Connect the patient journey, and assure the decisions that follow it
Healthcare does not suffer from a lack of data. It suffers from fragmented state and fragmented decisions: the patient's reality is spread across EHRs, claims, labs, pharmacy, imaging, devices, care management, specialists and social context, while the decisions have to be made continuously anyway.
At a glance
Sequence is part of the decision
Two cases in the same state can require different answers because of how they arrived there.
Information arrives in a different order every time
Results, notes and prior history land incrementally, and the order can change what is permissible.
Thin evidence carries the highest consequence
The populations with the least data are where autonomy should be most limited.
Oversight needs the record, not a second opinion
Assurance records what was relied on, without second-guessing clinical judgment.
Patient & member state
One changing state
ANANTDECISION
understand · predict · prioritize · decide
ANANTVERIFY
evidence · authority · test · assure
NEXT BEST ACTION
Payer
members, claims, utilization
Hospital & health system
flow, pathways, deterioration
Physician & specialty
longitudinal care
Dialysis & renal
protocols, labs, fluids
Pharmacy
adherence, therapy, safety
Allied & post-acute
rehab, SNF, transitions
Home health
visits, monitoring, escalation
Diagnostics
labs, imaging, pathways
Behavioral health
engagement, coordination
Life sciences
trials, protocol execution
“What should happen next?”
Patient State → Decision Intelligence → Assurance → Action → Outcome
Fragmented state, continuous decisions
Every part of the healthcare ecosystem holds a piece of the patient, and none of them holds the patient. The decisions, meanwhile, do not wait for the pieces to be joined up.
From healthcare data to healthcare state
The patient is not a collection of records
One patient, one longitudinal state
Fragmented sources
- claims
- EHR
- labs
- pharmacy
- devices
- procedures
- protocol
- care team
- social and operational events
State
what is true about this patient now
Trajectory
how it got there, and how fast
Next best action
what should happen, and to whom
Outcome
what happened after it was acted on
The last step returns to the first: the outcome becomes part of the state the next decision is made from.
An EHR knows encounters. A laboratory knows results. A pharmacy knows medications. A payer knows claims. A dialysis system knows treatments. A monitoring device knows measurements. A care-management platform knows interventions. Each holds part of the story, in its own system of record.
What the patient's reality is distributed across
- EHRs
- claims
- labs
- pharmacy
- imaging
- devices
- care management
- specialists
- dialysis centers
- post-acute providers
- home health
- behavioral health
- payer systems
- social and operational information
The patient is a changing state.
A decision layer across healthcare
Connect the journey without replacing the systems that run it
Connected state to next best action
HEALTHCARE SYSTEMS
CONNECTED HEALTHCARE STATE
ANANTDECISION
- Understand
- Predict
- Prioritize
- Decide
ANANTVERIFY
- Evidence
- Authority
- Test
- Assure
NEXT BEST ACTION
- Clinician
- Care team
- Patient
- Payer
- Pharmacist
- Operations
OUTCOME
The systems the organization already runs stay where they are; the decision layer sits across them rather than in place of them.
Nothing about this requires replacing the EHR, the claims platform, the pharmacy system, the laboratory system, the radiology systems, the care-management platform, the CRM, the ERP, the devices or the data platform. The decision layer sits across them.
Keep what you already run
- EHR
- claims platform
- pharmacy system
- laboratory system
- radiology systems
- care-management platform
- CRM
- ERP
- devices
- data platform
- specialty applications
Longitudinal patient intelligence
The unit is the journey, not the encounter
History to new state, as a loop
Fragmented sources
- claims
- EHR
- labs
- pharmacy
- devices
- procedures
- protocol
- care team
- social and operational events
State
what is true about this patient now
Trajectory
how it got there, and how fast
Next best action
what should happen, and to whom
Outcome
what happened after it was acted on
The last step returns to the first: the outcome becomes part of the state the next decision is made from.
One encounter and one claim are convenient units for the systems that produce them, and the wrong unit for the decision. What matters clinically is the direction of travel across all of them.
Connected state can also form a working representation of the patient, the pathway, the resources and the relationships around them — the substrate for prediction, simulation and protocol evaluation rather than only for reporting.
Not simply what happened, but what is happening, where it is going and what should happen next.
