The "We Only Charge When It Works" Positioning War in Agentic AI

Table of Contents

• Why "Only When It Works" Became the Default Pitch
• What "Works" Actually Means (and Who Gets to Say)
• The Economics Hiding Behind the Slogan
• The Positioning War: How Vendors Are Weaponizing the Phrase
• Where the Promise Quietly Breaks Down
• How Buyers Should Pressure-Test the Claim
• What This Means for the Broader GaaS Pricing Landscape
• Insights Most People Overlook
• Frequently Asked Questions
• Conclusion
• References

Why "Only When It Works" Became the Default Pitch

A few years ago, the standard objection to buying software was "will my team actually use it?" With agentic AI, the objection mutated into something sharper: "will this thing actually do the job, or will it confidently hand me garbage?" That single shift in buyer anxiety explains almost everything about why "we only charge when it works" spread so fast.

Seat-based and even pure usage-based pricing both put the risk of failure on the buyer. You pay for the license, or you pay for the tokens consumed, and whether the agent resolved the ticket or closed the candidate or reconciled the invoice is your problem. Outcome-aligned pricing flips that. The vendor is, at least rhetorically, putting skin in the game. For a category still fighting a credibility deficit, that posture is worth more than any feature list.

The clearest live example is customer support. Intercom's Fin agent popularized per-resolution pricing, charging only when the AI actually closes a conversation without human handoff. That model did something subtle and powerful: it turned a pricing page into a confidence signal. The implicit message is, "we're so sure this works that we'll only bill you when it does." Once one credible vendor plants that flag, everyone selling against them has to respond, which is how a pricing experiment becomes a positioning war. This dynamic connects directly to the per-resolution support playbook and the broader debate over outcome-based pricing that other nodes in this cluster examine in depth.

It also maps neatly onto how buyers now think about AI spend. Procurement and finance teams have been burned by usage bills that ballooned without obvious value, so "pay only for results" lands as the antidote to metered-billing anxiety. The slogan isn't just a price model. It's a story about who carries the risk.

What "Works" Actually Means (and Who Gets to Say)

Here's where the marketing gloss meets reality, and it gets messy fast. "Works" is not a fact. It's a definition, and definitions are negotiated.

Consider a support agent. Does "works" mean the conversation ended without a human? That the customer didn't reopen the ticket within 24 hours? That the customer was actually satisfied, not just exhausted into closing the chat? Each definition produces wildly different bills. A vendor that counts a resolution the moment the agent sends a plausible-looking reply has every incentive to declare victory early. A buyer who only considers it resolved when the underlying problem is genuinely solved is measuring something else entirely.

This is the audit problem at the heart of outcome pricing, and it's why "who defines and audits the outcome" is arguably the most important question in all of GaaS monetization. The party that controls the definition controls the invoice. When the vendor both performs the work and grades its own homework, the buyer is trusting a counterparty whose revenue depends on a generous interpretation of "success."

Three patterns have emerged for handling this:

• Vendor-defined, vendor-measured. Fastest to ship, weakest on trust. Fine for low-stakes, high-volume tasks where occasional misclassification washes out statistically.
• Mutually defined, vendor-measured. The contract spells out success criteria, but the vendor's telemetry reports the numbers. Most common in mid-market deals.
• Mutually defined, independently auditable. Success is logged in a way the buyer can verify against their own systems (CRM ticket status, deal-stage changes, reconciled ledger entries). This is the gold standard, and it's rare because it's hard to build.

The vendors winning the long game are quietly investing in that third pattern, because a slogan that survives a buyer's audit is a slogan that renews. A slogan that collapses under scrutiny becomes a churn driver.

The Economics Hiding Behind the Slogan

"We only charge when it works" sounds like the vendor is being generous. Often, they're being strategic about margin in ways the buyer never sees.

Every agent run costs the vendor real money in inference, whether it succeeds or fails. An agent that fails a task still consumed tokens, possibly a lot of them if it looped, retried, or wandered down a reasoning dead-end. So "free when it fails" means the vendor is absorbing the cost of every failure. That's only sustainable if two things are true: the success rate is high enough, and the price on successes is fat enough to cover the failures plus margin.

Run the math. If an agent succeeds 70% of the time, and each attempt (win or lose) costs the vendor a dollar in compute, then every successful outcome carries the cost of roughly 1.43 attempts. The vendor must price that success above $1.43 just to break even on compute, before any overhead or profit. The "free failures" the buyer celebrates are baked into the price of the wins. This is the same margin-safety logic that governs passing through volatile inference costs and the search for a minimum viable margin, both treated elsewhere in this beat.

This produces a counterintuitive result: outcome pricing structurally favors vendors with high success rates, which usually means vendors with more proprietary data, better fine-tuning, and tighter workflow scoping. A scrappy newcomer with a 50% success rate cannot offer the same "only when it works" deal without bleeding cash, because half their compute spend earns nothing. The slogan, in other words, is partly a moat, it's easiest to promise for the incumbent who's already good. That's why outcome pricing quietly favors incumbents with data, a tension worth sitting with.

