How should Snowflake price Cortex agents — per query or per outcome?

Per-Query vs Per-Outcome: Pros & Cons

Per-Query Advantages:

1. Simplicity: usage equals cost; no outcome tracking burden on customer or Snowflake operations.
2. Predictability: CFOs can forecast AI spend via message-per-user multiplied by cost per message.
3. First-mover friendly: lowers land friction compared to outcome risk (what if agent *doesn't* deliver?).
4. Developer-native: engineers optimize toward reduced API calls; aligns with Snowflake's "pay for compute" DNA.
5. Rapid adoption: no outcome-audit requirement slows pilot-to-production transition.
6. Competitive equivalence: matches Databricks Mosaic AI and OpenAI's per-token model; expected by market.

Per-Query Disadvantages:

1. Value leakage: agent might prevent significant churn risk but Snowflake earns minimal compute fees; customer gets outsized ROI while Snowflake starves.
2. Wrong incentive: vendor optimizes for *chatter*, not *impact*; encourages bloatware agents and long feedback loops.
3. Commoditization: Cortex Agents drift toward commodity utility pricing; no defensibility versus open-source agents running self-hosted.
4. Ceiling effect: per-message math cannot scale with customer value; a large customer using Cortex Agents pays the same per message as a small startup.
5. Churn risk: when outcome visibility increases, customers may ask "Why am I not paying for ROI?" and shop alternatives like Salesforce Agentforce or Databricks.
6. Sales complexity: sales reps must explain why agent-driven significant churn savings cost the same as a test script.

Per-Outcome Advantages:

1. Defensibility: outcome-linked pricing is becoming table-stakes for major AI platforms.
2. Land-expand: pilot at low per-query floor; outcome tier triggers only on production success; risk transferred to customer (vendor earns if customer wins).
3. Sales narrative: "Your agent prevented churn → you pay a percentage of savings" is framing customers increasingly expect from major vendors.
4. LTV flywheel: high-outcome customers become upsell targets; per-query never scales LTV.
5. Analyst moat: agentic-outcome pricing becomes core to TAM models; Snowflake owns the narrative versus commodity per-token shops.
6. Vertical bundling: outcome model enables industry-specific SLAs (retail: "agent accuracy above threshold or we discount"; healthcare: "agent decision audit within acceptable range").

Per-Outcome Disadvantages:

1. Operationalization: requires outcome-audit framework and measurement partner. Snowflake carries SLA risk.
2. Customer friction: "How do we prove the agent caused the outcome?" pushes ML discipline; mid-market organizations lack regression maturity.
3. Sales cycle drag: outcome pilot → measurement framework → outcome audit can take months; per-query sells quickly.
4. Sandbagging: sophisticated customers may under-report outcomes or hide agent value in broader process improvements; Snowflake chases audit.
5. Cannibalization: mid-market customers who cannot measure outcomes may move to cheaper competitors; outcome-model pricing works best at scale.
6. Downside protection: what if agent *increases* churn (bad prompt, hallucination)? Snowflake cannot charge negative; customer may seek damages.

What Snowflake Should Consider (Roadmap)

1. Establish baseline floor pricing. Standardize per-query pricing across development, staging, and production tiers. Maintain a reasonable pricing envelope to avoid customer shock.

1. Develop outcome-measurement capabilities. Launch a Cortex Outcome Tracker: drop-in agent hooks that capture decision events, churn-risk flags, forecast deltas, and revenue impact. Partner with measurement vendors to wire tracking into customer success workflows.

1. Pilot hybrid pricing with select accounts. Test a hybrid model: (a) per-query floor, (b) outcome bonus based on documented savings or improvements, (c) exclusive pricing on new Cortex Agent models. Gather case studies.

1. Expand outcome pricing to broader tier. Roll outcome model to more accounts. Offer buyout options: upgrade to per-outcome with discount on query floor, or stay per-query at adjusted pricing to subsidize outcome operations.

