预测退货与保修索赔趋势
按销售批次和滞后分布预测未来索赔量,拆分已售与未来销售来源,并提示异常质量风险。
完整提示词
<context>
Returns and warranty claims are an echo of sales: each month's sales generate claims spread over the following weeks or years. Forecasting claims directly from the claims time series misses this and fails at turning points (a sales peak at Christmas produces a returns peak in January; a warranty spike follows a product's sales by months or years). The standard approach treats each sale month as a cohort: estimate the claim rate per unit sold and the lag distribution (share of a cohort's claims arriving in month 1, 2, 3 after sale and so on), then forecast claims as the sum over cohorts of units sold times rate times lag share. Recent cohorts are immature, so their final claim rate is not yet visible and must be projected from older cohorts' development (the same idea as an insurance claims triangle). Rates differ by product and channel (online returns far exceed in-store), and a product quality issue shows up as a cohort whose early claims run above the usual curve, which is the early warning worth surfacing.
Use: {{use}}
</context>
<task>
<sales_and_claims>
{{sales_and_claims}}
</sales_and_claims>
{{#future_sales}}
<future_sales>
{{future_sales}}
</future_sales>
{{/future_sales}}
1. Build the cohort table (sale month by months since sale) from the data. If only claims by month are supplied, say the lag profile must be assumed and state the assumption.
2. Estimate the lag profile (share of eventual claims by month since sale) and the ultimate claim rate per unit sold, using mature cohorts; project immature cohorts forward using the profile.
3. Check for cohorts running above or below the usual curve and flag possible quality or policy changes.
4. Forecast claims by future month: from units already sold (known) plus from future sales (if supplied). Show the two parts separately.
5. Give a range from the variation in rates across cohorts.
6. Translate into the units the use requires (provision value, parts, staff hours) only when the needed conversion figures are supplied; otherwise list them.
7. Provide code or spreadsheet steps to rebuild the triangle monthly.
</task>
<constraints>
- Compute only from data supplied and show the cohort table.
- Do not treat immature cohorts' claims to date as their final rate.
- Accounting treatment of warranty provisions is for the finance owner; this forecast supplies the expected claims, not the accounting entry.
{{> output/uncertainty}}
</constraints>
<output_format>
## Summary
Expected claims for the horizon, split into already-sold and future-sales parts, with range and any cohort warnings.
## Lag profile
Table: Months since sale | Share of claims | Cumulative share.
## Claim rates
Table: Cohort | Units sold | Claims to date | Maturity | Projected ultimate rate.
## Forecast
Table: Month | From units sold | From future sales | Total | Low | High.
## Code
Code or formulas.
## Caveats
Bullets.
</output_format>可替换内容
- {{use}}
- {{sales_and_claims}}
- {{#future_sales}}
- {{future_sales}}
- {{/future_sales}}
- {{> output/uncertainty}}
分类与标签
商业营销、索赔预测、销售队列、售后分析、质量预警