Two complementary capabilities
One capability decides. A second decides whether it may be relied upon.
Together
- STATE
- DECIDE
- VERIFY
- ACT
- LEARN
The decision layer proposes within the protocol; the assurance layer releases within the authority the organization configured.
The two questions are different, and a system answering both would be grading its own work.
The exact decision authority remains with the healthcare organization and the appropriate licensed professionals.
One architecture across the healthcare ecosystem
Different organizations, one healthcare decision architecture
Nine parts of the healthcare ecosystem
Payers
Member intelligence, utilization, authorization workflows, care management, payment integrity and population health
Hospitals and health systems
Clinical operations, patient flow, care pathways, deterioration, discharge and resource decisions
Physician and specialty practices
Longitudinal care, protocols, risk, follow-up and next best action
Dialysis and renal care
Continuous CKD and dialysis state, treatment protocols, labs, medications and intervention
Pharmacy
Medication intelligence, adherence, safety workflows, specialty pharmacy and therapy management
Allied and post-acute services
Rehabilitation, skilled nursing, home health, hospice, DME and coordinated transitions
Diagnostics
Laboratory, imaging and diagnostic workflow intelligence
Behavioral health
Longitudinal engagement, care coordination and operational decision support
Life sciences
Protocol execution, evidence, trial operations and patient-support decisions
Different organizations, one healthcare decision architecture.
Payers, health systems, physician practices, renal care, pharmacy, post-acute services, diagnostics, behavioral health and life sciences each describe their decisions in their own vocabulary. The architecture underneath is the same one.
Where the decisions are immediate
Hospitals make thousands of decisions a day, most of them across a state that changed in the last ten minutes — which is precisely the case that retrospective dashboards handle worst.
Hospitals and health systems
From retrospective dashboards to continuous patient and operational intelligence
The questions a ward actually asks are operational and immediate, and answering them a day later is not answering them.
The questions that recur
- who needs attention?
- who is deteriorating?
- who is ready for discharge?
- which patient is deviating from pathway?
- where should a bed go?
- which case requires escalation?
- which follow-up is missing?
Example: patient deterioration
The value is not another alert
Three moments, twenty-two minutes apart
- 08:00vitals stablelaboratory trajectory acceptablepatient progressing normally
- 10:15respiratory rate risingoxygen requirement increasinglaboratory trend changing
- 10:22Multiple individually modest changes combine into a meaningful trajectory
AnantDecision
ESCALATE CLINICAL REVIEW
what changed · how quickly it changed · which observations contributed · relevant trajectory · why the patient moved into an intervention region
The appropriate clinician remains the decision authority; what the system adds is the reasoning behind the escalation, not a verdict in place of one.
Each observation on its own may be unremarkable. What makes this case worth escalating is the combination and the speed, and a system that reports the number without the trajectory has told the clinician very little.
Patient flow and capacity
The same architecture, applied to the facility rather than the patient
Flow decisions are made against four states at once, and any one of them can make an otherwise sensible decision wrong.
Supporting decisions such as
- admission prioritization
- bed placement
- ICU transfer workflow
- discharge readiness
- transport
- operating-room utilization
- staffing
- care-team workload
- post-acute placement
- resource allocation
Reasoned over
- patient state
- facility state
- resource state
- workflow state
Surgical care and ERAS
Perioperative protocols as continuously evaluated pathways
Represented as one changing patient state across the whole pathway, so a deviation in recovery can be read against the phase the patient was expected to be in.
A perioperative pathway is the clearest example on this page of a protocol that spans weeks and several care settings, and the state that matters is the change between its phases rather than the value at any one of them.
Example: postoperative recovery
Protocol-aware intelligence without replacing clinical judgment
A recovery that left its expected trajectory
- Observedmobility decreasespain increasesoral intake declinestemperature trends upward
- Supporting contextexpected pathwayobserved trajectoryrelevant protocol staterecent interventions
AnantDecision
REVIEW RECOVERY DEVIATION
AnantVerify evaluates
- required state
- evidence
- applicable clinical authority
Independent assurance evaluates whether the recommendation has the required state, evidence and applicable clinical authority before any downstream automation acts on it.
No single observation here determines the next action. The decision sits in the combination and in the protocol state the patient was expected to be in.
