Network analysis built on resolved provider identity, a stated attribution method and honest treatment of small numbers, because what comes out of this work becomes contract positions, network decisions and regulatory filings.
Provider performance comparison is the most consequential analysis a payer produces and the most fragile. It depends on knowing which entity you are measuring, which members and episodes belong to them, and whether the volume supports an estimate at all. Get any of those wrong and the output still looks authoritative, gets taken into a negotiation, and is very difficult to withdraw once a relationship has been affected.
A provider ranking built on unstable estimates is not a finding. It is a list, and somebody will act on it.
A provider is not one thing. The same clinician appears as an individual, within one or more groups, at several locations, billing under different entities, participating in multiple networks under separate contracts, each with its own effective dates. Network analysis requires deciding which of those is the unit of measurement, and most plans have never made that decision explicitly because no single function owns provider data end to end.
The consequences compound quietly. Cost attributed to the wrong entity. Adequacy measured against a directory that does not reflect who is actually accepting patients. Performance compared across providers whose panels are not comparable. Each output is precise, well formatted and unreliable in a way that is invisible from the report.
Provider identity is unresolved
Individual, group, facility, location, billing entity, network and contract are different units. Analysis that mixes them attributes cost and performance to the wrong place.
Attribution decides the answer
Which members and episodes belong to a provider changes the result substantially, and the method is frequently undocumented and inconsistent between analyses.
Most providers have too little volume
A large share of any network has too few attributed members for a stable estimate. Ranking them produces a list driven by noise.
Adequacy is measured against a directory
If the directory overstates who is participating and accepting patients, adequacy analysis measures a network that does not exist in practice.
Claims show utilization, not availability
Claims record where members received care. They cannot show members who could not get an appointment, abandoned the search, delayed care or travelled farther.
Leakage is reported without capacity context
Out-of-network use because nothing in network was available is an adequacy finding, not a steerage failure.
This analysis becomes a contract position. Build it accordingly.
Most analytics can be wrong and corrected. Network analysis is carried into negotiation, termination decisions, steerage design and regulatory filings. The standard of evidence has to be higher here than anywhere else in the payer analytics estate.
Three things have to be settled before any network finding is credible: which entity is being measured, which activity belongs to it, and whether there is enough of that activity to support a conclusion. Analysis built before those are agreed produces output that cannot be defended when a provider disputes it, and providers do dispute it.
Step 1
Resolve provider identity across individual, group, facility, location, billing entity, network and contract, with effective dates, since participation on the service date is not participation today.
Step 2
Choose and document the unit of measurement per analysis. A group-level finding and an individual-level finding are different claims and should not be presented interchangeably.
Step 3
Define attribution explicitly. Which members, which episodes, on what basis, over what period, and what happens when more than one provider has a claim.
Step 4
Establish reliability thresholds. Minimum volume for a stable estimate by measure, below which a provider is reported as insufficient rather than ranked.
Step 5
Apply risk and case mix adjustment wherever providers or populations are compared, with the basis and its limits stated alongside the result.
Step 6
Separate composition from performance. Network structure, adequacy and access are one set of questions. Cost and quality of those providers are another.
Step 7
Reconcile adequacy to reality. Directory participation tested against claims activity, so adequacy reflects providers who are actually seeing members.
Step 8
Analyse leakage with capacity context, distinguishing patient or referral choice from the absence of an in-network option.
Step 9
Route findings to the function that can act: contracting, network development, provider relations, medical management or compliance.
Step 10
Track what changed afterwards. Whether access improved, leakage declined, utilization shifted or provider behaviour moved, closing the loop from evidence to action to measured resul
Say how you attributed, every time.
The same provider can look above or below expectation depending on whether attribution is by plurality of visits, by assignment, by episode or by cost concentration. State the method with the result and the conversation becomes about performance rather than about arithmetic.
The analytical techniques here are well established. What separates network analytics that survives a provider challenge from analytics that does not is the measurement infrastructure underneath: identity, attribution, adjustment and reliability, built once and applied consistently.
Establish Measurement
Provider Identity Resolution
Individual, group, facility, location, billing entity, network and contract relationships mastered with effective dates.
Attribution Methodology
Member and episode attribution defined, documented and applied consistently, with sensitivity to method stated where a finding would change under a reasonable alternative.
Risk and Case Mix Adjustment
Adjustment applied to every provider or population comparison, with the model, its inputs and its limitations published rather than referenced.
Reliability and Small Number Handling
Minimum volume thresholds per measure, with insufficient-volume providers identified as such rather than ranked, and confidence intervals shown where an estimate is presented.
Analyse the Network
Network Composition, Adequacy and Access
Coverage by specialty, geography and product against applicable standards, reconciled to claims activity so adequacy reflects providers genuinely seeing members, alongside distance and availability analysis.
Provider Cost and Utilization Performance
Risk-adjusted cost, utilization and practice pattern comparison with volume reliability applied, and the drivers of variation identified rather than only the variation.
