Agentic Pricing in B2B SaaS

Ch 1. Market Reality

The agentic pricing market has not settled

Five concurrent pricing patterns dominate the 22-vendor inventory: pure outcome-based (Intercom Fin, Sierra), hybrid platform-fee plus meter (Decagon, Writer, Glean, ServiceNow, Microsoft Copilot), pure per-seat with absorbed inference (Harvey, Hebbia, Notion AI, Perplexity Enterprise), per-case or per-interaction flat (EvenUp, Cresta), and pure token-rate pass-through (Cursor, Cognition Devin, GitHub Copilot after June 2026). Salesforce alone runs five SKUs simultaneously for Agentforce. No single pattern has won.

The 14 named published voices on agentic pricing converge on six consensus claims:

Where AI gross margin compression lives

The hyperscaler trap, with a scope clarification

GitHub Copilot is the most credible cost-coverage signal in the inventory. Owned by Microsoft, backstopped by Azure. Two repricings in twelve months. If Microsoft's vertically-integrated economics cannot maintain flat per-seat for an agentic product at the application layer, most mid-market vendors will not.

Two named scaling exceptions show where the thesis does not generalize. Perplexity Enterprise Pro runs flat per-seat ($40-$325/user/month) for an agentic research product and has not repriced through May 2026; its agentic surface is structurally bounded. Notion AI's core seat ships at $20/user/month with unlimited core features for 2+ years and counting; Custom Agents moved to a credit model in May 2026, the core seat did not.

Five vendor archetypes in market today

Ch 2. The Value-Metric Decision Framework

Pick the value metric first, then design the contract

The most common pricing mistake at PE-backed B2B SaaS portcos is to copy the headline metric of a successful vendor without copying the structural cost-coverage architecture underneath it. Fin's $0.99 per resolution works because Intercom has Atlassian and Shopify-class scale spreading fixed costs. A $50M ARR portco copying $0.99 with a similar agent stack faces the Support Agent Volume Trap the moment query mix shifts.

The decision tree

Dual-vocabulary translation table

Three universal-claim exceptions, named

• Outcome pricing without a quality SLA fails in archetypes where outcome quality varies materially. Named exception: narrow-vertical deterministic-outcome categories (Replit Agent code-completion).
• Flat per-seat fails for application-layer vendors with unbounded agentic surfaces. Named exceptions: bounded-agent products (Perplexity Enterprise, Notion AI core) and vertical labor-anchored products (Harvey, Hebbia).
• Budget Hard Stops are required on enterprise procurement-driven contracts at $250K+ ACV. Named exception: SMB/individual segments (Replit Agent ships without one and scales).

Ch 3. The Cost-Coverage Risk Model

The Cursor June 2025 case, re-read

On June 16, 2025, Cursor moved Pro from $20/month for 500 fast model requests to $20/month for $20 of model credits at API rates. The mechanism was a transfer of the entire AI cost-coverage risk from Cursor's balance sheet to the customer's credit card.

CEO Michael Truell apologized publicly on July 4, 2025: "We have to be more careful about how we communicate pricing changes... The previous model was financially unsustainable." Cursor refunded affected users. Trustpilot settled at 1.7/5 across 203 reviews. ARR grew from $100M (January 2025) to $2B (February 2026); the company entered funding talks at a $50B valuation. The pivot triggered persistent trust friction that did not stop one of the fastest B2B scaling events in history.

Cost taxonomy

Six top-level cost categories ranked by margin-risk significance: inference (HIGH volatility, EXTREME on reasoning tokens), agent architecture (MEDIUM-HIGH), compute infrastructure (MEDIUM), model variability (LOW frequency, HIGH magnitude on migrations), human fallback, infrastructure overhead. The categories tagged HIGH or EXTREME are where contract terms must be most aggressive.

The eight hidden cost amplifiers

Amplifiers compound, they do not add. A fully-stacked agent (loop × multi-model × evaluator × fan-out × reasoning) can produce 100x+ cost on the same nominal task. Rule of thumb: assume 5-8x amplifier on baseline inference at P75 traffic for unbounded agentic surfaces. For bounded surfaces, 2-3x.