Members, claims and benefits
A payer sees the member through claims, benefits, authorizations and pharmacy rather than through care, and every one of those interactions opens a decision.
Payers and health plans
Turn member and claims data into continuous decision intelligence
The payer's view of a member is assembled from interactions that arrive separately, which is why the intervention so often happens later than it needed to.
Where the decisions sit
- members
- providers
- claims
- benefits
- authorizations
- care management
- risk
- utilization
- quality
- pharmacy
- payment integrity
Resolved to
- next best intervention
- care-management priority
- utilization workflow
- claim routing
- payment review
- member outreach
- provider engagement
- exception handling
Example: high-risk member intervention
Risk, context, decision, action
One member, four disconnected signals
- MemberMultiple chronic conditions
- Recent eventsemergency visitnew medicationmissed specialist follow-up
- ClaimsIncreasing utilization
- PharmacyMedication refill gap
- Current systemEach signal exists in a different workflow
AnantDecision
CARE MANAGEMENT INTERVENTION
contributing factors · suggested workflow
The output is not another risk score. It is a decision with the context that produced it, routed to a care manager who can act on it.
Every signal in this case already exists somewhere in the plan. The problem is that each one lives in a different workflow, so none of them has ever been read together.
Utilization management
The clinical, administrative and policy state, assembled before the workflow decision
Utilization decisions are made against policy, and the policy is only half of what determines the right answer — the other half is the clinical state the request actually describes.
Decision workflows
- authorization triage
- clinical-review prioritization
- documentation completeness
- level-of-care review
- continued-stay workflows
- exception routing
- appeal prioritization
Where organizational policy requires it, AnantVerify provides independent assurance for the AI-assisted decision rather than letting the request review itself.
Claims and payment integrity
Determine the appropriate next action for each claim
Payers process claim decisions at a volume that makes individual review impossible, so the decision has to be expressed in the claim's own vocabulary and evaluated against everything attached to it.
Decision vocabulary
- PAY
- PEND
- REVIEW
- REQUEST INFORMATION
- INVESTIGATE
- DENY according to applicable policy and authority
Evaluated in the context of
- member
- provider
- benefit
- authorization
- clinical history
- claim history
- relationships
- coding patterns
- policy
- payment patterns
The objective is not fraud detection. It is the right disposition for this claim.
Renal protocols and trajectory
Renal care is the most measurement-dense chronic environment in healthcare, and its variables interact continuously — which makes it the clearest case for state and trajectory rather than thresholds.
Dialysis and renal care
Continuous intelligence for a continuously changing chronic state
A dialysis patient generates repeated longitudinal observations that only mean something in relation to each other and to where they were a month ago.
Observations that arrive continuously
- laboratory results
- treatments
- weight
- blood pressure
- vascular access
- medications
- nutrition
- anemia management
- mineral and bone disorder
- dialysis adequacy
- fluid status
- hospitalizations
Renal protocol intelligence
The clinical organization defines the protocol; the platform evaluates against it
Every one of these protocol areas has a defined target, a defined tolerance and a defined owner. The platform's contribution is to notice movement before the next scheduled review, not to choose the target.
Protocol areas
- anemia management
- iron status
- CKD mineral and bone disorder
- fluid and dry-weight management
- dialysis adequacy
- vascular access
- blood pressure and volume
- nutrition
- electrolytes
- medication management
- infection prevention
- transplant workflow
Example: anemia management
A rule reacts to a value; the decision depends on the trajectory
A protocol value read in context
- Patient stateHemoglobin trending downward
- Iron stateTSAT decliningferritin available
- TreatmentESA active
- Recent eventshospitalizationtreatment interruption
- Evaluated togethercurrent statetrajectorymedication historytreatment eventsprotocolrelated physiology
AnantDecision
CLINICAL REVIEW — ANEMIA MANAGEMENT
relevant evidence · relevant trajectory
AnantVerify evaluates
- required state
- protocol evidence
Assurance evaluates whether the required state and protocol evidence are present before the recommendation enters a governed workflow.
The same hemoglobin reading means different things before and after a hospitalization, and a threshold rule cannot express that difference.
Beyond rules: mechanistic and learned intelligence
Where purely statistical prediction is not enough
Renal physiology is a domain where the mechanism is partly known, and a model that ignores it has thrown away information the clinician is using. The architecture supports both.