1. Publish pricing transparency. Create documentation showing per-query versus per-outcome TCO by vertical, organization size, and agent swarm size. Position Snowflake as outcome-pricing pioneer.

1. Introduce outcome insurance and dynamic pricing. Offer outcome insurance tier: Snowflake guarantees minimum agent performance and captures upside above guarantee. Consider dynamic pricing based on model complexity, concurrency, and customer vertical.

1. Launch vertical outcome bundles. Create Cortex Agents for specific industries: Cortex For Retail (agent optimizes inventory, priced on stockout reduction), Cortex For SaaS (agent surfaces churn signals, priced on churn prevention), Cortex For Healthcare (agent flags clinical variation, priced on quality-outcome capture). Each bundle ships with pre-wired measurement.

1. Offer enterprise seat-based tier. For large customers, offer seat-based tier including all queries and all outcomes. This can undercut competitors' per-conversation models while capturing all value creation.

Pricing Model Roadmap

Outcome Architecture

Implementation Roadmap for Hybrid Pricing

Snowflake's transition to a hybrid model requires a phased rollout to minimize customer friction and build trust in outcome measurement. Phase 1: Launch per-query with optional outcome tracking via Cortex's built-in observability. Customers opt into outcome monitoring for a subset of agents to validate ROI without immediate financial commitment. Snowflake provides dashboards showing "value generated" estimates alongside query counts. Phase 2: Introduce outcome premium as a voluntary tier—customers who enable tracking pay a small base rate plus a capped percentage of documented savings. Use automated auditing via Cortex's native logging to prevent gaming. Phase 3: Make outcome premium default for production agents, with per-query reserved for development and testing. This gradual approach avoids sticker shock and lets Snowflake refine measurement frameworks based on real-world data.

Vertical-Specific Outcome Metrics

The "outcome unit" must vary by use case to align with customer value perception. For RevOps agents, price based on forecast accuracy improvement or incremental revenue attributed to agent recommendations. For customer support agents, tie pricing to resolution time saved or customer satisfaction score lift. For risk/fraud agents, charge per prevented churn event or per false-positive reduction. Snowflake should offer pre-built outcome templates for common verticals (retail, finance, healthcare) to reduce measurement complexity. Customers can also define custom outcomes using Cortex's SQL-native event logging, ensuring flexibility without vendor lock-in.

Competitive Positioning and Defensibility

Snowflake's hybrid model must differentiate from rivals like Databricks (per-compute) and Salesforce Agentforce (per-conversation floor). By anchoring to outcomes, Snowflake captures value creation rather than just consumption—a defensible moat as agents become autonomous. For example, a churn-prevention agent that saves significant revenue would generate meaningful outcome premium under Snowflake's model, versus much less under pure per-query at high usage. This aligns incentives: Snowflake profits when customers profit, reducing churn risk. To further lock in customers, bundle outcome tracking with Cortex's governance features (audit logs, compliance certifications) that competitors cannot easily replicate. The hybrid model also discourages "agent stuffing" since outcome premiums reward high-impact interactions.

The Outcome Measurement Challenge

A per-outcome model demands rigorous attribution—proving the agent caused the outcome, not just correlated with it. Snowflake must invest in native telemetry hooks: event logs that tie agent actions to downstream business metrics (e.g., "agent recommended discount → customer stayed → churn averted"). Without this, customers will dispute billing. A pragmatic middle ground: offer outcome pricing only for well-defined, high-confidence events like "forecast generated with <5% error margin" or "support ticket auto-resolved without escalation," where causality is clear. For ambiguous outcomes (e.g., "improved customer satisfaction"), stick to hybrid query pricing.

Vertical-Specific Outcome Units

Not all outcomes are equal. Snowflake should define tiered outcome units by vertical to avoid one-size-fits-all friction:

• Finance agents: Price per "forecast variance reduction" (e.g., under a threshold).
• Sales agents: Price per "qualified lead passed to CRM" or "deal stage advancement."
• Support agents: Price per "resolution time drop below baseline."
• Risk agents: Price per "fraud alert that prevented chargeback."