Sensitivity scenarios (S1-S5)

Five scenarios stress-test six archetype-and-pricing combinations. S1 base; S2 inference cost +50%; S3 volume +200%; S4 agent loop overhead 3x; S5 combined adverse.

The hybrid and Writer-class architectures stay positive across all five scenarios. The pure-outcome and pure-seat-application-layer architectures go negative on a single bad quarter.

Five named margin-collapse scenarios

1. The Support Agent Volume Trap (S1+S3). Pure-outcome pricing loses 29 percentage points of margin during success-driven scaling. Remedy: outcome rate paired with a platform-fee floor OR a quality-weighted definition of "resolution."
2. The Developer Tool Loop Creep (S1+S4). 18 months of agentic feature additions without repricing compress margin from 18% to -110%. Cursor June 2025 is exactly this. Remedy: cost-class declaration at design review; surface the tail to customers before margin goes negative; grandfather existing customers.
3. The Vertical Reasoning Avalanche (S2+S4). A 50% upstream price increase on reasoning compounds with a 3x loop feature. Margin moves from 78% to 12%. Remedy: reasoning tokens as a separate meter; per-deliverable hard cap; pre-staged model-version pass-through clause.
4. The Combined Adverse "AI Winter Pivot" (S5). Provider repricing coincides with volume and complexity. Pure-seat application-layer developer tool reaches -280% margin. Remedy: pre-positioned model-version hedges (2+ providers, BYOK at enterprise, 25%-trigger pass-through); pre-staged customer caps; quarterly margin review.
5. The Quality Erosion Mirror (Klarna Deployment-Scope Failure). Vendor deploys AI beyond its quality envelope; outputs degrade; customer rehires humans. Remedy: contracted quality-weighted outcome metrics; customer-side telemetry; quality-engineering investment alongside the pricing clause.

Worked example: $100M ARR portco

The framework's $1.8M-$3.7M program investment and the multiple-protection claim apply to a specific representative profile: $100M ARR, 22% YoY growth, 55% GM baseline, 8x revenue multiple, 24-month transition, application-layer agentic surface.

Two-year program total: $1.8M-$3.7M. Protection band: 0.5-1.5x of the 8x baseline multiple, or $50M-$150M of EV on the representative profile. The methodology generalizes; the dollar magnitudes do not.

The three levers in one row

Ch 4. Customer Risk Mitigation Patterns

Three levers, top three patterns

The Predictability Architecture rests on three levers that together prevent every named margin-collapse scenario in the unbounded-agent application-layer regime. Every name in the Wave 1 inventory that has survived an agentic-pricing crisis without trust damage runs all three levers.

P4, P9+P15, P12

P4, Predictable Seat + Metered Power-User Usage (Writer Architecture). The broadly applicable pattern for application-layer agentic products. The only architecture in the Chapter 3 sensitivity matrix that stays positive across all five scenarios. Year 1 priority: retrofit every existing seat-priced contract that lacks a metered escape valve.

P9 + P15, Universal Riders for Enterprise Procurement. Budget Hard Stop + Model Version Stability, required on every multi-year metered contract in enterprise procurement-driven sales motions at $250K+ ACV. P9 is the enterprise-procurement Cursor-prevention clause; P15 is the universal "AI Winter Pivot" defense.

P12, Outcome Guarantee with Quality Assurance, scoped to $1M+ ACV. Highest aggregate moat score. Below $1M ACV, use the P12-Light variant (outcome metric + quality SLO + graduated credit remedies, without full audit infrastructure).

Five specimen contract clauses (excerpts)

Moat impact (Helmer 7 Powers)

Ch 5. Transition Playbook

Five phases over 24-30 months

Reversibility drops at each phase boundary. The cliff is between Phase 2 (80% reversible) and Phase 3 (50% reversible). Cross it deliberately.