What can be combined
- clinical protocols
- deterministic rules
- statistical models
- machine learning
- temporal models
- continuous-time models
- mechanistic models
- causal reasoning
- human clinical judgment
Medication decision intelligence
Medication state is one of the few places where the clinical, the operational and the financial views of a patient are the same object — which makes it a good place for one decision instead of three alerts.
Pharmacy
From medication records to medication decision intelligence
Medication-related state spans the prescription, the dispense, the laboratory result that says whether it is working, the coverage that decides whether it continues and the patient's behavior in between.
Medication state
- prescriptions
- dispensing
- adherence
- laboratory results
- diagnoses
- allergies
- interactions
- therapy response
- coverage
- authorization
- refills
- clinical outcomes
Pharmacy use cases
Clinical recommendations stay with the people licensed to make them
The workflows below are coordination and review decisions, and each one lands in front of a pharmacist, a prescriber or an operational owner rather than executing on its own.
Workflows
- adherence intervention
- refill outreach
- therapy review
- specialty pharmacy coordination
- prior-authorization workflow
- medication reconciliation
- high-risk medication review
- therapy monitoring
- inventory and operational decisions
- patient engagement
Appropriate clinical recommendations remain subject to pharmacist, prescriber and organizational authority.
Example: specialty pharmacy
One patient state, one coordinated next action
Three alerts that are one decision
- PatientComplex specialty therapy
- StateMedication active
- Pharmacy eventRefill overdue
- Clinical stateRecent lab monitoring incomplete
- CoverageAuthorization approaching expiration
AnantDecision
COORDINATE THERAPY REVIEW
refill · laboratory · authorization
One patient state and one coordinated next action, instead of three disconnected alerts routed to three different queues.
Three separate systems would raise three separate alerts here, and the person receiving them would have to reconstruct the situation before deciding anything.
Where care continues
Most of a patient's care happens outside the hospital, in settings that see them repeatedly and briefly. Those settings are where a longitudinal state is worth the most and is usually least available.
Allied health and post-acute care
Decision intelligence across the settings care continues into
Rehabilitation, skilled nursing, home health, hospice, respiratory therapy, DME and community care each make decisions that a hospital never sees and that the next setting always inherits.
Settings
- physical therapy
- occupational therapy
- speech therapy
- rehabilitation
- skilled nursing
- home health
- hospice
- respiratory therapy
- DME
- ambulatory services
- community care
Example: rehabilitation
Clinical trajectory, read with operational reality
A care plan and an authorization, together
- PatientPost-surgical rehabilitation
- Expected trajectoryProgressing toward a mobility milestone
- Actual stateattendance decliningfunctional improvement slowingpain score increasing
- Operational stateAuthorization has limited visits remaining
AnantDecision
REVIEW CARE PLAN + AUTHORIZATION
clinical trajectory · operational reality
The decision combines the clinical trajectory with the operational constraint, which is necessary because in post-acute care the two are the same decision.
In post-acute care the clinical decision and the administrative decision are rarely separable: what is appropriate depends on what remains authorized.
Home health and skilled nursing
Continuous awareness of who needs attention now
Home health distributes the observations across visits, which makes each one look isolated and the pattern invisible. Connecting them to the broader patient state is the difference between a visit record and a decision.
Home health decisions
- visit prioritization
- risk escalation
- care-plan adherence
- medication concerns
- missed visits
- wound progression
- remote-monitoring changes
- hospitalization risk
- staff allocation
Skilled nursing and post-acute decisions
- admission
- care-plan progression
- fall risk
- wound management
- therapy
- medication
- transfer risk
- discharge readiness
- staffing
- resource prioritization
Behavioral health
Longitudinal patterns rather than isolated measurements
Engagement and adherence in behavioral health are patterns over months, and the operational decisions that follow them — escalation, follow-up, resource allocation — need the same privacy and oversight controls the clinical record has.
Supported workflows
- engagement
- missed appointments
- care coordination
- treatment adherence
- follow-up
- utilization
- escalation
- resource allocation
With appropriate professional oversight and privacy controls maintained throughout.
