How should Snowflake price Cortex agents — per query or per outcome?
Snowflake should shift from per-query to hybrid per-outcome model by 2027, anchored to customer ROI (churn reduction, revenue lift, cycle time compression). Current per-message pricing ($0.005–$0.05/msg, live Q1 2025) is consumption-efficient but leaves 70–80% of agent value on the table.
A defensible 2027 frame: charge *base query tier* ($0.02–$0.05/msg for dev/staging), then *outcome premium* (5–15% of incremental revenue/cost saved, minimum $5K/agent/quarter) for production use. This mirrors Salesforce Agentforce ($2/conversation floor) but captures the *why*—the actual business event, not the interaction count.
Key terms:
- Per-query (status quo): Snowflake's current model; scales with agent chatter, not business value; incentivizes thin prompts + cold batch runs.
- Per-outcome (target): Price tied to measurable events ("agent decided on X churn-risk customer"; "agent generated forecast cut error by Y%"); customer sees ROI, Snowflake captures value creation.
- Hybrid floor: Low base fee ($0.01–$0.03/msg) eliminates free-tier abuse; outcome tier (5–15% of documented savings) drives expansion.
- Outcome unit: Varies by vertical—RevOps agents price on "forecast accuracy delta," support agents on "resolution time saved," risk agents on "$ churn prevented."
Per-Query vs Per-Outcome: Pros & Cons
Per-Query Advantages:
- Simplicity: usage = cost; no outcome tracking burden on customer or Snowflake ops.
- Predictability: CFOs forecast AI spend via message-per-user × cost/msg.
- First-mover friendly: lowers land friction vs. Outcome risk (what if agent *doesn't* deliver?).
- Developer-native: engineers optimize toward reduced API calls; aligns with Snowflake's "pay for compute" DNA.
- Rapid adoption: no outcome-audit requirement slows pilot→prod transition.
- Competitive equivalence: matches Databricks Mosaic AI + OpenAI's per-token model; expected by market.
Per-Query Disadvantages:
- Value leakage: agent might prevent $2M churn risk but Snowflake earns $0.50 in compute fees; customer gets 4000× ROI, Snowflake starves.
- Wrong incentive: vendor optimizes for *chatter*, not *impact*; encourages bloatware agents, long feedback loops.
- Commoditization: Cortex Agents drift toward commodity utility pricing; no defensibility vs. Open-source Claude/Llama agents running self-hosted.
- Ceiling effect: per-message math can't scale with customer value; a $500M customer using Cortex Agents pays same $/msg as a $50M startup.
- Churn risk: when outcome visibility lands (late 2026, early 2027), customers ask "Why am I not paying for ROI?" and shop alternatives (Salesforce Agentforce ROI-link, Databricks outcome insurance).
- Sales complexity: sales reps must explain why agent-driven $10M churn save costs the same as a test script.
Per-Outcome Advantages:
- Defensibility: outcome-linked pricing is *table-stakes* for Agentforce, Mosaic AI, and emerging agentic-pricing vendors (e.g., Octane).
- Land-expand: pilot at low per-query floor; outcome tier triggers only on prod success; risk transferred to customer (we earn if *you* win).
- Sales narrative: "Your agent prevented churn → you pay 5–10% of savings" is framing customers *already expect* from Salesforce / Microsoft / Databricks by 2027.
- LTV flywheel: high-outcome customers become upsell targets (expand agent swarm, add decision-layer agents); per-query never scales LTV.
- Analyst moat: agentic-outcome pricing becomes core to TAM models; Snowflake owns the narrative vs. Commodity per-token shops.
- Vertical bundling: outcome model enables industry-specific SLAs (retail: "agent accuracy ≥95% or we discount"; healthcare: "agent decision audit ≤5% of interactions").
Per-Outcome Disadvantages:
- Operationalization: requires outcome-audit partner (Pavilion, Bridge Group, Klue for measurement). Snowflake carries SLA risk.
- Customer friction: "How do we prove the agent caused the outcome?" pushes ML discipline; mid-market orgs lack regression maturity.