Referral and Leakage Analysis
Where attributed members receive care outside the network or outside a preferred group, with capacity context so an adequacy gap is not reported as a steerage failure.
Act and Govern
Contract and Rate Analytics
Expected versus actual reimbursement, rate position by market and specialty, and the cost impact of proposed terms modelled before negotiation rather than after.
Network Scenario Modelling
The effect of adding, removing or retiering providers on cost, access, adequacy and member disruption.
Source Authority by Attribute
Define which system governs which attribute, stated explicitly rather than resolved case by case during mastering.
Directory Accuracy Analytics
Discrepancy between directory state and claims-evidenced activity.
Directory Accuracy Analytics
Discrepancy between directory state and claims-evidenced activity.
What CaliberFocus does, and does not do?
We will not produce a provider ranking where the volume does not support one, and we will not present a comparison that cannot be adjusted credibly. Both are commonly requested, both look like the deliverable, and both create positions the plan has to defend to a provider who will bring their own numbers. We will also tell you when the finding is that provider data quality is the constraint, because network analytics built on unresolved identity is precise and wrong in a way that only surfaces during a dispute.
Every analysis below is routinely produced. What determines whether it survives contact with a provider, a regulator or a board is the measurement condition in the third column, and that is usually assessed after the analysis has been circulated.
| Analysis | The Question It Answers | What Has to Be True for It to Hold |
|---|---|---|
| Network Adequacy | Do we meet access standards by specialty and geography? | Directory reconciled to claims activity, or you are measuring a network that exists on paper. |
| Access and Availability | Can members actually get an appointment? | Evidence beyond participation status, since listed is not the same as accepting. |
| Provider Cost Performance | Which providers cost more than expected? | Resolved identity, stated attribution, risk adjustment and sufficient volume. All four. |
| Practice Pattern Variation | Where does clinical practice differ materially? | Case mix adjustment, and the discipline not to read variation as inappropriateness. |
| Referral Patterns | Where do our providers send members? | Attribution of the referring relationship, which claims data supports only partially. |
| Network Leakage | Where is care going outside the network? | Capacity context. Leakage where no in-network option existed is an adequacy finding. |
| Contract and Rate Position | How do our rates compare and what would a change cost? | Contract terms as data and expected reimbursement modelled, not inferred from paid amounts. |
| Network Scenario Modelling | What happens if we add, remove or retier providers? | Member disruption quantified, not only cost effect. The disruption is what generates complaints. |
| Directory Accuracy | Does our published directory reflect reality? | A regulatory obligation in its own right, and the input that makes adequacy analysis meaningful. |
A member receiving care proves access existed for that encounter. It proves nothing about the population. Claims cannot distinguish those who could not get an appointment and gave up.
Leakage Without Capacity Context Is a Misdiagnosis
Out-of-network utilization can mean no in-network provider was available, access was poor, the member or referring provider chose otherwise, or the service was emergent. Those causes require different responses.
Provider measurement rests on four choices that are frequently made implicitly: the unit of measurement, the attribution method, the adjustment model and the reliability threshold. Each materially changes the result, and a provider disputing a finding will interrogate all four.
| Choice | The Options | Why It Matters |
|---|---|---|
| Unit of Measurement | Individual clinician, group, facility, location or contracted entity | A group can perform well while individuals within it vary widely, and vice versa. |
| Attribution | Plurality of visits, assignment, episode-based, cost concentration or specialty-specific rules | The same provider can appear above or below expectation depending on the method chosen. |
| Adjustment | Risk, case mix, demographic, social and contract differences | Unadjusted comparison attributes population differences to provider behaviour. |
| Reliability | Minimum attributed volume, confidence intervals, multi-year pooling | Most providers have too few members for a stable estimate, and ranking them reports noise. |
Network: market, product, specialty, provider group, provider, location, service, member. Leakage: market, specialty, service, referring provider, out-of-network provider, member population. Cost variation: network, provider group, provider, service category, procedure, claim.
Publish the method with the result
Unit, attribution, adjustment and reliability stated on the output rather than available on request.
Report insufficient volume as insufficient
Not as average, not as unranked, and not quietly excluded.
Test sensitivity to method
Where a finding would reverse under a reasonable alternative attribution, it is not a finding.
Separate variation from inappropriateness
Practice variation is a signal to investigate. It is not evidence that care was unnecessary.
Benchmark only where comparison is legitimate
Different specialties, populations, contract structures and coding intensity make many comparisons meaningless.
Show the driver, not only the gap
Unit cost, utilization, case mix, site of service or referral role should explain the gap.
Assume the provider will see it and bring their own numbers.
Provider-facing performance reporting is increasingly expected and frequently contested. The useful test at design time is whether you would be comfortable presenting this analysis to the provider it concerns, with the method visible, and defending each choice.