Allied service groups
Care decision intelligence and operational decision intelligence
For a diversified allied-services organization the opportunity is broader than any individual clinical recommendation: it is the operating model that connects referrals, therapists, authorizations, utilization and reimbursement.
What the platform connects
- referrals
- patients
- therapists
- appointments
- care plans
- authorization
- documentation
- utilization
- outcomes
- staffing
- locations
- reimbursement workflows
Diagnostics, specialties and chronic care
Diagnostics produce the most consequential single events in the record, and chronic care produces the least noticeable ones. Both need the state around them to mean anything.
Laboratory, imaging and diagnostics
Connecting a diagnostic event to the pathway it belongs to
A result is only critical in relation to a patient, and imaging AI is only useful inside the pathway that acts on the finding. Both are stronger when they can see the state around them.
Diagnostic workflows
- result prioritization
- follow-up
- critical-value routing
- repeat-testing workflow
- diagnostic pathway progression
- operational workload
- quality exceptions
Imaging AI, complemented by
- patient history
- prior imaging
- diagnoses
- labs
- medications
- care pathway
- follow-up state
Supporting prioritization, follow-up workflow, care coordination and decision assurance around consequential AI-assisted actions.
Specialty care
Each specialty defines its own state, protocol and authority
Cardiology and oncology do not share a vocabulary, and a platform that flattened them into one would be useless to both. What they share is the architecture underneath.
Specialties
- cardiology
- oncology
- nephrology
- endocrinology
- neurology
- pulmonology
- orthopedics
- surgery
- gastroenterology
- rheumatology
Defined by each specialty
- state
- protocols
- evidence
- decision vocabulary
- authority
- outcomes
Oncology
Protocol-aware coordination across a treatment journey
Cancer care combines a long longitudinal state with a protocol that changes as the state does, and the coordination decisions between cycles are where the delays accumulate.
Longitudinal state
- diagnosis
- stage
- pathology
- biomarkers
- therapy
- laboratory results
- imaging
- toxicity
- response
- supportive care
- comorbidities
Supported decisions
- treatment readiness
- toxicity review
- supportive-care intervention
- follow-up
- protocol deviation
- care coordination
- operational next best action
Chronic disease management
Where most of the cost and most of the outcome is decided
Chronic conditions are managed between appointments rather than in them, which is why the useful decision is the next best intervention rather than the next measurement.
Domains
- diabetes
- chronic kidney disease
- heart failure
- COPD
- hypertension
- cardiovascular disease
- multiple chronic conditions
Continuously evaluated
- clinical state
- trajectory
- medication
- utilization
- engagement
- protocol
- risk
Life sciences and clinical research
The same state-based architecture, applied to a study
Trial operations run on protocol execution and eligibility, which are decision problems of exactly the kind this architecture already describes.
Decision environments
- clinical research
- trial operations
- protocol execution
- site operations
- patient eligibility workflows
- protocol deviations
- evidence management
- safety workflows
- study operations
Represented as state
- patient
- protocol
- site
- study
- event
- intervention
- outcome
Population health and healthcare operations
The operational half of healthcare decides as often as the clinical half and is usually given less intelligence. It runs on the same architecture.
Population health
Move from risk lists to intervention intelligence
Beyond the score
- Patient risk = HIGH
- Why?
- What changed?
- What intervention is appropriate?
- Who should act?
- When?
And then two more questions the risk list never asks: did it happen, and did it work?
Risk stratification is where population health usually stops: a list of names, sorted. The decisions that change an outcome all happen after that point.
Care management
A queue of thousands is not a prioritization
Scarce clinical resource spends its time re-deriving which case matters, because the list tells it who is high-risk and nothing about why now.
What the decision determines
- who needs attention
- why
- what changed
- what intervention may be appropriate
- which evidence matters
- what should happen next
Patient engagement
Next best action, extended to the patient
Many of the decisions that change an outcome are made by the patient between encounters, and the channel for reaching them usually already exists.
Decision opportunities
- schedule follow-up
- complete lab
- refill medication
- prepare for procedure
- respond to care manager
- complete screening
- review education
- contact provider
The communication channel can remain within the patient-engagement systems the organization already runs; the decision layer determines which opportunity is worth raising.
Healthcare operations
Not every important healthcare decision is clinical
Staffing, scheduling, supply, bed management, revenue cycle, procurement, facilities, equipment, transport and financial operations all make decisions under the same conditions: changing state, defined policy and a cost of getting it wrong.
