael Stothard, Sam Endacott and Lorcan Delaney

Over the past decade, pretty much no VCs have been looking to invest in service-heavy businesses. Why would they? Compared to software businesses, services performed by humans tend to be low margin, hard to scale exponentially, lumpy in revenue and generally lacking in any of the qualities that point a path towards a speedy multi-billion dollar outcome.

At firstminute, we believe that this model for thinking about venture and services business should be torn up. We are seeing some of the best founders in our portfolio, and some of the best founders we are meeting day to day, using AI to automate big parts of the services industry with a fresh and AI-first approach, and we are actively investing in this theme.

One of the companies we really admire is ClaimSorted, which is building a brand new AI-first TPA — which stands for Third Party Administrator, companies that process outsourced insurance claims. TPAs are part of a wider universe of outsourced services companies — sometimes called BPOs, which stands for Business Process Outsourcing — which we think are ripe for disruption and where we are spending a lot of our time at the moment.

Why are we so excited here about founders building in these service heavy sectors? Four main reasons:

1. The markets are gargantuan. The big 4 consultancies alone brought in $20bn in revenue last year. Then in each of the hundreds of specialist outsourced services sub niches that exists across different industries (e.g. supply chain billing, inspection and certification, medical prior authorisation, insurance claims processing etc…) there are under-the-radar behemoths doing billions in revenue. Many of these fall under the broad category of BPOs (Business Process Outsourcing) and TPAs, which both do outsourced tasks for larger companies. Conduent, one of the largest BPOs in the US, alone does $4bn of revenue.

2. AI has just changed what is possible: many of these services businesses employ tens of thousands of people (Tata Consultancy Services in India employs nearly 600,000 people on $25bn of revenue while Conduent around 60,000 people) in work that has just become ripe for disruption thanks to generative AI. They are also often run on relatively thin margins so a 10-50% margin increase would be profound for them and their valuations.

3. Legacy providers are ripe for disruption (unlike in pure software): the companies that have dominated these industries for the past 50 years have business models that rely on human labour and hourly billing and have little tech DNA. While software incumbents are having a relatively easy time adapting to the GenAI wave (see companies like Intercom or Gong or Canva building brilliant GenAI products) services incumbents will struggle.

4. Companies like Palantir can show us the way: Palantir started as a services business and is now a $100bn company (on $300m in net income, and so the market is giving the company a lofty multiple) with a mix of tech and services. There is, we think, about to be an explosion in the number of Palantirs in the world thanks to AI.

Many of the companies we are looking at at the moment are starting with software to serve big outsourced services companies. We are not against the pure-software play in this market, and indeed have backed many. But a large number are considering providing the services themselves either now or in a few years. We think this is an interesting shift and — if you don't mind us philosophising for a moment (!) — speaks to a bigger point about what a tech startup really is and will be over the next decade. Because while today we naturally draw a distinction between tech and services companies, in truth buyers don’t really care — they just want a job done well and efficiently. It would, in fact, be easier for many to have a single vendor to hold responsible for a job rather than some "tech" and some "services".

This point is backed up by how well Palantir is performing right now (shares are up 30% year to date), with one employee telling firstminute that in part thanks to how they have approached AI — with their blend of software and services — they were "busier than ever". CEO Alex Karp recently said that the company has received as many inbound phone calls in “like, a month” that it typically gets in a year. When it comes to AI tools, he said, “This is an infinite market.”

Below we’d like to lay out some of the work we’ve been doing on this tech enables services theme — based on dozens of investments as well a 100+ conversations — discussing: a) a broad framework for how we think about the opportunity in services and b) our work on TPAs and BPOs and some of the most exciting services industries and verticals that we think are ripe for disruption including in health, insurance and supply chains c) how we are battling with some question around PE roll ups, starting as a full stack play, go to market and the challenge of hyper-scale.

Our Framework

Service firms are hired by a company for two reasons. Either to:

a) Do a job that the company doesn’t have the bandwidth or expertise to do, such as audits or customer support or claims processing. 

b) Offer third-party advice in decision making, such as M&A banking services, litigation or tax planning.

Starting with b), there is no doubt that big businesses will be built around the second “third-party advice in decision making” service category. 

But there are reasons to like this less. One is that expert advice like M&A or tax planning is too complex for AI today and another is that advice is in large part an ass-covering exercise — blue chip brands matter and bots will never have the credibility of the likes of McKinsey to hand out advice. 

This means that really any startup in this category will in effect be building software tools to help the services companies make decisions. Examples of this include Harvey, the AI for legal services platform, which is there to help save lawyers time. We have seen hundreds of these across different bits of law and other areas such as due diligence and wealth management.

As we said, big companies will be built here — Harvey was valued at $715m at its last Series B round and backed by Sequoia and Open AI. But we think more exciting is startups in this first category of outsourced processes, particularly those that are tackling specific pieces of work that can be relatively easily automated and have objective and measurable outcomes.

This is exciting because we believe that here there is the low-hanging fruit where startups can find a wedge to sell into a services company — or just be an AI-first services company — and create immediate value. Down the line they can do more and more specific tasks well, until they are eventually a full service and largely automated solution. These companies can be sticky and build themselves a data moat.

