Cost per resolution is the number the enterprise support market has settled on, and it is the one number Cresta does not print anywhere. The company sells AI agents that handle enterprise support from first contact to closed ticket, and its site documents the outcomes those agents produce in generous detail. It documents what they charge nowhere at all.
That gap is easy to measure rather than assert. Across seven Cresta pages totalling 9,566 words — the AI Agent hub, the omnichannel product page, four named customer stories and the customer-stories index — there are 27 distinct percentage figures and exactly two dollar amounts. Both dollar amounts are revenue. Neither is a price. The words “pricing”, “price”, “per resolution”, “outcome-based” and “per seat” appear zero times on all seven. A request for cresta.com/pricing returns HTTP 404, and so does cresta.com/plans.
This matters more in 2026 than it would have in 2024, because the rest of the category has converged on a published unit. Salesforce, Intercom and Zendesk now bill enterprise support automation by the resolved conversation, and two of the three print the rate on a public page. This article compares what those vendors disclose, shows why the definition of a “resolution” moves the bill further than the headline rate does, and works through what Gartner’s 2030 forecast does to the arithmetic. It is a companion to our audit of Cresta’s Training Simulator disclosures, which found the same pattern in a different product line.
Table of contents
- What Cresta Launched, and the Price It Did Not Publish
- Cost Per Resolution Became the Unit of Enterprise Support AI
- What Cresta Publishes Instead of a Cost Per Resolution
- Two Vendors Print a Cost Per Resolution. Four Do Not
- The Definition, Not the Price, Decides Your Bill
- Gartner Says the Cost Per Resolution Stops Falling
- What a Cost Per Resolution Costs at Enterprise Volume
- How to Buy When the Cost Per Resolution Is Not Published
- What This Means for Enterprise Support Buyers
- Frequently Asked Questions About Cost Per Resolution
- References and Further Reading
What Cresta Launched, and the Price It Did Not Publish
Cresta’s AI agent line is positioned as end-to-end enterprise support: the agent takes the contact, works the case across voice and digital channels, and closes it without a human unless one is needed. The platform wraps that with discovery, build, test, deployment and optimisation tooling, plus an operations centre for supervising live conversations.
What the product pages claim
The AI Agent hub runs to 1,941 words and is unusually specific about results. It carries a 5.5x higher containment figure alongside a 23% CSAT improvement, and a second case showing a 74% containment rate with a $3.3M revenue lift. The omnichannel page repeats the 5.5x containment claim across 973 words.
What they never mention
Neither page contains the word “price” or “pricing”. Neither describes a billing unit. There is no rate card, no per-seat figure, no per-conversation figure, and no statement of whether the commercial model is a subscription, a consumption charge, or a cost per resolution. The only route to a number is the contact form.
Why a missing rate card is normal, and still worth noting
Enterprise software has always sold this way, and nothing about withholding a rate card is improper. What has changed is that Cresta’s closest competitors stopped doing it. When two of the vendors on a buyer’s shortlist publish a cost per resolution and the third publishes nothing, the asymmetry itself becomes a procurement fact.
Cost Per Resolution Became the Unit of Enterprise Support AI
The seat licence made sense when software was a tool a human operated. An AI agent that closes tickets on its own has no seat, so vendors needed a unit that tracked the work rather than the worker. The cost per resolution is what the market picked.
From per-seat to per-outcome
Under a seat model, a support team that automates half its volume pays the same and simply needs fewer people. Under a cost per resolution model, the bill moves with the automation, which is why finance teams like it and why it spread quickly through 2025 and 2026.
Who moved
Intercom priced Fin per resolved outcome. Zendesk moved to billing per automated resolution. Salesforce launched Agentforce Help Agent on an explicit pay-per-resolution rate. HubSpot has been shifting its AI charges from per-use to per-resolution on the same logic.