Operational decision areas
- staffing
- scheduling
- supply chain
- pharmacy inventory
- bed management
- revenue cycle
- claims
- procurement
- facilities
- equipment
- transportation
- financial operations
Revenue cycle
What should happen next for this account or claim?
A work queue assigns a document to a person. It does not tell them what the account needs, which is why the second and third touches are so often the same work done twice.
Decisions
- documentation review
- coding workflow
- claim readiness
- denial prevention
- authorization status
- claim follow-up
- appeal prioritization
- payment variance
- work-queue prioritization
Healthcare supply chain
Availability, read against clinical demand
Medications, devices, implants and supplies are procured and held far from the clinical decision that depends on them, which is why a shortage is usually discovered at the point of care rather than before it.
Decisions
- reorder
- substitute
- allocate
- expedite
- hold
- investigate
- reroute
Held
- medications
- devices
- implants
- supplies
- vendors
- inventory
- distribution
Protocols, evidence and clinical authority
This part is the one a health system will read closest, because it is where the platform states what it does not decide.
Protocol intelligence
From protocol documents to protocol intelligence
What a protocol becomes when it is executable
Usually lives as
- documents
- guidelines
- order sets
- rules
- human knowledge
Applied slightly differently each time, with no record of which version was used.
Represented as a decision structure, it holds
- eligibility
- state requirements
- labs
- medications
- procedures
- timing
- contraindications
- decision boundaries
- evidence
- authority
- next best actions
Which can be evaluated, tested against changed state and replayed.
A protocol that lives in a document, an order set and several people's memory is applied slightly differently each time, and nothing records which version was used. The same protocol expressed as a decision structure can be evaluated, tested and replayed.
Evidence and configured authority
AnantVerify evaluates configured authority. It does not invent clinical authority.
A recommendation is only usable if the person receiving it can see what it rests on and what governs it — including the evidence that argues against it, and what is simply missing.
What the record makes explicit
- what patient state supports it
- what evidence supports it
- which protocol applies
- which evidence contradicts it
- what is missing
- how current the information is
- what authority governs the action
Sources the organization configures as authoritative
- clinical guidelines
- peer-reviewed evidence
- approved local protocols
- formulary rules
- organizational policies
- regulatory requirements
- specialty guidance
- authorized clinical judgment
The healthcare organization determines which sources are authoritative for each decision context.
Test the decision, not just the model
When the patient changes, does the AI decision change in the right way?
Five ways to change the question
An ordinary test asks whether the answer was correct for this case.
- Change something irrelevantDoes the recommendation remain stable?
- Change a material labDoes the recommendation change appropriately?
- Change medication stateDoes behavior respond correctly?
- Combine two conditionsDoes an interaction expose unexpected behavior?
- Change the sequenceDoes timing matter?
Asking whether the answer was right for one case says nothing about what the system does when a material input moves.
Asking whether an answer was right for a case tells you about that case. It does not tell you what the system will do when a material input moves, which is the question that matters between the examples a model was validated on.
Clinical decision boundaries
Not to replace clinical judgment, but to show where AI behavior changes
Healthcare decisions cluster at boundaries, and a small change in a number can cross several of them at once. Which boundary applies is a matter of organizational policy, not of model inference.
What a small change can alter
- risk classification
- protocol status
- treatment eligibility
- monitoring requirement
- escalation
- discharge readiness
Human clinical authority remains explicit
AI intelligence should strengthen clinical judgment, not silently replace it
The distinction the page has to make unmistakable is between a recommendation the platform produced and the authority to act on it, which the platform does not hold.
Graduated autonomy for healthcare
Where AI may assist, where it may automate, and where human authority is mandatory
Four regions, four responses
Which region a decision sits in is defined by the healthcare organization, not chosen by the model.
One operating model for every decision would either over-constrain the routine ones or under-constrain the consequential ones. The organization defines where each decision sits.
Interoperability
Intelligence should participate in the healthcare workflow
A decision layer that requires a new data store and a new integration project will not be adopted by an organization that already runs a healthcare interoperability estate.
Standards and interfaces
- FHIR
- HL7
- APIs
- event streams
- databases
- files
- enterprise integration platforms
The objective is not another isolated AI data store.
