Per-Conversation Pricing: Examples & Companies

What is it

Per-Conversation Pricing is a billing unit where each complete customer conversation — from first message to resolution — is metered as a single chargeable event.

The unit’s appeal is legibility. A conversation is bounded in time and observable to both parties: it begins when a customer sends a first message and ends when the thread closes, through AI resolution or human escalation. Neither side counts tokens or agent messages to agree on what happened, which makes the conversation the natural default meter for AI customer-service platforms that want a usage signal without token economics.

Six corpus companies use conversations as a primary or secondary unit — Sierra, Decagon, Forethought, Maven AGI, Cresta, and Yellow.ai — all in AI customer service. It sits adjacent to, and is often layered with, per-resolution pricing: a conversation meter charges for every exchange the AI handles, resolved or not; a resolution meter charges only when the AI closes the issue without a human. Which meter a vendor uses, or whether it blends both, is the first question to settle before signing — the choosing the right usage metric guide frames the call.

How it works

The bill is structurally simple — total = conversations × rate — but three design decisions shape what it actually costs.

What bounds a conversation. Most platforms define one as a thread initiated by a single customer contact that stays within one session or ticket context. Abandoned sessions (a chat opened but never continued) are usually excluded, and a transfer to a human typically closes the AI’s conversation and opens a new human-handled thread. The edge cases — a reopened chat, a dropped-and-reconnected voice call, an email chain spanning days — are where contract language earns its keep.

Whether the charge stands alone or rides on a platform fee. In every disclosed model here, the per-conversation rate is only part of the bill. The table below pairs each vendor’s unit with its base structure.

Whether the meter counts conversations or resolutions. Decagon documents this trade-off most explicitly, offering enterprise buyers a choice: a flat rate per AI-handled inquiry charged regardless of outcome with volume discounts, or a higher per-unit rate charged only when the AI resolves an issue end-to-end — which its glossary defends as value “aligned with cost.” Most platforms bury the same choice inside a quote.

Unit math: bill = platform_fee + (conversations_handled × conversation_rate), or for outcome variants = platform_fee + (conversations_resolved × resolution_rate). The first form earns the vendor money on every exchange; the second, only on closed issues. See the introduction to usage-based pricing for how the two compare.

Companies using this

All six adopters operate in AI customer service — contact-center and support automation, where a customer initiates a thread and the value delivered is closing it. The table below sorts by pricing model, billing units, and free-tier availability.

Patterns observed

The unit is domain-locked. No corpus coding tool, document platform, or analytics product bills per conversation; the meter only appears where the workflow is a customer-initiated thread closed by an AI. It is the narrowest billing unit on the site by industry spread.

Positioning and contract diverge. Vendors market “outcome-based” language over meters that are not standardized underneath. Sierra frames its model as “pay for a job well done,” yet bills a blend of volume pricing for routine interactions and per-resolution charges for complex ones. The terminology travels; the meter does not — so a buyer must map each vendor’s copy to the precise unit in the contract.

Transparency is near-absent. Five of the six adopters are entirely sales-only, leaving Yellow.ai’s public rate the category’s single stated unit anchor. That opacity is why the meter definition, not the sticker, is the real negotiation.

Counterexamples & variants

The taxonomy can mislead: Cresta is seat-first. Cresta’s core commercial model is per-agent-seat, not per-conversation — a reflection of its human-in-the-loop, agent-assist origins. Conversations surface only in its autonomous AI Agent product, where containment-linked economics layer onto the seat fee. A buyer should not infer a per-conversation model from the conversations tag in its taxonomy.

The unit under-aligns incentives. A vendor billing per conversation earns whether the AI resolved the issue or escalated it, so at high escalation rates the buyer pays for interactions that deflected nothing. Decagon’s per-resolution alternative exists precisely to close that gap, trading a higher per-unit price for value alignment — the resolution meter simply refuses to bill the escalations.

Voice diverges from chat. Vendors separate messaging from voice, and voice consistently implies a higher per-unit price. Maven AGI structures contracts around autonomous-resolution volume across both channels, and Yellow.ai sells its Nexus Vox voice product separately from its chat-oriented free tier. A cross-channel deployment needs per-channel rates in writing, not the chat rate applied uniformly.

What this means for buyers vs vendors

For buyers

The rate per conversation matters less than the definition around it: how a conversation is bounded and whether escalations are excluded. A per-conversation-handled contract exposes you to full cost at low resolution rates; a per-conversation-resolved contract protects you but costs more per unit. Get the resolution definition in writing, confirm how abandoned and transferred threads are counted, and demand a worked example against your historical escalation rate. The choosing the right usage metric guide pressure-tests whether conversation or resolution fits your workload.

For vendors

Per-conversation billing buys forecastable revenue — the meter runs on every AI-handled exchange, not only successful ones — but it detaches revenue from AI quality, which increasingly reads as a liability as buyers benchmark against outcome-aligned peers. Forethought’s public position points to the durable hybrid: a platform fee for access plus a per-resolution component for the work delivered, a revenue floor that keeps outcome alignment on the usage dimension.

Explore this theme in the knowledge graph →