It also reshapes the vendor's internal incentives in a healthy direction. When you only get paid for successes, you obsess over reliability, model routing to cut the cost of each attempt, and knowing when to gracefully decline a task you'll probably fail rather than burn money attempting it. Pricing becomes a forcing function for engineering discipline.

The Positioning War: How Vendors Are Weaponizing the Phrase

The phrase itself has become contested territory, and vendors are deploying it in at least three distinct ways.

As a wedge against per-seat incumbents. Newer agent vendors use "you don't pay unless it works" to attack legacy SaaS pricing as a relic. The pitch writes itself: "Why are you paying $X per seat per month for software your team might not even use, when you could pay only for outcomes delivered?" It's a sharp contrast that makes the incumbent look greedy and the challenger look confident.

As a trust accelerant in the sales cycle. For high-consideration purchases, the slogan shortens the trust-building phase. A buyer nervous about agent reliability can be told, "then you literally lose nothing if it doesn't work." This collapses a long evaluation into a low-risk trial, which is why outcome framing changes the sales motion entirely. The catch is that "lose nothing" is rarely literally true, there's still integration cost, opportunity cost, and the risk of a bad customer experience when the agent fails.

As a defensive reframe by vendors who can't actually do it. This is the subtle one. Some vendors who structurally can't offer pure outcome pricing, because their margins or success rates won't support it, adopt the language while building escape hatches into the fine print. "We only charge when it works" becomes "we only charge when our telemetry records a success event," and the success event is defined loosely enough that almost everything qualifies. Watching how a vendor defines the billable event tells you whether the slogan is a promise or a costume.

The war is being fought on pricing pages, in sales decks, and increasingly in the fine print of contracts. As more vendors crowd into the same positioning, the differentiation moves from whether you charge only for outcomes to how honestly and verifiably you define them. Honesty becomes the new battleground.

Where the Promise Quietly Breaks Down

No pricing model is free of failure modes, and "only when it works" has several that buyers discover late.

Partial completion. Real work isn't binary. An agent that completes 80% of a multi-step task and stalls on the last step did something valuable, but is that a "success"? If the vendor bills nothing, they're undercompensated and may quietly stop attempting hard tasks. If they bill in full, the buyer feels cheated. The honest answer involves graceful degradation and partial-completion pricing, which most slogans paper over entirely.

Gaming the success metric. Whatever you measure, the agent optimizes toward. If "resolution" means "conversation closed," an agent can learn to close conversations prematurely. If "qualified lead" triggers a fee, the bar for "qualified" drifts downward. The metric becomes the target, and Goodhart's law does the rest.

The disputed invoice. When success is ambiguous, billing disputes follow. Buyer and vendor look at the same interaction and disagree about whether it "worked." Without clean, mutually trusted audit trails, these disputes poison the relationship and consume the exact account-management hours the model was supposed to eliminate.

Volatile, unpredictable bills. Ironically, "pay only for outcomes" can produce less budget predictability than a flat fee. If the agent suddenly gets better, your bill spikes precisely because it's working, automatic usage growth that finance didn't forecast. This is the annual-contract problem that haunts unpredictable usage, and it's why floor-and-ceiling structures and usage caps exist to protect both sides.

SLA and refund ambiguity. When an agent fails a task the buyer was depending on, "you weren't charged" is cold comfort if the failure caused real downstream damage. Mature contracts address refunds and SLAs when an agent fails, going beyond the simplistic "no charge" to define remediation, escalation, and liability.

How Buyers Should Pressure-Test the Claim

If you're evaluating a vendor leading with "we only charge when it works," a handful of questions separate substance from slogan:

• Define "works" in writing. Ask for the exact billable event, in contract language, with edge cases enumerated. If they can't or won't, that's your answer.
• Who measures it? Vendor telemetry, your systems, or a neutral log both sides can audit? Push toward verifiability.
• What happens on partial completion? A vendor who has thought hard about this will have a crisp, fair answer. A vendor who hasn't will improvise.
• Model the bill at scale. Get pricing per outcome, multiply by your real volume, and compare to flat alternatives. "Only when it works" can be more expensive than a subscription once it works a lot.
• Cap the downside. Negotiate a ceiling so a sudden spike in agent effectiveness doesn't blow your budget. Pair it with a floor the vendor needs to stay viable.
• Read the failure clause. What do you get, beyond a zero charge, when the agent fails something that mattered?

A vendor confident in their reliability welcomes these questions. A vendor relying on the slogan as cover gets evasive. That reaction is itself diagnostic.

What This Means for the Broader GaaS Pricing Landscape

"We only charge when it works" is best understood not as a pricing model but as a positioning layer that sits on top of several underlying models, per-outcome, per-resolution, hybrid, success-fee. It's the marketing skin over the machinery, and it belongs to the larger taxonomy of how agentic services get monetized.

The trajectory is fairly clear. As the category matures, the naked slogan loses power because everyone uses it. Differentiation shifts to verifiability, fairness on partial outcomes, and budget protection. The vendors who treated "only when it works" as a genuine engineering and contractual commitment, building real audit trails, honest partial-completion handling, and sane caps, will keep the trust they earned. The ones who used it as a costume will face disputes, churn, and a buyer base that learned to read the fine print.