The business case and how it is measured
Every figure in this part is a worked assumption rather than a result. The value of a healthcare decision program is measured on outcomes, and the hours released are only evidence that the capacity existed.
The business case
Six levers, each improving something the organization already reports
Healthcare AI should improve measurable outcomes for patients, clinicians, operations, payers and the organization. If it cannot be expressed in one of these six terms, it cannot be defended at the end of the year.
1 · Clinical capacity
- focus scarce clinical resource where human expertise matters most
2 · Operational efficiency
- less unnecessary manual review, coordination and rework
3 · Intervention timing
- identify meaningful changes earlier
4 · Care coordination
- connect decisions across fragmented settings
5 · Avoidable utilization
- support appropriate interventions that may reduce unnecessary utilization
6 · Decision assurance
- visibility into why consequential AI-assisted decisions were permitted, escalated or stopped
Lever 5 is expressed as support for appropriate intervention rather than as a reduction in utilization, because the organization's clinical judgment decides what is appropriate.
Illustrative business cases
Four settings, the same arithmetic
Four worked assumptions
No figure on this panel is a clinical result.
Care management
Assumptions
- 500,000 members managed
- 10% in active review = 50,000
- 20 minutes per member per review cycle
- a 5-minute reduction in initial prioritization effort
50,000 × 5 minutes, per review cycle
4,167 staff hours per cycle
What matters more than the hours
- faster intervention
- greater protocol adherence
- fewer avoidable exceptions
- better patient outcomes
Claims operations
Assumptions
- 1,000,000 claims requiring additional review a year
- 8 minutes of average review time
- a 2-minute reduction in avoidable review effort per claim
1,000,000 × 2 minutes
33,333 operational hours a year
What matters more than the hours
- the organization's loaded cost
- resulting capacity
- claim cycle time
Dialysis
Assumptions
- 10,000 patients
- multiple protocol dimensions reviewed monthly
- a 10-minute reduction in preparation and review per patient per month
10,000 × 10 minutes × 12 months
20,000 clinical-support hours a year
What matters more than the hours
- faster intervention
- greater protocol adherence
- reduced avoidable exceptions
- better patient outcomes
Pharmacy
Assumptions
- 100,000 specialty patients
- 30% of cases requiring coordination = 30,000
- 25 minutes of average coordination time
- a 5-minute reduction per affected patient
30,000 × 5 minutes, per intervention cycle
2,500 staff hours per cycle
What matters more than the hours
- therapy continuity
- adherence
- authorization cycle time
- avoidable abandonment
Every line above is a share of volume multiplied by a unit of time. None of them asserts that a patient outcome improved.
Each figure below converts a stated time saving on a stated volume into hours released. The hours are the easy part to calculate and the least important part of the argument.
Illustrative assumptions only. The actual value depends on review frequency, staffing costs and workflow design, and none of these figures is a clinical outcome.
Build ROI from the organization's own economics
No universal percentage should be assumed
A vendor improvement rate cannot be verified, audited or reproduced by the organization that buys it. A model built from the organization's own volumes, costs and baselines can be all three.
Enterprise healthcare KPIs
Four layers, because a single number would hide the trade-off
A program that reports only automation will eventually automate something it should not have; a program that reports only assurance will report a good score for doing less.
Clinical
- time to intervention
- protocol adherence
- clinical escalation rate
- avoidable deterioration indicators
- readmission
- care-gap closure
- treatment adherence
- outcome measures appropriate to the program
Operational
- decision cycle time
- manual review rate
- cases per clinician
- cases per care manager
- touches per case
- queue age
- rework
- SLA attainment
- automation rate
- cost per decision
Decision intelligence
- decision acceptance
- model calibration
- abstention
- decision stability
- boundary behavior
- decision drift
- next-best-action acceptance
- intervention completion
Assurance
- state sufficiency
- stale-state rate
- evidence conflicts
- authority conflicts
- ALLOW / HOLD / ESCALATE / BLOCK
- assurance coverage
- human overrides
- interaction failures
- execution mismatches
- replay completeness
Domain KPIs
Each setting measures what its own operation can move
A single cross-organization scorecard would hide exactly the differences that make the program work in one setting and not another.