What the unit promises a buyer
The pitch is that a cost per resolution converts a fixed platform commitment into a variable cost that only accrues when the software does the job. That promise is real, but it depends entirely on who writes the definition of “does the job” — and the vendors do not write it the same way.
| Cresta page | Words | Percentage figures | Dollar figures | Price mentions |
|---|---|---|---|---|
| AI Agent hub | 1,941 | 3 | 1 ($3.3M revenue) | 0 |
| Omnichannel AI Agent | 973 | 0 | 0 | 0 |
| Snap Finance story | 1,261 | 7 | 0 | 0 |
| Cox story | 1,046 | 7 | 0 | 0 |
| Windstar Cruises story | 1,741 | 3 | 0 | 0 |
| United Airlines story | 1,713 | 2 | 0 | 0 |
| Customer stories index | 891 | 5 | 1 ($2M revenue) | 0 |
| Total | 9,566 | 27 | 2 | 0 |
What Cresta Publishes Instead of a Cost Per Resolution
The table above is the whole argument in one view. Cresta is not a secretive company. It is a company that discloses one side of the ledger with real precision and the other side not at all.
Twenty-seven percentages, two dollar signs
Counted page by page, the seven documents carry 27 distinct percentage figures. Snap Finance and Cox contribute seven each. The two dollar amounts in the entire corpus are a $3.3M revenue lift on the AI Agent hub and a $2M figure on the customer-stories index. Both describe money earned by the customer, not money paid to Cresta.
The navigation promises a cost it never states
Cresta’s own platform navigation describes its Customer Care solution as delivering “brand-defining CX at a lower cost per contact”. That phrase appears in the shared navigation on all four pages probed for it, including the contact page itself. The site therefore markets a lower cost per contact on every page that carries the menu, while stating a cost on none of them.
The comparison this invites
A buyer reading these pages learns that containment can reach 74% and CSAT can rise 23%. They cannot learn what a resolution costs, what the floor commitment is, or whether an escalated conversation is billable. Every one of those is answerable from a competitor’s public page in under a minute.
Two Vendors Print a Cost Per Resolution. Four Do Not
Six vendors sell enterprise support agents into the same shortlists. Sorting them by what a buyer can read without a sales call produces three tiers, and only the top tier states a number.
The two that publish a cost per resolution
Salesforce prices Agentforce Help Agent at $2 per resolution, a figure given by Kishnan Chetan, EVP and GM of Agentforce Service, with general availability in July 2026 and a minimum purchase of 1,000 resolutions. Intercom prices Fin from $0.99 per outcome, with a stated minimum of 50 outcomes per month. Both numbers sit on public pages.
The one that publishes the model but not the cost per resolution
Zendesk states the mechanism plainly — “you pay only for customer requests that were successfully resolved by the AI agent, without any escalation to a human agent” — and lists seat prices of $19, $55 and $115 per agent per month. It does not publish the cost per resolution itself.
The three that publish neither
Decagon and Sierra both operate outcome-linked commercial models and disclose no figures publicly; third-party estimates circulate but carry no vendor confirmation, so they are not comparable evidence. Cresta is the fourth case and the strictest one: no cost per resolution, no billing model, and no pricing page to hold either.
| Vendor | Published rate | Billing unit | Stated minimum | Pricing page |
|---|---|---|---|---|
| Salesforce Agentforce Help Agent | $2.00 | Autonomous resolution | 1,000 resolutions | Yes |
| Intercom Fin | From $0.99 | Billable outcome | 50 outcomes / month | Yes |
| Zendesk | Not published | Automated resolution | Not published | Yes (seats only) |
| Decagon | Not published | Outcome-linked (reported) | Not published | No rate card |
| Sierra | Not published | Outcome-linked (reported) | Not published | No rate card |
| Cresta | Not published | Not stated | Not stated | 404 |
The Definition, Not the Price, Decides Your Bill
Here is the part that a rate-card comparison misses entirely. Salesforce and Intercom both bill per resolved conversation, and their headline rates differ by roughly two to one. Their definitions of a billable event differ in the opposite direction, and by more.
Salesforce does not bill an escalation
Salesforce’s position is explicit: “Pay when Help Agent resolves an issue autonomously, start to finish. If the customer asks for a human or walks away unhappy, there is no charge.” Chetan’s working definition is equally blunt — “we define successful resolution as the user’s question doesn’t need a human escalation.” Negative feedback, a request for a human, and abandonment are all free.