Where are we excited to invest?

One way to think about this as a founder is to think about specific tasks in cross industry verticals that could benefit from the work of an AI+Services company. For example:

• In IT support, the service could be app maintenance, pen testing, customer software building, cloud migration or quality assurance. 

• In finance, it could be bookkeeping, billing, financial reporting or auditing. 

• In legal it could be IP/Patent services or incorporate services

• In strategy and management, it could be data analytics

• In operations, it could be customer support, HR services or sales & marketing services.

Another way to think about this though — and what gets us the most excited — is thinking about one of these verticals as it applies to a very specific industry. 

There are too many of these industries to mention really and this piece would be too long if we went into all the possible areas ripe for disruption here. But some include, ESG service & monitoring providers, IP and legal services, call centres and freight billing - as well as the four we are going to talk more about here. 

1. Healthcare - Transforming patient care delivery, non clinical workflows, revenue cycle management.

AI is being embedded into frontline healthcare operations like diagnostics and medical imaging at a dizzying pace. 

But we’re actually most excited about improving the non-clinical workflows, which is a huge prize: of the nearly $4 trillion spent on healthcare annually in the US, a full $1 trillion of that is on admin (according to McKinsey).

There is lots to do here, but one of the most obvious areas to automate is PA (prior authorization), which has long been a source of frustration for clinicians. 

LLMs certainly promise more efficient PA workflows, freeing up physicians who would otherwise be spending up to 14 hours a week doing the process manually, and the rationale for payers and providers is clear where AI-enablement serves to increase accuracy and standardise authorisation outcomes.

Some of the startups in this space are building tech-enabled TPAs - most notably Cohere Health, which has raised $96M.

Some are building the architecture for tech-first TPAs, like Co:Helm, an LLM-powered co-pilot for medical administration, handling everything prior to clinical decision-making; Co:Helm raised $20m from investors including Sequoia.

There are several others as well:

• Basys.AI - Payer/Provider facing LLM engine for prior-authorisation and utilisation management (go forward recommendations based on individual records).

• Develop Health - Provider-facing, AI-driven platform designed to streamline the prior authorisation process. 

• Humata Health - AI and automated end-to-end workflow for providers across all high volume service lines, with payer connectivity. 

And we have just backed an incredible company in this space (although they are still in stealth!). Long before AI solutions are permitted to handle clinical diagnostics, they will eliminate much of the administrative burden and we are actively looking for teams here.

2. Testing, Inspection, and Certification (TIC). Inspection & certification services

Pretty much everything you touch — from food to furniture to car doors — has been inspected, tested, and certified. 

And this has mostly been done by people, with a clipboard and a piece of paper, working typically on behalf of giant testing, inspection, certification (TIC) companies. These workflows represent a bountiful harvest for AI-enablement. 

Traditional manual inspection processes are painfully error-prone and labour-intensive, and those companies that provide TIC services are lumbering giants; slow to change in a rapidly shifting landscape of industry. 

TICs have enormous value lurking on their income statements – giants like SGS Group (Switzerland, $17bn in revenue), ALS Limited (Australia, $4.1bn in revenue), Bureau Veritas SA (France, $13.3bn in revenue), and Intertek Group (the U.K, $10b in revenue) have personnel costs accounting for 60-65% of revenue, and operating margins lingering in the single digits.

How much of the manual inspection workflow could be expedited by implementing intelligent document processing, co-pilots, and image and video analysis — turning inspectors into power users? 

With sufficient proprietary training data, and successful pilots, the report-generation side of things could soon be expedited, and we might eventually expect to get some level of automation – where staff can shift their focus from manual data-gathering to supervising automated processes.

The incentives for TIC incumbents to streamline operations will create a great opportunity for TIC-enabling-platforms, as long as the hurdles to implementation can be removed (staff will need to be willing to use the new tech, new tools will have to slot directly into proprietary systems, etc.). TIC staff wanting to protect the billable hour will be a key dynamic!

We are optimistic about more entrepreneurs venturing into this space, and think that the right team will build a massive business around this thesis. 

 3. Financial Services: Modernizing banking processes, lending, payments, and regulatory compliance as services.

One great area we have been doing work on is the manual service of debt collection, which requires lawyers to write letters and people to turn up at houses. One company we admire is Respaid, which uses AI to “respectfully” collect debts from $50 to $100k ish and have a 50% collection rate within 20 Days (vs. human at 3%). Will the whole human-heavy sector be disrupted? 

We have also been thinking about loan servicing, new general ledgers and accounting platforms. A successful outcome here across the BPO / TPA thesis will be allowing companies to focus on core competencies while outsourcing routine yet critical tasks to AI-enhanced platforms. This shift is evident in the way regulatory compliance and risk management are being redefined. 

• ComplyAdvantage - AML & anti-fraud monitoring. By automating the compliance process, ComplyAdvantage has helped outsourced service providers manage regulatory risk more efficiently and accurately.

