> ## Documentation Index > Fetch the complete documentation index at: https://docs.wayak.io/llms.txt > Use this file to discover all available pages before exploring further. # Reserve estimation > Calculate initial claim reserves based on comparable historical claims using statistical analysis Setting accurate reserves is one of the most consequential tasks in claims management. Under-reserving creates surprise shortfalls that destabilize financial reporting. Over-reserving ties up capital unnecessarily. Most initial reserves are set by adjusters using experience and rough rules of thumb, which introduces inconsistency — two adjusters often set very different reserves for similar claims. This use case connects your claims history database to a Wayak agent that estimates reserves by analyzing comparable historical claims. A playbook uses Python to run statistical calculations — pulling median payouts, adjusting for claim-specific factors, and generating confidence ranges — so every new claim gets a data-driven reserve estimate within minutes of classification. *** ## What you need * **Claims management system** — Historical claims with final payout amounts, claim type, severity, region, and resolution details * **Policy administration system** — Policy limits and deductible amounts for reserve cap calculations * **General ledger** — Current reserve postings for reconciliation * **Reserving guidelines** — Upload your actuarial reserving methodology, including adjustment factors and minimum reserve rules * **Claims benchmarking data** — Industry benchmark payout data by claim type, if available **Semantic layer:** Define these in your ontology before setting up the agent. | Component | Name | Definition | | --------- | ----------------- | ---------------------------------------------------------------------------------------------------------------- | | Object | Claim | Maps to the claims management system. Represents a claim with type, severity, region, and current reserve amount | | Object | Historical Payout | Maps to closed claims with final payout data. Used as the basis for comparable analysis | | Metric | Estimated Reserve | Statistical estimate of expected payout based on comparable historical claims | | Metric | Average Payout | Mean final payout for closed claims matching a given type, severity, and region | | Dimension | Severity | Classifies claims by loss magnitude: minor, moderate, significant, severe, catastrophic | | Dimension | Region | Geographic area where the loss occurred, which affects cost benchmarks | See [building a semantic layer](/quickstart/build-semantic-layer) for a step-by-step guide. *** ## Agent setup Go to **Agent Space** > **New agent**. | Field | Value | | -------- | ------------------------------------------------------------------------------------ | | **Name** | Reserve Estimation Agent | | **Role** | Claims reserve analyst | | **Goal** | Provide data-driven reserve estimates for new claims based on historical comparables | > You estimate claim reserves using historical data. When asked to estimate a reserve, always start by identifying the claim type, severity, and region, then pull comparable closed claims. Present the reserve as a range (low, expected, high) with the number of comparable claims used. Explain which adjustment factors you applied and why. Never present a single point estimate without a confidence range. If fewer than 10 comparable claims are available, flag this as low confidence and recommend manual review. Use precise dollar amounts and always note the policy limit as a cap. Grant access to: * Claims management system data source (current and historical claims) * Policy administration system data source (policy limits and deductibles) * Reserving guidelines knowledge space * Claim and Historical Payout objects in the semantic layer **Trigger:** User requests a reserve estimate or a new claim is classified 1. Identify the claim type, severity, and region from the intake summary or user request. 2. Pull historical closed claims matching the same type, severity, and region from the last 3 years. 3. Calculate the median and mean final payout for the comparable set. 4. Adjust for claim-specific factors: documentation quality, liability clarity, injury severity, and claimant history. 5. Apply the adjustment factors from the reserving guidelines knowledge space. 6. Cap the estimate at the policy limit minus the deductible. 7. Present the reserve estimate as a range: low (25th percentile), expected (median), and high (75th percentile) with the number of comparable claims used. **Trigger:** User asks how accurate past reserves were or wants to validate an estimate 1. Pull closed claims from the specified period with their initial reserve and final payout. 2. Calculate the reserve accuracy ratio: final payout divided by initial reserve. 3. Group by claim type and severity to show where reserves tend to be most and least accurate. 4. Identify systematic biases (consistent over- or under-reserving) by category. 5. Present a summary table with claim type, average initial reserve, average final payout, accuracy ratio, and bias direction. *** ## Automation ### Playbook: Reserve estimate on new claim Set the playbook to trigger when a new claim is classified and assigned in the claims management system (after the intake pipeline completes). The playbook pulls comparable claims, runs a statistical reserve calculation, and posts the estimate to the claim record. 1. **Query step** — Pull the new claim's type, severity, region, and policy details. 2. **Query step** — Pull historical closed claims matching the same type, severity, and region from the last 3 years. 3. **Python code step** — Calculate the reserve estimate: compute the median, 25th percentile, and 75th percentile of historical payouts. Apply adjustment factors for documentation quality and liability clarity. Cap at the policy limit. 4. **Condition step** — If fewer than 10 comparable claims were found, flag the estimate as low confidence and add a note for manual review. 5. **Action step** — Post the reserve estimate (low, expected, high) to the claim record in the claims management system. The **Python code step** uses a Python code block to compute percentile-based reserve estimates and apply adjustment multipliers from your reserving guidelines. You can customize the adjustment factors, the comparable claim window (default 3 years), and the confidence threshold (default 10 comparable claims). Post the reserve estimate directly to the claim record. Send a notification to the assigned adjuster with the estimate range and confidence level. For low-confidence estimates, also notify the reserving supervisor. Click **Run now** to test with live data, then toggle to **Active**. *** ## What's next Score claims against known fraud indicators and flag suspicious submissions. See the full list.