Phase one — scoping

A conversation with the people who would actually run the work. What is the population, what is the problem you are trying to move, what have you already tried, and what data do you already hold. The output is a written statement of the problem and an honest assessment of whether this program helps with it. Sometimes it does not, and saying so at this stage costs an agency nothing.

Phase two — design

Which components, for which population, with what data flows. This is where data governance gets settled, because your counsel needs a specific description of what moves where and not a general assurance. It is also where the outcome measures are agreed — before the program starts, and including the ones that could make it look bad.

Phase three — pilot

A defined population, a defined geography, a defined duration and a pre-registered set of measures. See pilot design for what makes a pilot capable of producing a real answer rather than a testimonial.

Phase four — evaluation and decision

The measures agreed in phase two, reported whatever they say, with the analysis available to the agency’s own epidemiologists. A pilot that does not work should be discontinued, and a program that cannot survive that possibility is not worth running.

What we will not agree to

• Deployments in which participation in the program is a condition of receiving treatment.
• Use of clinical components as an enforcement or case-building instrument.
• Individual-level reporting of behavioral health screening, social determinants assessment or monitoring data to an agency.
• Building a per-person suspicion score. See AI integrated rule discovery.
• Anything that would require IntellaRx to make a legal determination about your agency’s obligations.

The full boundary is on what we do and do not do.