Pilot & MVP development customer
One practice, chosen because it is the hard case.
AiRadics is developed inside a working general practice rather than against a specification. Every module ships against real channel data, a real payer mix and a real patient book before it is offered to anyone else. That practice is Mi Smile Family Dental in North Houston, our pilot and MVP development customer; Dr. Soujanya Maddipati owns it and owns Quadfence, so it is a related-party engagement and we state that plainly.
A pilot is only worth something if what is learned there transfers. Most single-clinic pilots do not, because most clinics are narrow — one payer, one demographic, one service line. This one is deliberately the opposite.
4payer classes live
Medicaid, CHIP, Medicare Advantage and commercial PPO, in one book. Each behaves differently on acquisition cost, recall and lifetime value, so the model has to separate them rather than average them.
Mixedcatchment demographics
Owner-occupier and renter households, a wide income band and multiple language groups inside one catchment — the segment variance a single-demographic clinic never produces.
Fullservice line
Preventive through restorative, plus third-party patient financing. The full set of procedure categories a general dentist bills, which is what the ontology has to cover.
Since 2022operating history
Three years of channel spend, collections and procedure mix already in the book, which is the history a model needs before it can say anything about attribution.
The point is coverage. A practice serving one payer and one demographic teaches the ontology one path. This practice exercises nearly all of them at once, which is why the axioms it produces — what worked for which segment under which payer, and what did not — generalise to the next clinic instead of having to be relearned.
What success looks like here
Four things have to be true before we sell to a second practice.
A related-party first customer is only worth something if it is held to the same standard a paying stranger would hold us to. These are the tests, and we would rather publish them before we have passed them than describe them afterwards.
The engine moves the numbers
New patients per month, cost per acquired patient split by payer rather than blended, and lifetime value against acquisition cost. Measured against the practice’s own three-year history, not against a projection.
The governance gate holds at volume
Every generated asset clears FTC, ADA, state dental board and platform health-content rules before publication, with a pass rate we can quote and a provenance record for each decision. A gate that only works on a handful of assets is not a gate.
The ontology produces reusable axioms
Not conclusions about this clinic, but structured results — which channel produced which patient type at what cost, under which payer, in which catchment condition, and what failed. Enough of them, across enough segments, to be worth something to a practice we have never seen.
A second practice stands up in a day
Its own cloud project, channels connected, catchment modelled, site generated — provisioned from the blueprint rather than built. This is the number that decides whether an agent-led sale works at all, and it is the one the round exists to prove.
None of these are passed yet, and we are not going to claim otherwise while the work is in front of us. They are the criteria, published in advance, so that the answer when we are asked is a number rather than a story.
Why one clinic is enough to start
What the pilot produces is not a case study. It is the ontology.
The engine does not carry conclusions from clinic to clinic. It carries structure: the categories a dental market decomposes into, and an axiom set recording what moved which segment, under which payer, in which catchment condition — and what did not.
What transfers
Built once at the pilot, reused at every practice after it.
- The dental ontology itself. Payer classes, procedure categories, catchment variables, channel taxonomy, patient journey states. The vocabulary the engine reasons in.
- Segment axioms. Which channel produced which patient type at what cost, split by payer rather than blended. A Medicaid recall patient and a PPO restorative patient are not the same acquisition and should never share an average.
- Negative results. What was tried and did not work, held with the same weight as what did. Most marketing knowledge is lost because nobody records the failures.
- The governance rule set. FTC, ADA and state dental board advertising rules compiled once into a gate that runs before generation.
- The pipelines and the console. Channel ingestion, the client warehouse, the hub aggregation, the generation and publishing path, deployment and monitoring.
What does not, and we say so
A second clinic is not a copy of the first.
- The catchment. Every practice sits in its own market and needs its own model. What transfers is the method for building it, not the answer.
- The payer mix. Ours is unusually broad, which is the point — a narrower practice uses a subset. A practice with a mix we have not seen adds to the ontology rather than reading from it.
- Absolute numbers. A cost per patient at one clinic is not a forecast for another. The transferable object is the relationship, not the figure.
This is also the honest answer to why we are not running twenty pilots. Twenty shallow clinics would produce twenty thin datasets. One clinic exercising every payer class, a mixed catchment and a full service line produces an ontology the twenty-first can actually use.
What the pilot is meant to produce, and what the round is meant to prove.