• Zest.AI - Lending software for smart (and fair) loan approvals. Leverages machine learning to enhance credit decision processes, enabling lenders to analyze more data, improve accuracy in risk assessments, and increase fairness by reducing bias. Zest AI helps financial institutions boost approval rates and expand access to credit, while ensuring compliance and efficiency.

4. Freight and shipping - invoice processing

In the bustling world of logistics, goods flow one way and capital flows the other. Tariff compliance delays, customs clearances, and those inevitable snags at handover points often tie up goods. However, you would at least expect that even if goods flow slowly, cash would be able zip along instantaneously in the supply chain. How is it, then, that a truck driver in the US typically waits a staggering 50 days to get paid after a delivery? Cash flows encounter their own set of hold-ups due to regulatory and compliance frictions; shippers, carriers, and freight forwarders wade through a swamp of paperwork, grappling with mismatched document formats, and battling tedious reconciliation processes that belong in an era when goods went by horse and cart. 

So, logistics also suffers from a burdensome back-office. Tasks are typically handled through slow, error-prone manual processes, placing a significant strain on labor. This inefficiency is ripe for a technological overhaul. The introduction of better tooling – smarter data extraction and validation techniques, paired with AI models trained on domain-specific data – can drive significant improvements. Shipium and Loop are two interesting businesses in this space.

Who is building in this space already?

Loop - providing a suite of services designed to streamline financial transactions between shippers, carriers, and brokers; improving cost visibility, reducing operational expenses through automated audit and payment processes, and optimizing transportation budgets.

Expedock - a platform which specializes in automating and digitizing the paperwork involved in freight forwarding, such as airway bills and bills of lading, helping companies avoid tedious data entry tasks, and improving accuracy.

Shipium - a shipping enterprise platform, which focuses on delivery experience APIs, fulfillment network performance, customer delivery dashboards, and shipping insights – allowing smaller retailers to compete with larger players like Amazon by improving how they meet delivery promises.

5. Pharma

The most important outsourcing providers for drug manufacturers can be found in the groups of contract research organizations (CROs) – organizations doing research, development and/or manufacturing of drugs on a contract basis.

One of our portfolio companies Lindus Health (which raised a Series A last year) is taking on the CROs building an tech-enabled challenger. It describes itself as an “anti-CRO” running end-to-end clinical trials for life science pioneers.

We think that the tech-enabled administrator theme slots very neatly into the CRO space. CROs, like logistics, prior authorisation, and testing / inspection / certification, face a mountain of administrative responsibilities. We’re optimistic about services and tooling that helps CROs accurately and responsibly streamline their manual operations. Two examples are Ryght, which partnered with Avance in 2024, an American-Australian CRO, and Mendel.AI. 

The big questions

There are a bunch of questions that we are asking ourselves all the time as we invest in this broad tech-enabled services category. We don’t have answers to them but they are important for founders - and investors - to keep in mind. 

Will these companies end up as part of some PE roll up? There are a lot of people raising money right now to do a roll up of tech enabled services businesses. In theory at least, it makes sense to buy 20 niche services businesses doing the same thing, make them 10% more efficient with AI, and then flip the whole thing for a huge premium. If that is where this sector heads, do startups end up being bought out before they can achieve venture scale?

What if our portfolio founders start buying services businesses — and is that ok? A few of our more experienced founders have started buying small services businesses and then deploying their technology in the firm. If they can buy a business for 3.5x EBITDA and then 2x margins, clearly that makes financial sense. But it’s unorthodox from a traditionally venture case point of view. 

Is being full service even the end goal? Many of the companies we talk to see the opportunity in tech enabled services companies, but just want to sell software to do the enabling. Others are trying to build full-stack AI-first services companies from the ground up. The problem with the former approach is getting the tech in these 50-year-old companies. The problem with the latter is that building a services business from scratch risks not scaling well. 

How to get buy in from the non tech-savy? The non-digital nature of these service-heavy industries is an opportunity, but it also presents a challenge. How do you get buy in from a user base suspicious of technology? How do you find internal champions in your target customer companies? How also do you get users to shift the way they perform tasks and adopt new interfaces? Not sure anyone has solved this yet - and it might hold back growth.

How do you find a GTM wedge? Many of the companies we talk to have advisers in the industry that are helping them with intros — that’s great. Many are also solving very niche problems, which is also great, and have a solution which sits on top of existing workflows rather than forcing new ones (ideal if possible). Many also have a “share of savings” pricing model, aiming to take 30% of cost savings. The idea of this is to get the customer comfortable with buying in the knowledge that they won’t pay unless it works — and also acknowledged that many of these companies don’t have specific tech budgets to spend from. Also interesting is the idea of “piggyback software” — areas where you can sell software alongside services — and the importance of credibility and brand in these areas where trust is so important. 

–

If you are building a company with an AI wedge automating a niche bit of the services industry, and are interested in pre seed or seed financing, we’d love to hear from you and do reach out at michael@firstminute.capital, sam@firstminute.capital or lorcan@firstminute.capital.

Michael

P.S Thanks to all our founders and and everyone in the space that helped us think about our thesis here, and for Oliver Lavelle for advising as well.