Sierra AI Pricing: Sierra AI Agent Cost, Outcome Based Pricing and Published Resolution Rates

Sierra AI costs whatever your negotiated contract says, because Sierra publishes no prices anywhere. We checked this directly rather than repeating it: on September 2, 2026, sierra.ai/pricing returns HTTP 404, and a scan of the homepage markup found no link containing the word pricing. There is no plan table, no starting-from figure, and no self-serve signup path. Every route into the product is a sales conversation.

That leaves the reported figures, which are worth quoting only if you also state where they came from and how much weight they carry. The widely repeated numbers are $1.00 to $2.50 per resolution, an annual platform commitment starting near $150,000, and implementation of $50,000 to $200,000 depending on how many systems the agent has to reach. Those figures appear across many buyer guides, but essentially all of them are published by vendors who sell against Sierra, and none of them cite a Sierra document. Treat them as a directional range for budgeting, not as list prices.

Why the per-resolution rate is not the number that matters

The unit everyone quotes is a resolution. The unit your support organization actually counts is a contact. Converting between them is the whole game, and it is the step every competing page skips.

If Sierra charges $1.50 per resolution and its agent resolves 70% of what arrives, your effective cost is $1.05 for every contact that hits the queue, and you still pay human agents for the other 30%. Change the resolution rate to 90% and the same unit rate becomes $1.35 per contact, which is more expensive, not less, even though the agent is performing better. Outcome pricing has that counterintuitive property built in: a better agent produces a bigger invoice. That is by design, and it is defensible, but it needs to be in your model.

Sierra charges when its agent produces an outcome. Sierra set this out publicly in a post dated December 10, 2024, and the language is unusually specific for a company that publishes no numbers. Sierra defines an outcome as "a resolved support conversation, a saved cancellation, an upsell, a cross-sell, or any number of valuable outcomes," and gives as a worked example "managing an issue that would typically require a 20-minute call with L2 technical support."

Two commitments in that post are the ones to hold Sierra to, because they are the buyer protections in the model:

• "If the conversation is unresolved, in most cases, there's no charge."
• "If a case needs to be escalated, in most cases, there's no charge."

Read the qualifier. "In most cases" is doing real work in both sentences, and it is exactly the phrase your procurement team should convert into contract text before signing. Ask which cases are the exceptions, and get the answer in the agreement rather than in an email.

The definitional risk nobody prices in

The weakness of outcome pricing is not the rate. It is that the billable event is a judgment call. Who decides a conversation was resolved? What happens when a customer comes back two days later about the same issue? Is that one resolution or two? Does a resolution that the customer rates poorly still bill?

Decagon, Sierra's closest competitor, makes this argument openly on its own comparison page, saying most enterprises prefer per-conversation pricing for "predictable costs and to avoid negotiating what counts as a 'resolution.'" That is a rival talking its own book, but the underlying point is fair and it matches what buyers report. A per-conversation meter is measurable by both parties on day one. A per-resolution meter needs a shared definition, an audit trail, and a dispute process.

None of that makes outcome pricing a bad deal. It makes it a deal that lives or dies on one contract clause. Negotiate the definition harder than you negotiate the rate.

Because Sierra bills on resolutions, the resolution rate is the single input that sets your bill. Sierra does not publish a headline accuracy number, but it does publish customer case studies, and some of those carry a hard percentage. So we counted them.

Method, stated plainly. We pulled Sierra's own sitemap on September 2, 2026. It lists 83 English customer pages. We successfully retrieved 51 of them and parsed every stat block. Of those 51, only 14 publish a resolution or containment rate at all. This is a sample of Sierra's public case studies, not a census of Sierra's customers, and it measures what Sierra chooses to publish rather than what Sierra typically delivers.

The median is 70% and the mean is 73%. The spread runs from 60% to 94%, which is a 34-point range across companies that are all, presumably, reference customers Sierra was happy to publish. That range is the most useful thing on this page for a buyer, because it tells you the resolution rate is a property of your support mix rather than a property of the vendor. A retailer with narrow, repetitive order-status volume sits at the top of that range. A financial services firm with regulated, high-variance cases sits at the bottom.

The second finding is about disclosure. Of the 51 case studies we retrieved, 31 publish no performance metric of any kind. SiriusXM's page, for example, publishes only that SiriusXM has 34 million subscribers, which is a fact about SiriusXM rather than a result from Sierra. Big logos are not evidence. When you run your reference calls, ask for the resolution rate specifically, and ask over what time window.