This granularity lets Snowflake capture value where it's highest while customers see direct ROI. A base query fee covers simple lookups; the outcome premium triggers only on verified business impact.

The Risk of Per-Outcome Underpricing

Outcome pricing risks Snowflake leaving money on the table if agents become indispensable but the metric caps revenue. Example: a churn-prevention agent that saves millions annually but triggers only a few hundred outcome charges. Snowflake should include a revenue floor (minimum monthly commitment) and a revenue ceiling (cap on outcome charges) to align incentives. This prevents customer sticker shock while ensuring Snowflake captures a share of outsized value. A quarterly renegotiation clause tied to agent performance benchmarks adds flexibility.

Sources

• Snowflake Cortex Agents Documentation
• Salesforce Agentforce
• Databricks Mosaic AI
• Microsoft Copilot
• Octane AI
• Snowflake Pricing Page

FAQ

What is the main difference between per-query and per-outcome pricing? Per-query pricing charges for each message or interaction, regardless of the business value generated. Per-outcome pricing ties costs to measurable results, like revenue saved or time reduced, so customers pay based on the actual impact of the agent.

Why would Snowflake consider moving away from per-query pricing? Per-query pricing may leave significant agent value uncaptured, as it rewards volume rather than outcomes. A hybrid model with an outcome premium lets Snowflake align pricing with customer ROI, potentially increasing revenue while making agents more attractive for high-value use cases.

How would a hybrid model work in practice? A hybrid model would combine a low base fee per message for development and staging, plus an outcome premium on documented savings or revenue gains in production. This ensures a predictable floor while capturing a share of the value created.

What counts as an "outcome" for pricing purposes? Outcomes vary by use case—for example, churn reduction in customer support, forecast accuracy improvements in RevOps, or cycle time compression in supply chain. The key is that each outcome is tied to a specific, measurable business event that the agent influences.

How does this compare to other AI pricing models in the market? Salesforce Agentforce uses a per-conversation model, while many others charge per token or per message. Snowflake's proposed hybrid model is unique in directly linking pricing to documented business outcomes, which can justify higher spend for customers who see clear ROI.

When might Snowflake actually implement this pricing change? The suggested timeline would require Snowflake to develop outcome measurement tools and test the model with early adopters before a broader rollout. This gives time for refinement based on real-world data.

Bottom Line

Snowflake's per-query pricing is a landing board, not a runway. As customers increasingly compare Cortex Agents' ROI to competitors and ask why Snowflake doesn't offer outcome pricing, the answer is to shift toward a hybrid model (query floor plus outcome bonus), partner with measurement vendors, and own outcome pricing before competitors lock Snowflake out of the value layer. A customer preventing significant churn via Cortex Agents should pay proportionally for that outcome tier, not minimal query fees. The window for establishing this model is limited; after that, it becomes a race to parity rather than category creation.

Tags

["snowflake", "cortex-agents", "agentic-pricing", "outcome-based", "per-query", "per-conversation", "sales-ops", "cro-lens", "pricing-strategy", "competitive-pricing"]

Related on PULSE

• [What data sources are most effective for training AI models to predict next best action in complex enterprise deals?](/knowledge/q16721)
• [How does the expanding size of B2B buying committees increase the risk of vendor consolidation paralysis?](/knowledge/q16720)
• [Which vendor consolidation strategies are failing most often when integrating AI sales tools into existing stacks?](/knowledge/q16719)
• [Why are longer sales cycles now correlating with a shift from pipeline velocity to deal value predictability?](/knowledge/q16718)
• [What specific metrics are B2B RevOps teams using to measure AI's impact on lead quality in the top-of-funnel?](/knowledge/q16717)

Sources

• https://docs.snowflake.com/en/user-guide/snowflake-cortex/cortex-agents
• https://www.salesforce.com/agentforce/
• https://www.databricks.com/product/mosaic-ai
• https://www.microsoft.com/en-us/copilot/microsoft-copilot-pro
• https://www.getoctane.io/
• https://www.snowflake.com/pricing/