Contracting holds the agreement. Credentialing holds the verification. Rosters arrive from groups. Directory holds what members see. Claims show who actually billed. Each is maintained by a different function for a different purpose, and reconciling them is the foundational work of network analytics rather than a preliminary step.
Provider mastering across sources
Contracting, credentialing, rosters, attestation, directory and claims-derived identity reconciled into a governed master with effective dates and relationship hierarchy.
Contract terms as data
Rates, fee schedules, tiers and effective periods structured, since expected reimbursement is what separates a rate finding from a utilization finding.
Claims with final action resolved
Consistent claim state and amount definitions, drawn from the certified claims products rather than recomputed here.
Membership and attribution inputs
Enrollment, product, geography and assignment data, with retroactive change handled so attribution does not shift silently between analyses.
Credentialing and participation status
Effective-dated participation, since network status on the service date is the only status that matters for an analysis of that period.
Directory and access data
Published participation, panel status and availability, compared against claims evidence to establish whether the directory reflects reality.
Effective-dated joins throughout
Provider, contract and network attributes joined as of the service date. Current-state joins silently reassign historical activity and make trend analysis meaningless.
Hierarchy preserved, not flattened
Individual to group to system relationships retained, since analysis at one level must be able to roll to another without recomputation.
Unresolved identity stays visible
A provider that cannot be confidently matched is an exception with an owner, not a default assignment into the largest group.
Reconcile the sources rather than choosing one
Where contracting, credentialing and claims disagree, the discrepancy is the finding. Selecting whichever source is most convenient hides a data problem that is also a compliance problem.
Output that names providers and characterizes their performance affects contracts, reputations and livelihoods. It is also increasingly shared with providers themselves and sometimes published. That changes the governance requirement from analytical hygiene to something closer to the standard applied to a regulated disclosure.
Provider identity quality
- Match rate, unresolved identity volume and conflict rate monitored
- Effective-dated relationship history retained
- Named owner for provider mastering, with source authority per attribute
- Source conflicts surfaced, assigned and resolved
- Directory discrepancy against claims evidence tracked continuously
Method governance
- Each measure carries population, attribution logic, period, cost basis, utilization basis, adjustment methodology, minimum volume and owner
- Method and thresholds published and versioned
- Method changes treated as controlled releases
- Stated position on comparisons the plan will not make
- Sensitivity testing retained
Provider-facing use
- Method, adjustment, comparison group and period stated on shared reports
- Defined route for a provider to question a finding
- Legal and network leadership review before external sharing or use in a termination decision
- Reproducibility of any shared figure as at the date it was shared
Security and access
- Provider-identifiable performance output access controlled deliberately
- Member-level detail minimized, since provider analytics becomes member-level analytics when a user drills into utilization
Check whether your directory agrees with your claims.
Take a specialty and a market, list the providers your directory says are participating and accepting patients, and compare that to which of them have actually billed for a new member in the last year. The gap is usually larger than expected.
Network analytics is usually measured by reports produced and providers analysed. Neither establishes that a finding held up, that anyone acted on it, or that the provider relationship survived the conversation.
| Category | What We Measure | Why It Matters |
|---|---|---|
| Measurement Integrity | Provider match rate, unresolved identity volume, and share of providers meeting reliability thresholds | Determines whether anything downstream is defensible. |
| Directory Reality | Discrepancy between published participation and claims-evidenced activity | A regulatory measure and the input that makes adequacy analysis meaningful. |
| Finding Durability | Findings challenged by providers, and the share upheld after review | The honest test of whether the method was sound. |
| Action | Findings routed to contracting, network development or medical management, and what changed | Distinguishes network analytics from network reporting. |
| Network Performance | Risk-adjusted cost position, leakage by cause, access and adequacy by specialty and market | The business outcome, reported by cause rather than as a single rate. |
| Relationship Effect | Provider disputes, abrasion indicators and engagement with shared performance reporting | The cost that appears in contracting rather than in an analytics report. |
Two findings are likely.
A meaningful share of your providers will not have enough attributed volume to support a performance estimate, which reduces the population you can act on and is the correct answer. And your directory will disagree with your claims by more than expected.
We will take one specialty and market, resolve provider identity across your sources, apply attribution and reliability thresholds, and show you how much of that network is actually measurable and how much of your directory is evidenced in claims. That assessment usually reduces the population you can make claims about, which is uncomfortable and considerably safer than discovering it during a contract dispute.
Bring us a specialty or clinical setting where clinicians are spending too much time creating notes. We will assess where ambient documentation fits, what must remain clinician controlled, how it should integrate with your EHR, and how to measure whether it is actually reducing burden.
One conversation with people who have run these deployments, and a written readiness view you can use with or without us.
- AI Agents and Workflow Automation
- Voice and Conversational AI
- Document AI and Intelligent Processing
- Generative AI and Enterprise Copilots
- AI Strategy and Governance
- HCC and Risk Adjustment Analytics
Security & Compliance
