Predict Marketplace Performance¶

Capability Slug: predict_performance Plan Tier: Command Credit Cost: 8 base + 5 (recommendations, default on) ≈ 13 typical Category: Intelligence & Attribution Default mode: prediction_type: 'full' Backend: Prediction Engine (PredictionEngineService::predictListingPerformance())

Overview¶

ML-powered prediction for AWS Marketplace listing performance. Given a listing version (or raw listing content), the Prediction Engine returns CTR, conversion rate, time-on-page, monthly revenue estimates, and benchmark comparisons against the listing category. Use it before publishing to forecast the performance lift from a listing change.

Shared handler with two sibling capabilities¶

PredictPerformanceCapability is the underlying handler for three registry slugs that dispatch by prediction_type. See Predict Co-Sell Partner Success for the full shared-handler explanation.

Slug	`prediction_type`	What it predicts
`predict_performance` (this page)	`full` (or `listing`)	Listing CTR / conversion / time-on-page / revenue
`assess_launch_readiness`	`readiness`	Launch readiness score against blocking issues
`predict_partner_success`	`partner`	Partnership success probability
(undocumented)	`optimization`	Field-level optimization suggestions for a listing

What you get back¶

{
  "success": true,
  "prediction_type": "full",
  "credits_used": 13,
  "prediction": {
    "metrics": {
      "ctr": {
        "predicted": 0.045,
        "range_low": 0.038,
        "range_high": 0.052,
        "confidence": 0.81
      },
      "conversion_rate": {
        "predicted": 0.024,
        "range_low": 0.019,
        "range_high": 0.029
      },
      "time_on_page": 142
    },
    "benchmark_comparison": {
      "category": "Application Performance Monitoring",
      "benchmark_ctr": 0.038,
      "performance_vs_benchmark": "+18%",
      "percentile": 72
    },
    "revenue_estimate": 12500,
    "model_version": "v2.4.1",
    "features_analyzed": 34
  },
  "interpretation": "...",
  "recommendations": [ /* if include_recommendations=true */ ]
}

model_version is surfaced so you can correlate predictions against model deployments. The benchmark comparison is category-aware — the Prediction Engine looks up the listing's category and compares against that category's CTR baseline.

Required and optional parameters¶

Parameter	Required	Description
`version_id`	One of these required	Saved `ListingVersion` to predict against
`content`	One of these required	Raw listing content (title, short_description, highlights, category, pricing_model)
`prediction_type`	Implicit (`full`)	Defaults to `full` for this slug
`include_recommendations`	No	Default `true`. +5 credits.
`include_partner_prediction`	No	Default `false`. +10 credits if enabled — but if you want partner prediction, use `predict_partner_success` instead.

When neither version_id nor content is provided directly, the capability searches the agent run's dependency outputs for content or refined_content from upstream steps. This makes it composable after a content-generation chain — generate, refine, then predict performance on the result.

When to use it¶

Scenario	Recommended mode
Pre-publish performance forecast on a new listing version	`predict_performance` (this)
Score the same listing against publish-readiness gates	`assess_launch_readiness`
Field-level rewrite suggestions	`predict_performance` with `prediction_type: 'optimization'`
Forecast partnership outcomes	`predict_partner_success`

API access¶

Exposed as predict_performance. Partner API keys scoped with view_listings and manage_listings plus the agent capability slug can run predictions programmatically. See Partner API Capabilities.