[object Object]

The Model

Customers pay for outcomes — qualified leads, resolved cases, completed research reports — not seat licenses or token consumption. The vendor provides the AI agents (and, in most cases, humans-in-the-loop for the long tail) and absorbs all infrastructure, model, and operational risk. Industry shorthand: AaaS, AI-Agent-as-a-Service. The accounting treatment shifts the cost from a software CapEx-style commit to a variable COGS line item — finance teams need to model it differently.

The category is real enough that Bessemer, a16z, and Iconiq all published 2026 AaaS market maps in Q1, and three Gartner Magic Quadrants are being scoped for outcome-priced customer service, sales agents, and research agents.

Concrete Examples

  • HubSpot Breeze outcome pricing (April 2026): $0.50/resolved customer ticket, $1.00/qualified lead.
  • Decagon: per-resolved-conversation pricing for customer service, contracts based on baseline deflection rate.
  • Sierra (Bret Taylor): per-resolution outcome pricing with SLA guarantees.
  • 11x.ai and Artisan: AI BDRs at $X per booked meeting.
  • Cresta: per-handled-conversation in contact centers.
  • Harvey and EvenUp: per-completed-deliverable pricing in legal services.
  • Salesforce Agentforce: $2/conversation as the originally-published rate, now hybrid with platform fee.

Customer Benefits

Spend aligns with realized value — no idle capacity. Vendor handles scale, reliability, and model upgrades. Reduced integration complexity (the vendor often pre-builds connectors). Faster time-to-value than DIY agent building, which routinely runs 3–9 months. Pricing comparability against the labor cost it’s displacing — board conversations get cleaner.

Customer Risks

The vendor controls the agent. You depend on their quality, availability, and pricing discretion. Switching costs grow as integrations deepen and historical context accumulates inside the vendor’s environment. You have less control over process, escalation, and brand voice than with an in-house build. Margin trap: at high volume, per-unit pricing exceeds the marginal cost of insourcing — vendors price to extract value, not to share it. Run the breakeven math at year three, not year one.

Lock-in points to watch:

  • Conversation history and learned context that can’t be exported.
  • Custom workflows wired into the vendor’s UI, not your CRM.
  • Outcome definitions written in the vendor’s contract, not auditable independently.
  • Knowledge bases ingested into proprietary stores.

Cost Considerations

Per-outcome pricing is rational at low-to-medium volume and irrational at high volume. Crossover threshold (rough rule): if your sustained run rate is more than ~50,000 outcomes/month and the per-outcome price exceeds $0.30, build-vs-buy starts to favor build. Below that, AaaS wins on time-to-value and operational burden almost every time.

Negotiate for: outcome definition transparency (what counts as resolved), volume-tier discounts, an exit clause with data export, and an SLA with credit consequences.

Implementation Sequence

  1. Pilot one vendor on a bounded use case (single intent class, one channel).
  2. Validate the outcome definition matches your business definition — they will diverge.
  3. Compare per-outcome price against your fully-loaded internal cost.
  4. Negotiate annual commit only after 90 days of clean data.

What to Watch in 2026

Outcome definition disputes (the next “what counts as a unique visitor” battle). Standards bodies floating a metering schema. Hybrid AaaS+software pricing as vendors hedge variable risk.

[object Object]
Share