Salesforce has unveiled Agentforce Help Agent, a prepackaged AI customer support agent built on its Agentforce 360 platform, designed to ground itself on a company’s own knowledge and deploy across channels within minutes. The product itself is noteworthy, but perhaps even more significant is the pricing. Rather than charging per seat, per conversation or per AI action, Salesforce says customers pay only when Help Agent resolves a customer issue autonomously from start to finish. If a customer asks for a human or leaves unhappy, there is no charge.
Kishan Chetan, EVP and GM of Agentforce Service at Salesforce, frames the move as providing a closer connection to the end goal: “with our pay-per-resolution pricing, our success is directly tied to our customers’ success.” The CRM giant is not alone in making this move. As AI moves from experimentation to production, vendors are under increasing pressure to prove measurable value rather than simply impressive technology.
What Help Agent actually does
Help Agent is built on Salesforce’s Agentforce 360 platform and is designed to be set up with guided configuration rather than built from scratch, grounding itself automatically on a company’s existing Salesforce Knowledge base, with the option to add further files or crawl a web URL. An agent preview pane lets teams test responses before going live.
The agent comes with a library of prepackaged actions, allowing it to manage cases, schedule appointments and update orders on a company’s existing workflows, with further actions such as account management available through Agentforce Builder.
It can also be switched on across voice, web, portal and messaging channels from a single screen, and a reimagined Customer Service Portal lets customers describe what they need through a single conversation bar, with the interface adapting in real time as it works through the request.
Why outcome pricing matters
Traditional software pricing has long centred on users or licences. Generative AI introduced new models built around tokens, conversations or individual AI actions. Those approaches often left customers paying regardless of whether the AI actually delivered value, since usage and outcome are not the same thing.
Pay-per-resolution pricing, however, attempts to bridge the divide by tying vendor revenue directly to customer success, which in turn reduces some of the financial risk enterprises take on when investing in AI automation. Instead of asking how many AI interactions an organisation had in a given month, the more useful question becomes how many customer issues were actually resolved.
A wider shift in enterprise AI
Enterprises increasingly expect measurable return on investment, productivity gains, automation rates and resolution outcomes rather than AI capability for its own sake, and a growing number of CX vendors are leading with business metrics instead of model specifications. Zendesk recently moved towards a similar resolution-based commercial model alongside its own autonomous service agents, while 8x8’s chief executive has argued that pricing needs to shift away from per-seat models as AI takes on a larger share of customer interactions.
The catch: defining a resolution
Outcome-based pricing raises its own questions. What actually counts as a resolution, and does handing a customer to a human agent count if the underlying issue was eventually solved? What happens if a case reopens shortly after being marked resolved? And could vendors, consciously or not, end up optimising for straightforward cases while steering clear of more complex interactions where resolution is harder to guarantee? These questions are likely to become more pressing as outcome-based pricing gains wider adoption across the industry.
Whether Salesforce’s pricing model becomes the industry standard remains to be seen, but it highlights a broader shift in enterprise AI, from selling access to AI towards selling measurable business outcomes. It is a bold move, although perhaps a necessary one, as it demonstrates a confidence in the product living up to its promises. Therein lies the key to the success of Help Agent and other outcome-based AI pricing models: how many successful resolutions will they achieve and for how long?