Outcome-based AI pricing and the problem of defining a resolution

Several of the largest customer experience platforms now charge for their AI agents by outcome rather than by conversation. Intercom charges $0.99 per resolution. Zendesk charges $1.50 per automated resolution on committed volume, or $2.00 pay-as-you-go. HubSpot, as of 14 April, charges $0.50 per resolved conversation, down from $1.00 per conversation regardless of whether the AI solved anything. Sierra uses outcome-based pricing as its core model, and Leena AI dropped consumption pricing after finding that usage-based billing made customers reluctant to use the product.

Vendors are moving away from per-seat, per-conversation, and per-credit charging, towards charging only when their AI delivers a result. On paper this favours the buyer, who pays for value not volume. The catch sits in what each vendor counts as a resolution.

Every vendor defines resolved differently

Intercom counts a resolution when the customer confirms the answer helped, or simply does not ask for more help after Fin responds, so a customer who walks away is billed even if they gave up. Zendesk waits 72 hours, then runs a separate AI model to judge whether the issue was likely solved, which is more rigorous but still one AI marking the work of another. HubSpot has not yet published how it defines resolved under the new model. Some vendors stretch the term on purpose: Siena AI documented one that counted a conversation as resolved because no human agent took over even when the problem was never fixed, and another that billed a resolution every time its AI served a help-article link the customer clicked, whatever the relevance.

When a vendor’s revenue rises with the number of resolutions, the vendor has every reason to read the definition generously. Zendesk’s chief executive, Tom Eggemeier, said as much, noting that too many companies count deflections and non-answers as resolved. BCG’s research into outcome-based AI pricing found buyers value the alignment of cost to value but argue often about whether an issue was truly resolved, with some setting up arbitration for disputes. A customer who gets a mediocre answer and gives up is a resolution on the vendor’s invoice, and silent churn on yours.

The bill grows as the AI improves

The headline rates look small, but they add up. A contact centre handling 10,000 conversations a month at a 50 per cent AI resolution rate would face monthly AI charges between $2,500 with HubSpot and $10,000 with Zendesk pay-as-you-go, on top of platform and seat spend. Costs also climb as the AI improves: resolution rates that start around 30 per cent in month one can pass 50 per cent by month six, so the cost per resolved issue drops while total spend rises.

Australian buyers carry an extra edge, because every vendor above prices in US dollars. HubSpot’s $0.50 sits closer to $0.78 AUD and Zendesk’s $2.00 rate near $3.12 AUD, and exchange-rate movement adds cost that per-seat models never had. Seasonal peaks sharpen it: a retailer handling 5,000 conversations a month in March might handle 25,000 in December and watch the AI bill scale five times alongside. Compliance matters too. Australian Consumer Law requires that consumers can reach complaint resolution and obtain genuine remedies, so in regulated sectors overseen by ASIC, APRA, and the ACMA, resolved in the billing system needs to match resolved in a regulatory sense.

Questions to settle before you sign

Before signing any outcome-priced platform, press on the detail.

• Get the contractual definition of a resolution in writing: whether the customer must confirm, whether a timeout window applies, and what happens if they return with the same issue days later.
• Confirm you can audit the resolution data yourself. Zendesk offers an automated resolutions dashboard, so ask whether you get equivalent visibility.
• Understand the dispute process, including whether arbitration exists and whether you can flag false positives.
• Ask for committed tiers, caps, and overage policies, and model the bill at five and ten times your current volume.
• Check whether CSAT or CES is tracked per interaction, since without it the vendor cannot tell a genuine resolution from a customer who gave up.

Outcome-based pricing is a better model than per-seat or per-conversation pricing, because it ties vendor incentives to buyer value. It only holds where the outcome measured is the customer’s and not the bot’s. Without an industry-standard definition of resolved, and without independent tools to verify it, it shifts the complexity into business logic rather than removing it. The organisations that handle it well will be the ones who asked the hard questions before they signed.

CC Tech is an independent technology advisory that helps Australian organisations evaluate and select CX, contact centre, and AI platforms. Our advice is vendor-funded, so there is no cost to the client. If you are evaluating AI agent pricing models and want an independent view, book a call.