Total Cost Components
Platform licenses cover Salesforce Sales/Service Cloud editions, Dynamics 365 Sales Premium, HubSpot Sales Hub Enterprise, and the per-seat math that compounds with org growth. AI consumption now sits as a parallel line — Agentforce at $2 per conversation in the standard SKU, OpenAI GPT-5 token spend, Anthropic Claude pricing per million input/output tokens, plus prompt-caching credits where applicable. Data infrastructure includes Data Cloud credits, vector database fees (Pinecone, Qdrant, Weaviate), observability platforms (Datadog, Splunk, Langfuse, Langsmith), and the warehouse compute (Snowflake credits, BigQuery slots) that backs them. Integration platform fees from MuleSoft, Workato, Boomi, or n8n. Third-party apps from the AppExchange, Microsoft AppSource, and HubSpot Marketplace. Professional services from the SI partner. Internal engineering time — the line most often forgotten and usually the largest.
Visibility
Tag every cost by team, feature, customer segment, and use case. Build dashboards that show total cost per outcome — cost per supported customer, cost per resolved case, cost per closed-won opportunity. Aggregate numbers hide optimization opportunities; the team that knows their EMEA enterprise segment costs $4.20 per ticket while their NA SMB segment costs $11.80 has the data to act. Salesforce Digital Wallet, AWS Cost Explorer, Azure Cost Management, and FinOps-specific tools (Vantage, Finout, CloudZero) all support tag-based slicing. The discipline is upfront: if every new agent, app, and integration starts life with required tags (team, owner, use_case, environment), reporting is trivial; if tagging is retrofitted, it never finishes.
2026-04 Cost Snapshot - Service Org
Component Spend Tag
Salesforce licenses $480K cloud=sf, team=service
Agentforce conversations $122K ai=agentforce, use_case=triage
Anthropic Claude API $38K ai=claude, use_case=summary
Data Cloud credits $54K data=dc360
Pinecone vector DB $11K data=vector, use_case=kb
Datadog + Langfuse $19K obs=apm+llmops
Total $724K
Cost per resolved case $1.18Optimization Levers
License utilization — Salesforce Optimizer and equivalents identify users who logged in fewer than five times in 90 days; retire those seats. Edition rightsizing — many teams pay for Unlimited when Enterprise plus targeted add-ons would cover the workload. Committed-use discounting on AI — Anthropic, OpenAI, and the hyperscalers all offer 15-40% discounts for annual commits. Tier-shifting AI features — route the easy 70% of queries to a smaller, cheaper model (Haiku, GPT-5-mini, Llama 4 self-hosted) and reserve the expensive model for the hard 30%. Vendor consolidation where two tools overlap. Prompt caching to reuse system prompts across conversations (Anthropic’s prompt caching cuts repeated-prompt cost by 90%).
Governance Cadence
Weekly review for spike detection — a single Agentforce flow looped during a misconfiguration can burn $50K in a long weekend. Monthly review against budget with the budget owner, not the data team alone. Quarterly vendor and consolidation review with renewals on the table. Annual comprehensive audit. The cadence absent means cost surprises; the cadence present means controlled posture.
What Changed in 2026
Outcome-based pricing — Salesforce’s $2-per-conversation Agentforce SKU, Sierra’s per-resolution pricing, Decagon’s per-case model — moved CRM AI from an opex-as-license to opex-as-consumption category. The implication: cost is now a product metric, not just a finance metric. Engineering teams have to ship against a unit-economics target, which changes how they prompt, retrieve, and route.
Common Failure Modes
The recurring failures: discovering AI spend a month after it tripled, treating engineering time as free, not committing on AI volume because “we might not use it” and then paying list price all year, and consolidating vendors at the wrong layer (giving up the better tool to save 5%).
What to do this week
Pull last month’s CRM-related spend, tag it by use case, and compute one cost-per-outcome metric. Bring it to the budget owner. If they cannot name the metric without coaching, make that conversation the recurring one.