Sierra's opacity is normal for this category, which is worth knowing before you conclude it is a red flag. We probed the pricing URL of nine AI customer service agent vendors on September 2, 2026 and recorded what came back.

Three of nine publish anything, and only two publish a usable per-unit rate for the agent itself. There is also a middle category that is easy to miss: Zendesk publishes its seat prices ($19, $55 and $115 per agent per month billed yearly) and publishes the AI billing model in plain language, stating that AI agents are billed by Automated Resolutions and that "you pay only for customer requests that were successfully resolved." What Zendesk does not publish is the dollar value of one automated resolution. Knowing the unit without knowing the rate is partial disclosure, and it is more common in this market than either full transparency or full silence.

One detail from that check is fresh enough to be worth flagging: Zendesk's own pricing FAQ now describes the product as "Forethought AI agents by Zendesk," which confirms Forethought sits inside Zendesk rather than competing with it. If you built a shortlist from a guide written before that, you have one fewer independent vendor on it than you think.

Here is the comparison that matters, priced on a single realistic workload: 100,000 inbound support contacts a year, which is roughly 8,300 a month and typical of a mid-market US consumer brand. We hold Sierra at the reported $1.50 midpoint and hold the resolution rate at 70%, Sierra's own published median.

Read the last column, because that is where the difference lives. On pure usage, Sierra at $1.05 per contact and Fin at $0.75 per contact are within striking distance of each other, and both beat Agentforce conversation pricing, which charges on every conversation whether or not anything gets resolved. The gap opens on fixed costs. Fin publishes that there are no platform or setup fees. Sierra is reported to require a six-figure annual commitment plus implementation, which in year one can be larger than the usage itself.

That produces a clean rule of thumb. At a reported $150,000 floor and $1.50 per resolution, you need about 100,000 resolutions a year before the meter catches the minimum. At a 70% resolution rate, that is roughly 143,000 contacts a year, or about 11,900 a month. Below that volume you are buying a minimum commitment rather than a meter, and the effective price per resolution is higher than the quoted rate. Above it, Sierra's model behaves the way the marketing describes.

The market context that changes this shortlist

The published-price alternative is being absorbed by the largest incumbent. Salesforce signed a definitive agreement on June 15, 2026 to acquire Fin, the company formerly known as Intercom, for about $3.6 billion, with the deal expected to close in the fourth quarter of Salesforce's fiscal 2027. Salesforce reports Fin's agent closes roughly 76% of incoming support requests without a human.

If you are shortlisting today, that matters in two directions. It validates outcome pricing as the direction of the market, since the buyer is the company that already sells per-conversation Agentforce. It also means a bet on Fin is increasingly a bet on Salesforce, and buyers who chose Fin specifically to stay out of the Salesforce stack should ask what the roadmap looks like after close. Sierra, for its part, raised a $950 million Series E in May 2026 at a valuation reported near $15.8 billion, led by Tiger Global and GV, so vendor viability is not a live concern on either side.

Sierra is built for large, complex, brand-sensitive support operations, and its pricing is shaped to match. Being honest about where that fits, and where it does not, is more useful than a verdict.

Sierra fits when you run enough volume that a six-figure commitment is a rounding error against your support payroll, your cases are complex enough that a simple deflection bot has already failed, you need voice and chat under one agent, and you have an executive sponsor who can hold a vendor to a resolution-rate target. The customers Sierra publishes at the top of its resolution range, like Ramp at 90% and Airtable at 80%, look like this. Sierra's forward-deployed model, where its own engineers configure the agent, is a genuine advantage for a team without AI engineers of its own.

Sierra does not fit when your annual contact volume is under roughly 143,000, because the reported platform floor dominates and you pay for capacity you never use. It also does not fit when you need to be live in weeks rather than months, when you cannot get procurement to sign a contract whose billable unit is defined by the vendor, or when you want to try the product before committing, since there is no trial and no self-serve tier. For those cases a published-rate vendor with no floor, or a ready-made agent you can deploy and cancel, is the honest answer.

If you are earlier in the process and still deciding whether to buy an agent at all rather than which one, our AI agent pricing guide covers the pricing shapes across the whole market, and AI agent platforms compared puts the published rates of thirteen platforms in one table. For the Salesforce side specifically, Agentforce pricing has the full Flex Credits arithmetic.