Operational forecasting for the demand that determines staffing and service: claims volume, call volume, authorization workload, enrollment transactions and the seasonal peaks everybody knows are coming and nobody has modelled.
A health plan has a professional forecasting function. Actuaries forecast cost, reserves and pricing to a standard and under obligations that a consultancy should not be offering to improve. What they do not forecast, because it is not their remit, is the work: how many claims will arrive next month, how many calls in the week after a mailing, how many authorizations in a specialty next quarter, and how many people are needed to handle it. That planning is usually done on last year plus a percentage.
We do not do actuarial work and we will not pretend otherwise. We forecast the operation.
Open enrollment, benefit changes, mailings, January deductible confusion and new-group go-lives are known in advance. Yet many plans still meet the resulting demand by asking people to work harder.
Known Events Are Not Modelled
Dates are available, but the demand they create is frequently met with overtime rather than planned capacity.
Operational Forecasting Has No Owner
Cost sits with actuarial, budget with finance and the volume of work with nobody.
Capacity Is Planned on Averages
An average month conceals the week that breaks the operation. Planning to the mean guarantees failure at the peak.
A Forecast Without a Decision Is a Dashboard
If no decision changes because of the projection, it is observation with a confidence interval.
Forecasts Are Produced at the Wrong Level
Enterprise totals can be accurate and still useless if the actual staffing decision sits at queue, specialty or team level.
Forecast Error Is Never Decomposed
Without separating model miss from genuine business change, neither the model nor the business understanding improves.
Put your known calendar events against your staffing plan.
Take the next twelve months and place open enrollment, renewals, mailings, group go-lives, regulatory deadlines and known seasonal patterns against planned capacity. The gaps are often visible in one afternoon.
A forecast earns its place only if a decision depends on it. Cost, reserving, pricing, completion and risk revenue remain with actuarial.
Step 1
Identify the decision first.
Who decides what, when, and what would change with a better forecast?
Step 2
Establish the decision horizon.
A six-week staffing decision needs a six-week forecast, not an annual one.
Step 3
Confirm the boundary with actuarial.
Cost, reserving, pricing, completion and risk revenue stay with them.
Step 4
Assemble the drivers.
Include known calendar events, not only historical transactions.
Step 5
Build the simplest model that supports the decision.
Understandable methods are more likely to be challenged and used.
Step 6
Establish the simple baseline first.
Complexity has to prove value against the method the operation already uses
Step 7
Validate out of time.
Test the model on periods it never saw during development.
Step 8
Express uncertainty in decision terms.
Use a range and probability rather than a point estimate treated as a commitment.
Step 9
Build the scenario capability the decision needs.
Let leadership ask what-if rather than receive one number.
Step 10
Monitor accuracy and decompose error.
Separate model revision from real business change and retire models that stop earning their place.
If it cannot beat last year plus a percentage, it has not earned its keep.
Every forecast should be tested against the naive benchmark the operation currently uses before deployment and periodically afterwards.
Everything here serves an operational decision with a short horizon and a named owner. Nothing here trespasses on cost, reserving or pricing.
Forecast the Demand
Transaction and Work Volume Forecasting
Claims, enrollment transactions, authorizations, appeals, documents and contacts forecast at the horizon the staffing decision requires.
Calendar Event Modelling
Model the operational effect of open enrollment, renewals, benefit changes, mailings, go-lives and regulatory deadlines.
Contact and Channel Forecasting
Forecast calls, messages and portal demand by driver so known demand is planned rather than absorbed.
Seasonality and Pattern Decomposition
Separate trend, seasonality, calendar effects and one-offs so movement is understood rather than extrapolated.
Plan the Capacity
Capacity, Staffing and Peak Planning
Translate forecast demand into required capacity by skill and function, planned to the distribution rather than the mean.
Scenario and Sensitivity Analysis
Test what happens under a larger group win, a benefit change, a volume shift or a capacity loss.
