How Rely works / Learning the workflow
Rely observes the local workflow, defines owners and stops, and tests the approved path before live work.
The real local path is recorded, bounded, tested, and approved before live work.
Observed practice is not automatically approved practice.
Automation begins with observation. The real local path is recorded, bounded, tested, and approved before live work. The diagram shows Observe actual work, Map variants, Name owners and stops, Test synthetic cases, Approve the lane, Monitor and revise. Observed practice is not automatically approved practice.
Stage gates
Buyers should not have to guess when something becomes real. These are the stages a program moves through and what can honestly be said at each one. They describe an operating design, not a guaranteed outcome: what it produces in your program depends on volume, on data access, and on how fast your team can approve a change. If approvals take a quarter, the loop runs at the speed of approvals.
Pre
One workflow is scoped. Permitted actions, prohibited actions and the clinical boundary are agreed. Security and legal clearance is complete, access is provisioned, and routine and edge cases are tested. Nothing runs against a live population until these are signed.
Day 1
Where navigator support is included in scope, navigators onboard into your existing EHR and processes under named, provisioned access, the way a temporary nonclinical hire would. They work the nonclinical steps with patients and document coordination outcomes in the record. Anything a licensed user must author stays with your team. The program can begin while integration work is still being scoped, so discovery is not blocked behind an interface project. That depends on your organization granting named access under the agreed terms, and provisioning, least-privilege access and security review are defined for each program.
Day 45
The working target is that by day 45 the ideal path for this population is understood well enough for the agent to guide against it. That is not a claim that AI handles 95% of the work by then. Guiding on the ideal path can still require substantial human involvement, and in the source model that share is around 40%. We state the human share on purpose, because a number that does not survive month two is worse than no number. Reporting before this point describes what is being learned, not steady-state performance.
Q1
Service levels have real history, the exception taxonomy is written down, and reporting reconciles against your source system. This is the first point at which an outcome conversation is fair to either side.
6 mo
There is enough operating evidence to decide what becomes a system standard, what stays local to one department, and which adjacent transition is worth adding. Expansion is a decision against pre-agreed evidence, not an automatic next phase.
We will identify the real routes, owners, exceptions, and launch gates.