Intercom does bill an escalation
Intercom’s own pricing page lists “Procedure handoff: Fin completes a Procedure you’ve configured to hand off to a human” as one of its billable outcome types, and the Intercom pricing FAQ confirms an outcome counts when “Fin completes a workflow (Procedure), including handoffs“. A conversation that ends with a human taking over is therefore free at Salesforce and chargeable at Intercom.
Silence is billable at one vendor and not the other
The divergence repeats on the quietest case of all. Intercom counts an outcome when “no further help is requested after Fin’s last answer” — a customer who reads the reply and closes the tab is billable. Salesforce treats a customer who “walks away unhappy” as no charge. The same abandoned conversation lands on opposite sides of the invoice.
What that does to a comparison
A buyer comparing $0.99 against $2.00 concludes the first is 50.5% cheaper. That arithmetic holds only if both vendors bill the same events, and they demonstrably do not. Model the mix of resolutions, handoffs and abandonments in your own contact centre, then apply each definition to it. The cost per resolution that matters is the one your traffic produces, not the one on the page.
| Conversation ending | Salesforce Agentforce | Intercom Fin | Zendesk |
|---|---|---|---|
| Agent resolves it outright | Billed $2.00 | Billed from $0.99 | Billed |
| Customer stops replying | No charge | Billed from $0.99 | Not stated |
| Configured handoff to a human | No charge | Billed from $0.99 | Not billed |
| Customer asks for a human | No charge | Not stated | Not billed |
| Negative feedback given | No charge | Not stated | Not stated |
| Lead disqualified by the agent | Not stated | Billed from $0.99 | Not stated |
Gartner Says the Cost Per Resolution Stops Falling
Every business case for support automation assumes the unit price falls as the technology matures. Gartner’s January 2026 forecast argues the opposite, and it is the single most important input to any cost per resolution model built today.
The forecast
Gartner predicts that by 2030 the cost per resolution for generative AI will exceed $3, higher than many B2C offshore human agents. The stated drivers are rising data centre operating costs, a pivot by AI vendors from subsidised growth to profitability, and increasingly complex use cases that consume more computing power and require more expensive talent.
The regulatory multiplier
Gartner also expects assisted service volume to rise 30% by 2028 as rules guarantee customers the right to reach a human. Patrick Quinlan, Senior Director Analyst in Gartner’s Customer Service and Support practice, put it directly: “Regulations mandating easy access to human agents will encourage customers to request a human by default, bypassing AI agents.” Organisations will have to keep or rehire staff, possibly at higher salaries.
The strategic conclusion
Quinlan’s summary is unsparing: “Customer service leaders are determined to use AI to reduce costs, but return on those investments is far from guaranteed. Full automation will be prohibitively expensive for most organizations.” Gartner expects 10% of Fortune 500 firms to double customer service spending by 2030 — not to cut cost, but to differentiate. Anyone weighing automation against IT outsourcing should read that as a warning against single-variable business cases.
What a Cost Per Resolution Costs at Enterprise Volume
Unit prices under a pound feel harmless, which is exactly the problem. The arithmetic below multiplies each published rate by a stated volume; the volume is a worked assumption, not a vendor claim, and the rates are the ones quoted above.
The worked example
Take a contact centre where the AI agent produces 60,000 billable resolutions a month. At Intercom’s $0.99 that is $59,400 a month, or $712,800 a year. At Salesforce’s $2.00 it is $120,000 a month, or $1.44M a year. At Gartner’s 2030 figure of $3.00 it is $180,000 a month, or $2.16M a year.
The gap between the two published rates
The difference between the two disclosed rates at that volume is $60,600 a month, or $727,200 a year. That is the size of the decision a buyer is making when they treat a published cost per resolution as the whole story — and it is smaller than the swing produced by the definition differences in the previous section.
Why the definition swing is larger
Assume 40% of the agent’s conversations end in a configured handoff rather than an outright resolution. Under Salesforce’s terms those are free. Under Intercom’s they are billable outcomes. On 100,000 handled conversations that is 40,000 extra billable events at $0.99 — $39,600 a month appearing on one invoice and not the other, before either rate is compared.