- Sales cycle drag: outcome pilot → measurement framework → outcome audit = 120–180 days; per-query sells in 14 days.
- Sandbagging: smart customers under-report churn prevention or hide agent value in broader process improvements; Snowflake chases audit.
- Cannibalization: mid-market customers who *can't* measure outcomes move to cheaper competitor (e.g., open-source Claude agents); outcome-model pricing only works at scale.
- Downside protection: what if agent *increases* churn (bad prompt, hallucination)? Snowflake can't charge negative; customer sues for damages.
What Snowflake Should Charge (2027 Roadmap)
- Q3 2025–Q4 2025: Baseline floor ($0.015–$0.03/msg for all tiers). Kill free-tier abuse; standardize per-query pricing across dev/staging/prod. Maintain current $0.005–$0.05 envelope to avoid customer shock.
- Q1 2026: Outcome-measurement API (GA). Launch Cortex Outcome Tracker: drop-in agent hooks that capture decision events, churn-risk flags, forecast deltas, revenue impact. Partner with Pavilion (go-to-market motion) + Klue (competitive win/loss signals) to wire measurement into customer success workflows.
- Q2 2026: Hybrid pricing pilot (100 accounts, Strategic tier only). Test 3-tier hybrid on accounts >$2M ARR: (a) per-query floor ($0.02/msg), (b) outcome bonus (5% of documented churn saved or forecast-accuracy improvement), (c) exclusive pricing on new Cortex Agent models. Gather 50+ outcome-audit case studies.
- Q3 2026: Outcome pricing GA (ABM tier expansion). Roll outcome model to 500+ ABM accounts ($500K–$5M ARR). Offer buyout options: (i) upgrade to per-outcome + 15% discount on query floor, or (ii) stay per-query at +10% price increase to subsidize outcome ops. Use Bridge Group (peer benchmarking) to socialize outcome norm; Force Management (sales training) to arm reps with ROI-attach narrative.
- Q4 2026: Pricing-model transparency (public docs + analyst relations). Publish Snowflake Cortex Agents Pricing Benchmark (3,000+ customer cohort): per-query vs. Per-outcome TCO by vertical, org size, agent swarm size. Position Snowflake as outcome-pricing pioneer; brief Gartner/Forrester on agentic-AI economics. Cite Octane (new vendor) as outcome-pricing reference platform; position Snowflake as enterprise superset.
- 2027 Q1+: Outcome insurance + dynamic pricing. Introduce outcome insurance tier (8–12% of query + outcome layer): Snowflake guarantees minimum agent performance (e.g., "agent maintains ≥90% forecast accuracy or we refund 25% of month") *and* captures upside above guarantee. Announce Dynamic Cortex Agents pricing: model complexity, concurrency, and customer vertical automatically set per-query + outcome floor via machine-learning pricing engine.
- 2027 Q2+: Vertical outcome bundles. Launch Cortex Agents for [Industry] bundles: 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 + outcome audit via Pavilion.
- Competitive wall: Enterprise seat + outcome floor. For customers >$10M ARR, offer seat-based tier ($50K–$200K/year per Cortex Agent seat, all queries + all outcomes included). Undercuts Salesforce Agentforce's per-conversation model ($2/convo × 100K conversations = $200K/year) while capturing all value creation; defense against "Why not just use Salesforce?"
