Why we stopped recommending shared-tenant AI vendors to dental groups
After three engagements, the same vendor lock-in pattern shows up. Here’s the version of the conversation we now have on day one.
We have run three deployments now where the practice owner came to us because something they had bought from a vendor stopped working the way they expected. Every time, the root cause was the same: a shared-tenant architecture where the vendor controlled the data, the model, and the terms.
The pattern we kept seeing
The shape of the conversation is identical. Practice owner signs up for a chatbot or an intake tool. Six months in, the vendor changes something. New pricing, new terms, new owner, new model. The practice has nowhere to go without losing the data they accumulated.
A shared-tenant vendor is not a partner. It is a landlord with a short lease.
What actually changes when you isolate
Three things change the day a deployment becomes single-tenant. First, custody. Your data is on infrastructure you can audit. Second, exit. If the consultancy folds, gets acquired, or just becomes annoying, the deployment keeps running. Third, terms. The vendor cannot unilaterally rewrite what you signed up for.
The cost most operators miss
The honest part: isolated infrastructure costs more up front. A dedicated VPS is more expensive than a per-seat SaaS subscription for the first year. The math flips around month fourteen, and dramatically by month twenty-four.
But the real cost is not on the invoice. It is the cost of replatforming when your vendor sunsets the product, raises prices three times, or sells to private equity. That cost is hard to model because it has not happened to you yet. It happens to someone every quarter.
What we recommend now
For dental groups, CRE firms, and any operator handling regulated or sensitive data: skip the shared-tenant vendors entirely. Pay for a real deployment once. Own the keys, the code, and the runbook. The operating cost stabilizes, the audit trail is real, and you stop being a tenant.
That is what Salvus does. We are not the only people who can do it. We are happy to lose the engagement to a competent local engineer who builds the same thing for less, as long as the operator ends up with infrastructure they own.
Filed by
John McIntosh
Founder, Salvus AI
Talk to us
Want this kind of work shipped for you?
Salvus AI deploys per-client isolated MCP servers and the automations that run on them. Client success with real numbers, no shared tenancy.