Backlog and Queue Projection
Show where forecast demand exceeds capacity, when backlog forms and when it clears.
Govern and Sustain
Validation Against Naive Benchmarks
Every model compared with the simple method it replaces before deployment and periodically afterwards.
Uncertainty Communication
Ranges, probabilities and assumptions presented so decision makers can weigh them rather than treat a point forecast as a commitment.
Structural Break Detection
Identify when the operating environment has moved beyond the conditions the model was built for.
Forecast Accuracy Monitoring
Track performance by model and horizon, with error decomposed into model revision and real business change.
Model Lifecycle and Retirement
Define ownership, review cadence and a retirement path so models do not quietly degrade and remain in use.
What CaliberFocus does, and does not do?
We do not do actuarial work. Cost trend, reserving, completion, pricing and risk revenue projection sit with your actuaries. We forecast operational demand and capacity, work alongside actuarial and tell you when an existing model is not beating the simple method it replaced.
The ownership boundary is explicit. Where the table says actuarial, CaliberFocus supports the data and stays out of the projection.
| Domain | What Is Being Forecast | Who Should Own It |
|---|---|---|
| Claims Volume | How many claims arrive, by type and channel, in the next weeks | Operations. Drives staffing and is rarely forecast at all. |
| Claims Cost and Trend | What that care will cost | Actuarial. We supply governed inputs and stay out of the projection. |
| Reserving and Completion | Incurred but not reported, run-out development | Actuarial, entirely. A regulated professional judgement. |
| Contact Volume | Calls, messages and portal sessions by driver and week | Operations. Highly forecastable and usually planned on last year. |
| Authorization Volume | Requests by service type and specialty | Utilization Management. Determines clinical reviewer capacity. |
| Enrollment Transactions | Volume by channel and group, including renewal peaks | Enrollment Operations, concentrated in known windows. |
| Membership Projection | How many members, by product and market | Finance and Actuarial jointly. We support rather than produce. |
| Provider Network Capacity | Whether supply meets projected demand by specialty and market | Network. A genuine gap in most plans and squarely in scope. |
Do not forecast at a level where nobody can act.
The useful level is the line of business, market, product, queue, specialty or team where a named person makes a decision. Error should be measured at that level too.
Most forecasting effort in payer organizations can be screened before any modeling work begins.
Decision
What decision changes because of this forecast, and who makes it?
Horizon
Does the forecast arrive early enough to influence that decision?
Benchmark
Does it beat the simple method the operation currently uses?
Actionability
Can the organization actually respond to what the forecast says?
Prefer the Model People Will Use
A simpler method people understand, challenge and adjust beats a technically better one they treat as a black box.
Model the Calendar Explicitly
Known events should be model inputs rather than residual surprises.
Forecast the Distribution, Not the Point
Capacity decisions depend on the plausible range, not only the midpoint.
Decompose the Error Every Time
Separate model revision, calendar effect, data change and genuine business movement.
Validate on Held-Out History and Forward
Backtesting shows whether the method worked; forward performance shows whether it still does.
Know When the Model Is Outside Its Experience
Benefit, network or regulatory changes can create structural breaks that invalidate historical relationships.
Separate Signal From Noise
Not every month-to-month movement deserves operational response.
Retire Models That Stop Earning Their Place
Unused models or models that no longer beat the benchmark should be stopped.
Forecasting demand you cannot respond to is not planning.
If the operation cannot flex capacity within the forecast horizon, the useful work may be on the constraint rather than the forecast: flexible capacity, cross-training or moving the decision earlier.
Operational forecasting needs transactional history and a structured record of the things the organization is about to do to itself.
Transactional History
Claims, adjustments, payment detail, edits and pend reasons, ingested with change capture so adjustment history survives rather than being overwritten.
The Operational Calendar
Mailings, benefit changes, implementations, renewals, regulatory deadlines and campaigns captured as dated events.
Membership With Effective Dating
Exposure matters, and retroactive change affects the history the model learns from.