The Phase 3 sales-comp precondition gate (new in v2)

Before Phase 3 launches, the portco must complete five preconditions. If any one is not completed, Phase 3 does not launch:

• AE comp impact modeling. Run the new plan against actual W-2 history of the top 20% of reps. If any top-quintile rep loses more than 15% of cash comp in the model, the plan needs adjustment.
• Retention conversations. 1:1s with the top 3 quota carriers (CRO-led) explicitly framing the comp change, the rationale, and the AE's place in the post-transition org.
• Hire-back / replacement runway. Recruiting pipeline filled to 1.5x of expected top-decile attrition (15-25% range). Signing bonuses pre-approved at the CEO level.
• Competitor comp benchmarking. If the new comp is more than 10% below competitor comp at the top-quintile AE level, the plan creates an active flight incentive.
• Executive sponsor. A single CRO or CEO accountable for AE retention through Phase 3 by name, with explicit P&L for top-rep churn.

Stakeholder communications: six templates, one vocabulary rule

The vocabulary rule: "shared risk," "transition," "cost-coverage," "vendor margin" are operator-grade vocabulary used internally only. The customer hears predictable cost, hard ceilings, value floors, quality assurance. Mixing destroys the framing. The Cursor execution friction was partly a vocabulary failure.

Grandfathering: eight rules

• Honor every existing contract through its current term.
• Grandfather window of 12-24 months from new pricing launch (up to 36 months for flagship).
• Long-term contracts (3-5 year deals) honored in full; voluntary migration with credit incentive.
• Migration sequencing: smaller and less strategic first; flagship accounts in Wave 4 of Phase 4.
• Partial migration allowed (some BUs on new, some on legacy).
• Voluntary migration credits decay over the window (10% to 7.5% to 5% to 0%).
• P9 (Budget Hard Stop) added universally at next renewal for enterprise multi-year deals.
• P15 (Model Version Stability) retrofit offered free to every multi-year customer.

Ch 6. The JustKeenAI Advisory Engagement

Operating the framework internally

The framework is fully self-applicable. A portco with a capable CFO, CPO, CTO, and GC can execute the entire Predictability Architecture program from the chapters above. The pattern library is drop-in usable. The specimen contract clauses lift directly into MSAs and Order Forms. JustKeenAI does not gate the framework. The intellectual asset is the framework itself. The engagement is the optional accelerant for the conditions where it adds value an internal team cannot replicate.

Three engagement paths

Three trigger conditions that map to paths

• A provider repricing event looks imminent and the multi-year book is exposed. Maps to the P15 Contract Portfolio Audit (Path 2).
• The portco is 18-30 months from a planned exit and the diligence question is forming. Maps to Quarterly Board Narrative Review (Path 2) or the Full Engagement (Path 3).
• The portco's product is on the wrong terminal node of the Chapter 2 decision tree. Maps to the Full Engagement (Path 3), starting with the Phase 1 Diagnostic.

Two packaged products

P12 First-Mover Engagement ($1M+ ACV portcos only). A 6-9 month engagement to design, ship, and market the first operationalized P12 contract. Eligibility: at least one $1M+ ACV deal in active pipeline; quality-conscious buyer profile; capacity to invest in audit infrastructure ($400K-$880K Year 1). Typical scope: $250K-$500K.

P15 Contract Portfolio Audit. 60-90 day engagement to retrofit a multi-year contract book with Model Version Stability riders. Retrofit cost $250K-$1M legal + ops; GM-protection band 5-15 percentage points on the representative profile. Typical scope: $150K-$300K.

What JustKeenAI brings that internal teams cannot easily replicate

• External adversarial review. Internal teams will not consistently surface the failure modes they are personally invested in not seeing.
• Cross-portfolio benchmarking. Public data does not show where peer portcos sit on the decision tree.
• Vendor negotiation leverage. Multi-provider model commitments, BYOK pricing, capacity-deal terms.
• Stakeholder facilitation across the board, CFO, CCO, CS lead, and sales VP.
• Phase 3 reversibility-cliff discipline. The single highest-stakes call in the program.