Enterprises have always bought outcomes. Long before anyone used the phrase AI-native services or services-as-software, a business that needed GTM support hired a marketing agency. A carrier that didn’t want to staff an underwriting desk pushed submissions to an MGA. An employer that didn’t want to build a benefits function hired a broker or TPA. The work got done by somebody else, priced by the hour or the head, and the buyer never looked inside the box.
That arrangement held for decades because the economics held. The vendor's cost was labor and its price was labor plus a margin, so nobody could get dramatically cheaper or better without hiring more people.
That's the part that broke. Once a meaningful share of the work could be done by AI, the hundredth account stopped costing what the first one did. A company that takes on nearly the entire job with technology can price against a labor budget and still carry a margin that looks like software.
We've been underwriting this model before it had a name
This is not a new position for us. Several of our best companies have always been part software and part service, and the software is what made the service work at a price the market could absorb.
Midi Health delivers specialized care for perimenopause and menopause, delivering clinician-led support rather than solely selling a care platform. CertifyOS rebuilt credentialing (long outsourced to CVOs) as verified data through an API, and is now the provider data backbone for eight of the ten publicly listed US health plans. Stepful runs a school-as-a-service for entry-level healthcare roles, covering enrollment through placement so employers can skip contract staffing.
What has changed with AI is how much more of the job software can take on, and therefore how large the prize is. The market is proving this out in realtime, and below is how we think about the ones that will get big.
1. Capture outsourced spend that already exists
Services budgets are structurally larger than software budgets in most of the functions we care about. Therefore, the strongest wedge we see is a line item the buyer already pays for and probably resents. You are replacing outsourced vendors, BPOs, agencies, staffing firms, consultancies, offshore teams.
This changes the sale in ways founders consistently underrate. The budget is approved, so nobody has to invent one. The incumbent has a name, a contract, and a price per unit, so the comparison is simple math rather than a vision exercise. And you are underwritten against a service the buyer is already measuring, which is a much shorter path than convincing a team to adopt a tool and then waiting to see whether they actually use it. Nobody has to change their behavior for you to get paid, and you won’t have to jump through AI-vendor hoops with their security and IT teams.
2. Win the workflow, then own the record
This is the progression we care most about. You deliver the work, which puts you inside the workflow, eventually making you a potential system of record.
Delivering the service is what gets you access. Sitting in the workflow is what gets you the data, because you see every submission, every exception, every appeal, and every correction as it happens. Becoming the record is what makes you permanent, because at that point the customer is not renewing a vendor contract, they are deciding whether to rip out their source of truth.
Plenty of services companies stop at the first step. They do good work, they stay replaceable, and they get repriced at every renewal. The jump from doing the work to owning the record shifts the dynamic, and it is usually a deliberate product decision made early rather than something that happens on its own.
3. Build data that compounds
We want data you can only get by doing the work, especially if it’s hard to infer. Anything you can buy, scrape, or license is available to everyone competing with you, including the incumbent you are replacing.
The test we apply is simple. What is better about job number ten thousand that was not true about job number ten? The good answers are almost always about accumulated failure. The denial that got overturned and why. The credential that did not verify against the primary source. The submission that got mispriced and the exception that explained it. That information exists nowhere except inside the workflow, which is why the incumbent BPO cannot replicate it even after it buys the same models you use.
This is also the honest test of whether the margin story is real. If the ten thousandth job costs about what the tenth cost and comes out about the same, the data is not compounding, and what you have is a cheaper vendor rather than a better business.
4. Find labor shortages
Some of the best markets are the ones where the people simply are not there. Nursing, medical coders, radiologists, claims adjusters, actuaries, pharmacists, care coordinators, skilled trades and field services.
You are not asking a manager to displace their team, which is the objection that quietly kills most enterprise AI deals. You are picking up work that is not getting done at all. The buyer already has the backlog and vacancy rates they have been apologizing for, which means your ROI is sitting in their existing reports. These markets also tend to be the ones where the buyer will tolerate a hybrid model for longer, because the alternative is nobody doing it.
We still believe humans-in-the-loop early on is expected. What matters is whether you can tell us which parts of the work are on a path to automation, which parts never will be, and why.
Where we’re excited
While we don’t believe this shift will be limited to certain sectors, a few of the areas we're excited about are:
- Horizontal work still owned by agencies
- Financial services, including wealth and tax
- Education and learning
- Security operations, audits, and third-party risk
- Pharma services, including trial operations and market access
- Benefits eligibility, enrollment, and plan administration
Apply
If you're building an AI-native services company or know an amazing founder who is, we would love to talk. Applications are also open for our Accelerator program. You can apply here.
Stay connected
Get our monthly newsletter to explore how we're building a world that works better.