For anyone building or buying in this space, the lesson is to treat the phrase as the beginning of a conversation, not the end of one. The interesting questions all live downstream of the slogan, in the same territory the rest of this pricing-and-monetization cluster maps: how outcomes are defined and audited, how margins survive volatile inference, and how the whole landscape is likely to consolidate over the next couple of years.

Insights Most People Overlook

The slogan is a confession about success rates. A vendor who can comfortably offer "only when it works" is implicitly telling you their reliability is high, otherwise the model bankrupts them. Conversely, a vendor offering it with suspiciously generous "success" definitions may be hiding a low success rate behind a loose metric. The pricing model is a leak of information about the product quality.

Outcome pricing can punish the buyer for the vendor's improvement. Most buyers assume "pay for results" aligns incentives perfectly. But when the vendor ships a better model, your bill goes up through no decision of your own, the agent simply succeeds more often. The buyer who didn't negotiate a ceiling effectively signed an open-ended commitment to fund the vendor's R&D wins. Alignment cuts both ways, and the buyer's side is often overlooked.

"Free failures" are never free, they're prepaid in the success price. Buyers celebrate not paying for failed attempts, not realizing the cost of those failures is amortized into the price of every success. You're paying for the failures; you just can't see the line item. A vendor with a 95% success rate can price successes far cheaper than one at 60%, which means the cheapest outcome pricing comes from the most reliable vendor, the opposite of how buyers usually assume discounts work.

The real moat isn't the model, it's the audit trail. Anyone can copy the slogan. What's genuinely hard to replicate is an outcome-measurement system both parties trust. The vendor who builds verifiable, tamper-evident success logging owns something competitors can't slap onto a pricing page overnight. In a war fought on positioning, the durable advantage is infrastructure, not language.

Pure outcome pricing may be a transitional phase, not the endpoint. The current enthusiasm assumes outcome pricing is where the category lands. More likely, it's a credibility-building bridge. Once buyers trust that agents work, hybrid models, a base fee for availability plus outcome fees for results, will reassert themselves, because vendors need predictable revenue and buyers eventually want predictable bills. "Only when it works" may be how the category earns trust, not how it ultimately prices.

Frequently Asked Questions

Is "we only charge when it works" the same as outcome-based pricing? Not quite. Outcome-based pricing is the underlying mechanism, you pay per result. "Only when it works" is the marketing positioning layered on top, emphasizing that failures cost nothing. A vendor can use outcome pricing without leading with the slogan, and some use the slogan loosely over models that aren't truly outcome-pure.

How do I know if a vendor's "success" definition is fair? Get it in writing, check who measures it, and enumerate edge cases, especially partial completion and reopened/reversed outcomes. If the definition is vendor-controlled, vendor-measured, and vague, treat the slogan as marketing rather than a guarantee.

Does outcome pricing always cost less than a subscription? No. It costs less when the agent rarely succeeds, and potentially much more when it succeeds constantly. Always model your real volume against the per-outcome price before assuming it's cheaper. A highly effective agent under outcome pricing can exceed a flat subscription.

Why don't all agent vendors offer "only when it works"? Because it requires a high enough success rate and fat enough margins to absorb the cost of every failed attempt. Vendors with lower reliability or thin margins would lose money offering it, which is why the model tends to favor data-rich incumbents.

What's the biggest risk for buyers in this model? Two: an unbounded bill if the agent suddenly gets much better, and a metric-gaming agent that "succeeds" on paper while delivering poor real outcomes. Negotiate a ceiling and insist on outcome definitions tied to genuine results, not proxy events the agent can game.

How does partial completion get handled? Poorly, in most cases, the slogan assumes binary success. Mature vendors define partial-completion pricing and graceful degradation explicitly. If a vendor has no clear answer, expect billing disputes on every multi-step task that stalls.

Conclusion

"We only charge when it works" earned its place as the defining slogan of agentic AI pricing because it answers the category's core anxiety: can I trust this agent to actually do the job? As a confidence signal and a wedge against per-seat incumbents, it's genuinely effective. But the phrase is the start of the real conversation, not the end. "Works" is a negotiated definition, "free failures" are prepaid in the price of successes, and the model quietly favors reliable, data-rich vendors while exposing buyers to volatile bills and metric-gaming. The vendors who win the long game treat the slogan as a contractual and engineering commitment, with verifiable audit trails, honest partial-completion handling, and sane caps, rather than a costume. For buyers, the takeaway is simple: admire the confidence, then read the fine print. Everything interesting about agentic AI monetization lives in how "works" gets defined, measured, and disputed, the same questions that run through the rest of this pricing cluster.

References

• Intercom Fin: Resolution-Based Pricing, Intercom's per-resolution model for AI customer support agents.
• a16z: The Economics of Outcome-Based AI Pricing, Andreessen Horowitz analysis of how outcome and consumption pricing reshape software economics.
• HBR: How to Price Your AI Products, Harvard Business Review on aligning AI pricing with delivered value and managing buyer risk.
• Gartner: Emerging Pricing Models for Generative AI, Gartner research on consumption, outcome, and hybrid pricing for AI services.