Payer
- care-management productivity
- authorization cycle time
- manual review rate
- claim processing time
- payment integrity yield
- member outreach completion
- care-gap closure
- avoidable utilization indicators
- cost per case
Hospital
- time to intervention
- length of stay
- discharge cycle time
- readmission
- bed turnaround
- protocol adherence
- clinical escalation
- patient-flow delays
- staff workload
- decision-to-action time
Dialysis
- protocol adherence
- anemia management indicators
- mineral and bone disorder indicators
- dialysis adequacy
- fluid-management indicators
- vascular-access events
- hospitalization
- missed treatments
- time to clinical review
- intervention completion
Pharmacy
- adherence
- refill gap
- therapy abandonment
- time to therapy
- authorization cycle time
- pharmacist intervention rate
- therapy review completion
- coordination touches
- patient engagement
Allied and post-acute
- functional improvement
- care-plan adherence
- visit utilization
- missed visits
- time to escalation
- hospital transfer
- authorization utilization
- discharge readiness
- staff productivity
- patient outcome measures
Revenue cycle
- clean-claim rate
- denial rate
- days in A/R
- touches per account
- authorization-related denials
- appeal cycle time
- work-queue productivity
- payment variance
- cost to collect
Measure intelligence and outcomes together
Prediction is only the beginning of healthcare value
The complete chain
- Was the right patient identified?
- Was the recommendation appropriate?
- Was it acted upon?
- How quickly?
- What happened to the patient or process?
- What did the organization learn?
Model accuracy and the number of predictions made are the two measures most easily reported and least related to whether anything improved for a patient.
One journey, then the next
Nothing here requires transforming a health system at once. It requires one decision journey, its baseline measured before anything changes, and a willingness to expand only after the numbers move.
One patient, one journey, continuous intelligence
The patient is not fragmented; the systems are
Care is split across organizations, systems, specialties, encounters, claims and workflows, and every one of those boundaries is a place where the state is dropped and the decision has to be made without it.
A decision architecture organized around the changing healthcare state does not remove those boundaries. It stops them from being where the decision is made.
Start with high-value decision journeys
Start where the timing and the state both support it
The first journey should be chosen because the decision repeats, the state exists somewhere already and the outcome can be measured — not because it is the most visible program in the organization.
Choose a journey where
- state changes frequently
- decision volume is meaningful
- human capacity is constrained
- timing matters
- outcomes are measurable
- the next action can be clearly defined
Strong starting points
- care-management prioritization
- CKD and dialysis protocols
- perioperative pathways
- claims review
- authorization workflow
- pharmacy coordination
- post-discharge follow-up
- revenue-cycle exceptions
- hospital patient flow
- chronic disease management
Prove one decision journey, then expand
One journey becomes a healthcare decision architecture
Baseline before Define would be cheaper and would produce a number nobody trusts; Measure before Verify would report the automation without the control.
The order matters: the baseline is only credible if it was captured before the workflow changed, and the second journey is only funded if the first one survived review.
From systems of record to systems of care intelligence
What is happening, what is changing, and who has authority?
Healthcare has spent decades digitising records, orders, claims, medications, labs, images and encounters. That work is done, and the next step is not more data.
It is turning that state into decisions, which means answering:
- what is happening?
- what is changing?
- who needs attention?
- what should happen next?
- what evidence supports it?
- who has authority?
- what happened after we acted?
That is the transition from healthcare information to healthcare decision intelligence.
Better intelligence for every journey
Observe the patient journey. Understand the trajectory. Recognize what changed. Determine what should happen next. Assure the consequential AI-assisted decisions. Learn from the outcome.
ANANTDECISIONHealthcare Decision Intelligence
ANANTVERIFYIndependent Decision Assurance
- State
- Decide
- Verify
- Act
- Learn
Assurance where decisions become consequential.
Worked at decision level: Clinical Protocol Decision Support and Pre- and Post-Operative Care.
AnantVerify is not a medical device claim or a substitute for clinician judgment. Healthcare deployments require appropriate clinical validation, governance and regulatory review.
Illustrative throughout. Every population, volume, duration and hour on this page is a worked assumption for building a value model — not a measured result, not a clinical outcome, and not a promised benefit. Replace them with the organization's own economics.