TOOL-011 — Pricing Sensitivity Analyzer

Tier 3 Specialist Tool · Stateless · Reads quote acceptance/rejection patterns + competitive context to classify deal-level pricing elasticity · Augments AGT-406 deal desk decisions

Tier 3 · Tool Specced · v32 Domain 1 / Domain 2 Sonnet

Purpose

For an active deal at the quote stage, classifies how pricing-sensitive the prospect is likely to be based on (a) cohort-level patterns from QuoteLog history (similar deals, similar competitive context, similar segment), (b) deal-specific signals from ConvIntelligence (objection patterns around price, budget mentions, procurement involvement), and (c) any explicit pricing pushback already observed. Output is a sensitivity classification + suggested concession path + risk-flagged "do not concede here" guidance for the deal desk reviewer.

Augments AGT-406 CPQ & Deal Desk's existing 4-tier approval flow. AGT-406 owns the approval gate; TOOL-011 informs the human reviewer with structured pricing-sensitivity context. This is the "should we discount, and how much" decision support layer — not the approval mechanism.

Calibration risk. Pricing decisions have outsized financial impact and are highly contextual. The tool is informational input, not a recommendation engine for actual discount amounts. Hard rule: tool never recommends specific discount percentages — only classifies sensitivity and flags risk patterns. Discount magnitude remains a human decision per AGT-406 spec.

Input schema

{
  "opportunity_id": "uuid",
  "current_quote": {
    "quote_id": "uuid",
    "list_price_total_usd": 0,
    "current_proposed_price_total_usd": 0,
    "current_discount_pct": 0.0,
    "skus_in_quote": [
      { "sku_id": "string", "list_unit_price": 0, "proposed_unit_price": 0, "quantity": 0 }
    ]
  },
  "deal_context": {
    "current_stage": "string",
    "deal_health_score": 0,
    "icp_tier": "T1" | "T2" | "T3",
    "vertical": "string",
    "estimated_acv_usd": 0,
    "competitor_detected": "string | null",
    "champion_qualified": true | false,
    "economic_buyer_identified": true | false
  },
  "conversation_signals": {                // from AGT-407 trailing 60d for this deal
    "price_objections_raised": [
      { "objection_text_summary": "string", "raised_at": "ISO 8601",
        "addressed": true | false }
    ],
    "budget_mentions": [
      { "mention_text_summary": "string", "mention_at": "ISO 8601",
        "stated_budget_range_usd": "string | null" }
    ],
    "procurement_involvement_detected": true | false,
    "procurement_engagement_stage": "not_yet" | "introduced" |
                                    "actively_negotiating" | "blocking"
  },
  "cohort_history": {                      // from QuoteLog
    "cohort_definition": "string",         // similar segment/vertical/ACV band
    "cohort_size": 0,
    "cohort_avg_final_discount_pct": 0.0,
    "cohort_p25_final_discount_pct": 0.0,
    "cohort_p75_final_discount_pct": 0.0,
    "cohort_avg_revisions_to_close": 0.0,
    "cohort_acceptance_rate_at_initial_quote": 0.0
  },
  "previous_quote_revisions_in_deal": [    // if any prior versions
    { "version": 0, "discount_pct": 0.0, "outcome": "rejected" | "negotiated" }
  ]
}

Output schema

{
  "tool_call_id": "uuid",
  "sensitivity_classification": "low_sensitivity" | "moderate_sensitivity" |
                                "high_sensitivity" | "highly_constrained_budget",
  "classification_confidence": "high" | "medium" | "low",
  "key_drivers": [
    {
      "driver": "string",
      "evidence": "string",
      "input_field": "string",             // citation
      "directional_impact": "increases_sensitivity" | "decreases_sensitivity"
    }
  ],
  "cohort_comparison": {
    "current_discount_vs_cohort": "below_p25" | "p25_to_p50" |
                                   "p50_to_p75" | "above_p75",
    "interpretation": "string"
  },
  "concession_path_observations": [
    {
      "observation": "string",
      "rationale": "string"
    }
  ],
  "do_not_concede_flags": [
    {
      "flag_type": "weak_qualifier" |
                   "discount_above_authority" |
                   "competitive_response_not_warranted" |
                   "contract_term_concession_higher_leverage" |
                   "champion_not_confirmed",
      "explanation": "string"
    }
  ],
  "deal_desk_review_recommendation": {
    "review_priority": "standard" | "elevated" | "executive",
    "reviewer_focus_areas": ["string"]
  },
  "ungrounded_assumptions": ["string"],
  "data_completeness": "high" | "medium" | "low",
  "tool_metadata": {
    "model": "claude-sonnet-4-6",
    "input_tokens": 0, "output_tokens": 0,
    "cost_usd_estimate": 0.0,
    "latency_ms": 0
  }
}

Hard rule: Output never contains a specific recommended discount percentage or unit price. Tool produces sensitivity classification + cohort context + observation/flag patterns; human deal desk reviewer determines actual discount magnitude per AGT-406 approval flow.

