Construction

The six-figure integration bill your product competes against

Legacy operators dislike the software they own and still cannot switch. Annija Eizenarma on why sunk integration cost is a product spec, not an objection.

Your competitor in a legacy industry is not the incumbent software. It is the integration the operator already paid for. Annija Eizenarma — a founder, investor and advisor with seven years selling into legacy markets — puts observed integration costs at "from 50 to 150, sometimes even up to 500,000 or a million, I have heard, for integrating a single system," on licences that run five and ten years.

She does not state a unit for the first two figures; Artificial Reality reads them as thousands of dollars from context. This is a practitioner's observation, not a published benchmark. It is also the number that decides whether your deal closes.

Isn't the demand already there?

Yes, and that is the confusing part. Dissatisfaction with the installed stack is not the constraint. Eizenarma:

"Often it's the case when there's a big fragmentation and they're almost always disappointed. They never like it. They use all the systems because they have it sold on them. And usually they have these licenses and their managers and they have contracts for five, ten years."

Read that as a market state rather than a complaint. Demand for the replacement already exists. What does not exist is the operator's ability to act on it, which is a different problem and has a different fix.

Why doesn't dissatisfaction convert?

Because switching means defending a decision internally, not evaluating a feature set. Eizenarma:

"You have to fight against big onboarding and integration costs that legacy systems have. Like when they have all these Salesforces or whatever, they have paid a lot of money to install it in the first place. ... And it's really hard for them to switch over because they have already paid, they have invested."

An AI vendor competing against a six-figure sunk integration and a multi-year contract is not competing on product quality. Products that add a layer without displacing the contract avoid that fight entirely. Products that ask for rip-and-replace inherit it, and then lose it to an internal conversation they are not in the room for.

What this changes in the roadmap

Treat the sunk cost as a specification rather than an objection to be handled. The version that ships against this market:

Works while the incumbent contract is still running. Deployability, not displacement, is the entry requirement.
Produces its own ROI case in the first month. Eizenarma's approach is arithmetic: "you pay this for this system, this for that system — and making the math for them and showing the difference in terms of maintaining the existing systems."
Commits to one workflow, not a platform migration. "Proving that you can deliver one thing really well is really, really crucial."
Keeps onboarding out of the commitment. Her framing: "making the commitment very simple and easy for them so it doesn't require some setup, some crazy onboarding."
Is still installed when the incumbent licence comes up for renewal. That is when the switching decision becomes available, and it is years out.

The pricing advantage nobody is using

Modern tooling has a structural cost position against installed enterprise systems, and Eizenarma treats it as the wedge rather than a talking point: "usually [the] solutions nowadays don't require any licenses, right? So you can really compete with many things that the systems have."

That only converts if the buyer can see both numbers on one page. The comparison an operator can act on is current per-system spend and current days-sales-outstanding against projected — produced by you, in their numbers, before they have to ask anyone above them for approval.

What this reframes about connectors

Deep native connectors into the systems operators are stuck with are not integration debt. They are the moat. Every year a contractor cannot leave the system they bought is a year the layer above it becomes more load-bearing, and the incumbent's own lock-in starts working for you instead of against you.

The gap Artificial Reality sees in this category: nearly every product enters through finance. None appear to enter through the field, where the data that makes the automation accurate is actually created. A tool that captured job-site reality accurately enough to make billing automatic — rather than automating the billing of whatever imperfect data reached the office — would be solving one layer deeper than the category currently does.

So the question for a founder selling into construction is not whether your product is better than what the operator owns. They already agree that it is. It is whether your product can be running before their contract ends.