How to Buy When the Cost Per Resolution Is Not Published
An unpublished cost per resolution is a negotiating position, not a dead end. The task is to extract the same six facts a published page would have given you, in writing, before signing.
Ask for the billing unit first
Before any number, get the unit defined in the contract: per conversation, per resolution, per seat, per minute, or a platform fee with consumption on top. A cost per resolution quoted without its definition is not a price, and the definition belongs in the agreement rather than the deck.
Get the billable-event list in writing
Ask the vendor to enumerate every conversation ending that generates a charge, using the six rows from the table above. Handoffs, abandonments, negative feedback and disqualifications are where the money hides. Salesforce and Intercom answer all of this publicly, which makes them a useful benchmark to hold a quieter vendor against.
Model your own traffic, not the vendor’s
Take last quarter’s real contact mix and apply each vendor’s definition to it. The output is your effective cost per resolution, which is the only figure worth comparing. Teams already running intelligent automation will usually have the ticket data to do this in an afternoon.
Price the floor and the ceiling
Establish the minimum commitment — Salesforce’s 1,000 resolutions and Intercom’s 50 a month differ by a factor of 20 — and ask what happens when volume falls. Then ask for a price-review clause: Gartner’s forecast is a direct argument for capping increases across a multi-year term.
What This Means for Enterprise Support Buyers
None of this says Cresta is expensive, and nothing here suggests its results are overstated. The containment and CSAT figures on its pages are specific, attributed to named customers, and more concrete than most of the category manages.
The finding is about disclosure, not value
The measurable claim is narrow: across 9,566 words of product and customer-story pages, Cresta publishes 27 outcome percentages, two revenue figures and zero prices, while two direct competitors publish a cost per resolution and a third publishes the billing model. That asymmetry is a fact about the buying process.
Where the real risk sits
The risk is not a hidden rate. It is signing an outcome-linked agreement whose definition of a billable outcome was never examined, in a market where Gartner expects the underlying cost per resolution to rise rather than fall. Both of those are addressable in procurement, and neither is visible on a product page.
A reasonable position
Shortlist on capability, then force pricing symmetry before the technical evaluation ends. Ask every vendor — including the ones that publish a rate — for the billable-event list, the minimum, and a price-review clause. Organisations building an autonomous AI agent capability should treat the definition of “resolved” as a contract term, because that is exactly what it is.
Frequently Asked Questions About Cost Per Resolution
What is a cost per resolution?
It is the price a vendor charges each time its AI agent resolves a customer conversation, replacing the per-seat licence with a unit that tracks work completed. The cost per resolution has become the default commercial model for enterprise support automation.
Does Cresta publish a cost per resolution?
No. Across the seven Cresta pages measured for this article there is no rate, no billing unit and no pricing page — cresta.com/pricing returns 404. Pricing is available only through the contact form.
What is Salesforce’s cost per resolution?
$2 per autonomous resolution for Agentforce Help Agent, with general availability in July 2026 and a 1,000-resolution minimum. There is no charge if the customer asks for a human, gives negative feedback, or abandons the conversation.
What does Intercom Fin charge?
From $0.99 per billable outcome, with a 50-outcome monthly minimum. Intercom’s outcome list includes resolutions, configured handoffs to a human, and disqualifications, and it charges once per conversation regardless of how many actions the agent takes.
Is the cheaper cost per resolution actually cheaper?
Not necessarily. Intercom bills configured handoffs and unanswered replies that Salesforce explicitly does not, so a contact centre with a high handoff rate can pay more at $0.99 than at $2.00. Apply each definition to your own traffic before comparing.
Will the cost per resolution fall over time?
Gartner forecasts the opposite: above $3 by 2030, exceeding many B2C offshore human agents, driven by data centre costs, vendor margin pressure and more complex use cases. Any multi-year business case should model a rising cost per resolution, not a falling one.
References and Further Reading
GenAI customer service costs to top offshore agents by 2030: Gartner
Salesforce unveils AI Help Agent with pay-per-resolution pricing
Salesforce Launches Agentforce Help Agent That Deploys in Minutes and Only Charges for Resolutions
Huge Agentforce Pricing Shift: Salesforce Introduces Pay-Per-Resolution
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.