Pricing Model Roadmap
| Pricing Model | Customer Profile | 2025 (Current) | 2027 (Target) | ARR Impact |
|---|---|---|---|---|
| Per-Query | Dev/Staging, cost-conscious startups | $0.005–$0.05/msg | $0.015–$0.03/msg (floor) | Baseline; 0% expansion |
| Hybrid (Query + Outcome) | Mid-market ($500K–$5M ARR) | N/A (pilot Q2 2026) | $0.02/msg + 5–10% churn saved | +40–60% LTV per customer |
| Per-Outcome | Enterprise strategic ($5M+ ARR) | N/A | 8–15% of documented value creation (min $50K/agent/year) | +120–180% LTV; 3–5 agents/customer |
| Outcome Insurance | Risk-averse enterprise | N/A | 10–14% of query + outcome + SLA guarantee | +70–100% LTV; stickier churn |
| Seat-Based | Large-scale agent swarms (10+ agents, $10M+ ARR) | N/A | $50K–$200K/seat/year (all queries + outcomes) | $500K–$2M ARR per customer |
| Open-source Defense | SMB/price-sensitive | N/A (no model) | $0.008/msg self-hosted claude/llama agents | Retention lever; prevent churn |
| Vertical Bundles | Industry-specific (SaaS, Retail, Healthcare) | N/A (pilot Q1 2027) | $100K–$500K/year + outcome bonus | High-touch, high-margin |
Outcome Architecture (Mermaid)
Bottom Line
Snowflake's per-query pricing is a landing board, not a runway. By late 2026, customers will openly compare Cortex Agents' ROI to Salesforce Agentforce ($2/conversation, outcome-implicit) and ask why Snowflake doesn't offer outcome pricing. The answer: shift now to hybrid (query floor + outcome bonus), partner with measurement vendors (Pavilion, Klue, Octane), and own outcome pricing before Salesforce / Databricks / open-source lock you out of the value layer.
A $500K customer preventing $10M churn via Cortex Agents should pay $500K–$1.5M/year for that outcome tier, not $500. The window closes Q2 2026; after that, it's a race to parity rather than category creation.
Tags
["snowflake", "cortex-agents", "agentic-pricing", "outcome-based", "per-query", "per-conversation", "sales-ops", "cro-lens", "2027-roadmap", "competitive-pricing"]
FAQ
What hybrid pricing structure does the article recommend for Cortex agents by 2027? It recommends a base query tier of $0.02-$0.05/msg for dev/staging plus an outcome premium of 5-15% of incremental revenue or cost saved, with a minimum of $5K per agent per quarter for production use.
A low per-message floor of $0.01-$0.03 eliminates free-tier abuse. The article says the current per-message model ($0.005-$0.05/msg, live Q1 2025) leaves 70-80% of agent value on the table.
How does the proposed model compare to Salesforce Agentforce? The article notes Agentforce uses a $2/conversation floor, and its hybrid recommendation mirrors that floor structure but captures the underlying business event rather than interaction count. Per-outcome pricing is framed as table-stakes that customers will already expect from Salesforce, Microsoft, and Databricks by 2027.
Outcome-linked framing ("your agent prevented churn, so you pay 5-10% of savings") is presented as the expected sales narrative.
What is the "value leakage" problem with per-query pricing? The article gives the example of an agent preventing $2M in churn risk while Snowflake earns only $0.50 in compute fees, meaning the customer gets a 4000x ROI while Snowflake starves. Per-query pricing also creates the wrong incentive, rewarding chatter over impact and encouraging bloatware agents.
A $500M customer pays the same per-message rate as a $50M startup, capping Snowflake's value capture.
What are the main disadvantages of per-outcome pricing the article raises? It requires an outcome-audit partner and carries SLA risk, since Snowflake must measure whether the agent caused the outcome. Customers may sandbag by under-reporting churn prevention, the sales cycle stretches to 120-180 days versus 14 for per-query, and mid-market orgs that can't measure outcomes may defect to cheaper open-source Claude agents.
There's also no way to charge negatively if an agent increases churn through a bad prompt or hallucination.
What rollout timeline does the article lay out for the pricing shift? A baseline floor of $0.015-$0.03/msg ships Q3-Q4 2025, a GA outcome-measurement API (Cortex Outcome Tracker) launches Q1 2026, a hybrid pricing pilot on 100 accounts over $2M ARR runs Q2 2026, and outcome pricing reaches GA across 500+ ABM accounts in Q3 2026.
Pavilion and Klue are named partners for wiring measurement and win/loss signals into customer success workflows. Bridge Group is cited for socializing the outcome norm via peer benchmarking.