Predict Co-Sell Partner Success¶

Capability Slug: predict_partner_success Plan Tier: Command Credit Cost: 8 base + 5 (recommendations, default on) + 10 (partner prediction) ≈ 23 credits typical Category: Co-Sell Default mode: prediction_type: 'partner' Backend: Prediction Engine (PredictionEngineService::predictPartnerMatchSuccess())

Overview¶

You're considering investing in a co-sell partnership. You can read partnership decks all day; what you can't easily do is quantify the probability of success. Predict Co-Sell Partner Success runs the Prediction Engine's ML scoring against your company record + the partner record and returns a structured numerical read — match score, success probability with confidence, strength indicators, risk factors, recommended actions, and an estimated deal value.

This is the quantitative complement to Partner Relationship Intelligence. That capability gives you AI synthesis of partner data; this one gives you ML-backed predictions. Most partner-investment decisions benefit from both.

Shared handler — same code, three slugs¶

PredictPerformanceCapability is the underlying handler for three registry slugs that dispatch by prediction_type:

Slug	`prediction_type`	What it predicts
`predict_partner_success`	`partner`	Co-sell partnership success probability (this page)
`predict_performance`	`full` (or `listing`)	Marketplace listing CTR / conversion / revenue (documentation in progress)
`assess_launch_readiness`	`readiness`	Launch readiness score against blocking issues (documentation in progress)
(undocumented)	`optimization`	Field-level optimization suggestions for a listing

Same handler, same include_recommendations / include_partner_prediction flags, same credit math. The registry exposes them as separate capabilities so partners can scope API keys precisely.

What you get back¶

{
  "success": true,
  "prediction_type": "partner",
  "credits_used": 23,
  "prediction": {
    "match_score": 78,
    "success_probability": 0.62,
    "confidence": 0.81,
    "strength_indicators": ["...", "..."]
  },
  "risk_factors": ["...", "..."],
  "recommended_actions": ["...", "..."],
  "estimated_deal_value": 250000,
  "interpretation": "...",
  "recommendations": [ /* if include_recommendations=true */ ]
}

Field-by-field:

Field	Type	Meaning
`match_score`	0-100	Partner-fit score across product, market, ICP, engagement
`success_probability`	0-1	ML-predicted probability that the partnership produces qualified opportunities
`confidence`	0-1	How much weight the model puts on its own prediction (driven by data sufficiency)
`strength_indicators`	array	Factors driving the score upward
`risk_factors`	array	Factors pulling the score downward
`recommended_actions`	array	Concrete next-step suggestions tied to the indicators and risks
`estimated_deal_value`	dollars	Directional sizing of the partnership opportunity
`interpretation`	string	One-paragraph natural-language read of the prediction

How to read the numbers¶

The three numbers tell you different things and pair together for decision support:

If you see...	It means...
High `match_score` + high `success_probability` + high `confidence`	Strong investment signal. Move forward.
High `match_score` + low `confidence`	Promising fit but not enough data yet. Run a small pilot, gather signal, re-predict.
Moderate score + tight confidence	The model is sure the partnership is mid-tier. Often a "yes, but not a priority" signal.
Low `success_probability` + high `confidence`	The model is sure this won't work. Read `risk_factors` before deciding to override.
Any prediction with very low `confidence`	Data sufficiency problem. The model can score but isn't sure. Treat the prediction as illustrative.

The credit math (read this)¶

The registry shows predict_partner_success at 15 credits — but the actual cost is computed by getEstimatedCredits() with multipliers:

total = 8 (base)
      + 5 if include_recommendations is true  (default ON)
      + 10 if include_partner_prediction is true

For a typical partner-success run with default recommendations and partner prediction enabled: ~23 credits, not 15. Disable recommendations to save 5; disable partner prediction (relevant for the other prediction modes) to save 10.

This is the same hidden-credit-cost pattern that shows up in agent_tool_card, cppo_proposal_generator, and several other capabilities. We're documenting it openly across feature pages.

Required and optional parameters¶

Parameter	Required	Description
`partner_id`	Yes	The partner record to score against
`company_id`	Required if not in context	Your company record — pulled from context if available
`prediction_type`	Implicit (`partner`)	Required to be `'partner'` for this slug. The shared handler dispatches by this value.
`include_recommendations`	No	Default `true`. Adds AI-generated next-step recommendations. +5 credits.
`include_partner_prediction`	No	Set to `true` to enable partner-success scoring. +10 credits. Required to true for this capability.

When to use it (vs. Partner Relationship Intelligence)¶

Question	Use this capability	Use Partner Intelligence
"Should we invest in this partnership?"	✅ — quantitative read with confidence
"What's the dollar opportunity here?"	✅ — `estimated_deal_value`
"Brief me before my call with the partner"		✅ — `sales_brief` mode
"What deals could we run together?"		✅ — `co_sell_opportunities` mode
"Is the partnership healthy?"		✅ — `partner_engagement` mode
"Compare three potential partners"	✅ — run for each, compare scores

It is not:

A committed forecast. The estimated_deal_value is directional. Use it for sizing, not for revenue commits.
A substitute for human judgment. The risk_factors are hints, not blockers — read them, then decide.
A replacement for partner discovery. Use Co-Sell Partner Discovery & Matching (cosell_partner_matching) to find candidates; use this capability to evaluate one in depth.

API access¶

Exposed through the agent runtime as predict_partner_success. Partner API keys scoped with view_partners plus the agent capability slug can run predictions programmatically. The shared handler dispatches based on the prediction_type parameter; other prediction-engine modes are gated by the predict_performance and assess_launch_readiness capability slugs respectively. See Partner API Capabilities.