Sensitivity taxonomy

Classification	Indicators	Interpretation
`low_sensitivity`	Champion qualified + economic buyer engaged + minimal price objections + no procurement blocking + cohort accepts at list-or-near	Concession unnecessary. List or near-list pricing likely to close. Discount here erodes margin without advancing close probability.
`moderate_sensitivity`	Some price objections (addressed) + procurement introduced but not blocking + cohort typical discount in p25–p75 range	Standard concession path expected. Cohort-typical discount range is the reasonable negotiation envelope.
`high_sensitivity`	Multiple unaddressed price objections + active procurement negotiation + competitor detected + cohort discount typically above p75	Significant negotiation likely. Multiple revisions likely. Consider non-price levers (term length, payment terms, ramp) before deeper price concessions.
`highly_constrained_budget`	Explicit stated budget < current proposed price + procurement blocking + objections framing as "can't afford" not "want better deal"	Different negotiation altogether. Either reduce scope (remove SKUs, smaller term) or qualify out. Discounting to meet a constrained budget may set a precedent that hurts later expansion.

Called by

Caller	Invocation context
AGT-406 CPQ & Deal Desk	On every quote generation requiring approval. Output included in the deal-desk review packet alongside existing 4-tier approval routing data. Reviewer reads sensitivity classification + flags before approving discount magnitude.
AGT-401 Deal Health Monitor	For deals at proposal/negotiation stage with discount under consideration. Output augments deal health context for AGT-902 narrative queries.
AGT-902 Account Brain	For "should we hold the line on this deal?" queries. Brain integrates classification into BrainAnalysisLog narrative.
RevOps direct (workspace UI)	For ad-hoc quote analysis and pricing strategy work. RevOps drops in deal context, gets structured sensitivity read.

Design principles

Decision support, not decision authority. Tool informs the human deal desk reviewer with structured context. Approval and discount magnitude remain in AGT-406's existing 4-tier approval flow, with humans deciding.
Cohort comparison anchors interpretation. Absolute discount levels mean little; relative-to-cohort matters. The cohort_history input provides the anchor. Tool degrades gracefully when cohort data is sparse (small segment, new vertical).
Surface anti-patterns explicitly. The do_not_concede_flags are at least as valuable as the sensitivity classification. "Don't discount here because the champion isn't qualified" is more actionable than "this looks high-sensitivity."
Contract-term concessions before price concessions. Tool's observations may suggest non-price levers (term length, payment terms, ramp schedule, multi-year commit) when those have higher leverage than equivalent price concession. Contract terms preserve list-price baseline for renewals.
Avoid race-to-the-bottom dynamics. Aggregate use of the tool across deal desk should not produce monotonically rising discounts over time. Quarterly retrospective compares pre-tool and post-tool cohort discount distributions; sustained drift is a calibration concern.

Cost ceiling

Constraint	Value
Per-call input budget	20K tokens (deal context + cohort history + conversation signals)
Per-call output budget	2.5K tokens
Default model	Sonnet (synthesis-heavy, multi-signal reasoning)
Per-call cost	~$0.10
Monthly cap	$300/mo (~3,000 calls)
Frequency	Moderate — one call per quote generation requiring deal-desk approval; deals with multiple revisions get multiple calls.

Eval criteria

Criterion	Pass threshold
Schema compliance	100% (hard)
No discount magnitude in output	0 instances of specific discount % recommendations (hard, regex-enforced)
Sensitivity classification accuracy	20 retrospective deals with known final discount and acceptance pattern; ≥ 70% classification match.
Anti-pattern flag precision	For 10 retrospective deals where weak champion qualification preceded discount-and-loss outcome: ≥ 60% had `weak_qualifier` flag set in TOOL-011 output at quote time.
Cohort-baseline degradation	For 5 cases with sparse cohort data: tool returns confidence = medium or low; does not produce inflated cohort comparisons. 100% (hard).
Aggregate discount drift monitoring	Pre-tool vs post-tool cohort discount distribution: median discount should not drift > 2pp. Quarterly retrospective; sustained drift triggers calibration sprint.
P95 latency	≤ 4s