Workforce and Capacity Data
Staffing, skills, schedules and productivity so demand can be linked to the decision it is meant to inform.
Governed Data Products
Use certified definitions so forecasting does not become another version of the truth.
Finance and Actuarial Outputs
Consume relevant outputs as inputs while respecting their ownership instead of reproducing projections independently.
Integration principles
Capture the calendar as data. Match the grain to the decision. Preserve what was known and when. Keep training data reproducible. Do not reproduce actuarial projections.
A forecast that drives staffing and service decisions may carry less regulatory weight than an actuarial projection, but it still needs documentation, validation, monitoring and ownership.
Model governance
- Every model documented with its purpose, decision, horizon, inputs, method, limitations, owner and review date
- Intended use recorded and enforced, since an operational demand model becoming an input to a financial projection is a boundary crossing nobody noticed
- The boundary with actuarial documented explicitly, so nobody later mistakes an operational forecast for a cost projection
- A named business owner accountable for whether the forecast is used and whether it is right, distinct from whoever maintains it
Validation
- Backtested on held-out history and compared to the naive benchmark before deployment, with both results retained
- The boundary between clinical decision support and regulated medical device software, assessed per use case
- Forward accuracy tracked by model, horizon and segment, including persistent bias toward over or under-forecasting and whether stated prediction intervals actually contain the outcome at the claimed rate
- Error decomposed into model revision, calendar effect, data change and business movement rather than reported as one figure
- Assumptions stated and reviewed, since an assumption that quietly stopped holding is the most common cause of silent degradation
Explainability
- The drivers of a forecast visible to the person making the decision, because an unexplained number is either overridden or followed blindly
- Uncertainty expressed as a range with a probability rather than as a point estimate that will be quoted as a commitment
- Forecasts distinguished from actuals in every artifact, and scenarios labelled as scenarios, since both become indistinguishable in a slide within one forwarding
Operational control
- Version control on method, inputs and assumptions, since a method change moves the forecast without the business having changed
- Monitoring for input data changes, because a source that shifted silently degrades the model before anybody sees the accuracy drop
- An override is data. Where a business team repeatedly adjusts a forecast for the same reason, capture the original, the adjustment and the rationale. Recurring correction is the clearest available evidence of a missing variable, a structural change or a business rule the model does not represent
Check whether anyone is still using your existing forecasts.
For each model, identify which decision it informs, who looks at it, when they last changed a decision because of it, and whether it beats the naive benchmark. Models that fail those tests are maintenance cost.
Accuracy matters, but an accurate forecast nobody acts on has delivered nothing. A slightly less accurate forecast that changes a staffing decision six weeks early can deliver far more.
| Category | What We Measure | Why It Matters |
|---|---|---|
| Decisions Influenced | Decisions changed because of a forecast, by model and owner | Establishes whether the capability is operational rather than analytical. |
| Benchmark Improvement | Accuracy against the naive method previously used | Shows whether the model earns its maintenance. |
| Forecast Accuracy | Error by model and horizon, decomposed into causes | Tells you what to fix rather than only how wrong you were. |
| Peak Preparedness | Known events with capacity planned in advance and service maintained | The practical outcome visible to members and providers. |
| Response Capability | Share of forecast demand the operation can actually flex to meet | Identifies where the constraint is capacity rather than information. |
| Model Estate Health | Models in production that are used, beat the benchmark and have an owner | Prevents accumulation of forecasts nobody reads. |
Expect the assessment to narrow the programme.
Some existing models may not beat the naive benchmark. In another area, the real constraint may be capacity rather than forecasting. Both findings are useful because they prevent investment in models that would not change the operation.
We will take one operational decision, establish its horizon and its owner, build the simplest forecast that supports it, and validate it against the method you use today. We will also review your existing models against the four tests, which usually identifies a few that should be retired. None of this touches cost, reserving or pricing, which stay with your actuaries.
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
