Every enterprise software category is drifting toward the same endpoint: the company owns the model. Fine-tuned on proprietary data, trained on institutional history a competitor will never see, running on infrastructure the company controls end to end. Five years ago that was only viable at the scale of a handful of tech giants. Open-weight models, cheaper fine-tuning, and better tooling have pushed that same option down to mid-sized operators, and the line keeps moving.
This is probably the right call, long-term. A model trained on your own data doesn't leak your data to a vendor. Owning the weights means no one else can raise your price, deprecate your access, or shut off the feature you built your workflow around. It encodes what your business has actually learned, not a generic approximation of it, which is exactly the kind of advantage a competitor can't buy off a shelf. The economics tighten as you scale, instead of loosening. Every reason a company insources a core system is present here, stacked in the same direction.
Extend that logic far enough and you get a clean prediction: the enterprise stack of the future is mostly in-house. Whatever intelligence layer touches your product, your customers, your operations, you own it. The support model that knows your actual return policy instead of a generic one. The pricing model tuned on your real margin structure instead of an industry average. The ops model that has seen every job your company has ever run. None of that is exotic anymore. It's the direction almost every serious operator is already moving, at whatever speed their budget allows.
None of it works, though, if trust isn't the thing the whole stack gets built around. An in-house model that quietly erodes a customer's trust is a liability no matter how good the fine-tuning is. Every system worth owning gets built with trust as the starting assumption, not something bolted on at the end. That part belongs inside the walls.
The one exception
There's a structural break in that pattern, and it's not a matter of preference or budget. It's definitional.
Escrow only works because the party holding the funds has no stake in the outcome. That neutrality isn't a feature of escrow, it is escrow. The whole product is "neither side controls this money until the agreed condition is met." Remove the neutral third party and you haven't built a leaner version of escrow. You've built something else: one side holding money and asking the other to trust them anyway.
You can fine-tune your own model. You cannot be your own referee. A company that holds its own escrow isn't a company with efficient escrow, it's a company with no escrow, wearing the word as a label. The moment the fund-holder and the fund-beneficiary are the same entity, the thing that made escrow worth using in the first place is already gone.
You can't fix this by trying harder at neutrality internally, either. An internal "independent" review team is still on the same payroll. A separate legal entity owned by the same parent company is still owned by the same parent company. Every internal fix to the neutrality problem is a more elaborate way of describing the same conflict of interest, not a resolution of it. Neutrality that can be overridden by the org chart was never neutrality, it was a policy, and policies change when it's convenient for the people who wrote them.
This is why escrow is close to alone among software categories in getting more valuable as everything around it moves in-house, not less. Every other layer trends toward "why am I paying a vendor for something I could build myself." Escrow trends toward "the fact that I can't build this myself is the entire point." As trust in outside vendors erodes for everything else, the appeal of a genuinely neutral one goes up, not down, because it's the rare vendor relationship that was never supposed to be about convenience.
The cornerstone you still can't build yourself
Put those two ideas together and the shape of the future stack gets specific. Trust is the cornerstone: every in-house system, every model, every automated decision a company makes has to be built with trust as the foundation, not an afterthought. That part is entirely buildable in-house, and probably should be. What can't join it inside the walls is the live mechanism that proves that trust while money is actually changing hands. Escrow isn't the opposite of trust. It's what trust looks like at the exact moment two people who don't fully trust each other yet need to act as if they do.
Construction payments are the cleanest version of this problem there is. Large sums move between people who mostly haven't worked together before and may never again. The homeowner doesn't know if the contractor will finish. The contractor doesn't know if the homeowner will pay once the work is done. Both sides are right to worry, because both sides have been burned by someone in the other's position before. A handshake doesn't scale to that kind of stakes, and a payment app that just moves money faster doesn't fix the actual problem, it just makes the wrong outcome arrive quicker.
Put the money entirely in the homeowner's hands and you're exposed to nonpayment after finishing the work, a documented, common failure mode. Put it entirely in your hands upfront and the homeowner is exposed to a deposit that walks away, also documented, also common. Every fix that keeps the money on one side of the transaction just moves the exposure from one party to the other. It doesn't remove it.
What fixes it is a third party with no reason to favor either side, holding the funds until the work everyone agreed to is actually done. That's not a nice-to-have layered on top of a job. It's the cornerstone that makes a stranger-to-stranger transaction survivable at all, which is exactly why it's also the one piece that can never be built by either of the strangers in it, including you.
If you're still collecting payment the old way, a deposit up front and final payment chased after the walkthrough, that's exposure you're carrying that doesn't need to exist. Clearbuilt runs payments through a neutral, escrow-first rail that neither you nor your client controls, precisely because that's the one system trust requires you not to build yourself. Everything else in your business, build in-house. Start with this.
Trust is the cornerstone. Build everything else around it. Just don't try to be your own proof of it when someone else's money is on the line.