Per-Resolution Is the New Per-Seat: How Intercom Fin, Zendesk AI, and Agentforce Changed the Pricing Rules

Atlassian's most recent earnings disclosures contain a number that should worry every enterprise buyer who has clicked "yes" on a Rovo AI prompt without checking what it costs: Rovo credit consumption is growing more than 20% month over month, and customers who actively use it are growing their ARR roughly twice as fast as customers who don't. Atlassian has not yet sent a single overage invoice for that usage. The company's own support documentation confirms it: requests that exceed a customer's monthly Rovo credit allowance still go through, no charge applied, no hard stop.

That is not an oversight. It is a pricing strategy, and it is one of the more interesting ones running in enterprise software right now, because Atlassian has explicitly reserved the right to end it. Atlassian's billing documentation states that before any consumption-based overage charge becomes real, the company will give customers at least 90 days' notice and require an explicit opt-in — no automatic billing, no surprise invoice. That single sentence is doing a lot of work. It is simultaneously a customer-trust commitment and a soft countdown clock on free AI.

What Atlassian is running, whether it intended the framing or not, is one of the largest live experiments in trust-based AI pricing in enterprise software: give away the meter, let usage patterns mature, build the product into renewal-driving workflows, and only then decide what the meter actually costs. Understanding how the credit system works — and what changes when the free period ends — matters for any enterprise buyer evaluating an AI add-on with a similar structure, because Atlassian's playbook is becoming the template, not the exception.

How the Rovo Credit System Actually Works

Rovo is Atlassian's AI layer across Jira, Confluence, and Jira Service Management — a chat interface, a set of AI agents, and a retrieval system that searches across an organization's accumulated work. It rolled out to Premium and Enterprise plans starting in April 2025 and reached Standard-plan customers by October 2025, which means by mid-2026 it has had well over a year to accumulate real usage data across Atlassian's installed base of more than 300,000 customers.

The credit system is what makes Rovo's pricing structurally different from a flat AI add-on fee. According to Atlassian's own usage-allowance documentation, each licensed Standard-plan user receives a monthly allowance of 400 Rovo credits and 1,500 indexed objects. Those credits get consumed at different rates depending on what the AI does:

Credits pool at the organization level rather than resetting per individual, which means a single team running heavy Deep Research workloads can draw down the shared allowance faster than the average user in the same org would suggest. Credits also do not roll over — whatever isn't used by the end of the billing cycle disappears, which is a familiar shape to anyone who has dealt with mobile data plans or cloud compute commitments.

Teamwork Collection customers — the bundle that pairs Jira, Confluence, and Loom-style collaboration tools — get a meaningfully larger allowance: Atlassian has said these customers receive roughly 10 times the credits of standalone subscribers and consume more than twice as many credits per user as a result. That's a deliberate bundling incentive: the more of Atlassian's suite a customer buys, the cheaper AI usage becomes per unit, which pushes upsell conversations toward the bundle rather than the point product.

Rovo Dev — Atlassian's separate AI coding agent, distinct from the Jira/Confluence Rovo experience — runs on its own pricing track entirely: $20 per developer per month for an allowance of 2,000 Rovo Dev credits, with metered overage already live at $0.01 per credit beyond that. Rovo Dev is the control group in this story. It shows that Atlassian is fully capable of shipping straightforward consumption billing when it wants to — which makes the decision to hold off on core Rovo overage charges look more like patience than technical limitation.

The 90-Day Promise: What Happens When the Meter Turns On

The most consequential sentence in Atlassian's Rovo documentation isn't about how credits work today — it's about how billing will change tomorrow. Atlassian has committed that no consumption-based overage charge will be introduced without at least 90 days' advance notice and an explicit customer opt-in. That commitment is unusually specific for a SaaS vendor, and it's worth asking why Atlassian made it.

The straightforward answer is trust-building during a sensitive product transition. Enterprise buyers have been burned before by AI pricing surprises — GitHub Copilot's switch to token-based billing on June 1, 2026 caught agentic coding teams off guard with monthly cost increases of 10x to 50x for heavy users, and that backlash is fresh in every procurement team's memory. Atlassian's 90-day, opt-in commitment is a direct answer to that fear: it tells buyers they will see the bill coming and have to actively agree to it, rather than discovering it on a renewal invoice.

But the commitment also functions as a controlled fuse. Once Atlassian flips the switch on overage billing — and the language "before that changes" in its own documentation makes clear that it considers this a matter of when, not if — the 90-day countdown starts, and procurement teams that haven't been tracking their organization's credit consumption will be negotiating from a position of total surprise about what their actual usage costs. The notice period buys time to react; it does not buy time to have already built the internal visibility needed to react well.

This is the structural tension at the center of every grace-period pricing model: the vendor benefits from maximum usage during the free period, because usage drives the product attachment that justifies a renewal at a higher tier. The customer benefits from minimum usage during the free period, because every credit consumed today is a data point the vendor will eventually use to calibrate a price the customer will have to pay. Those two incentives point in opposite directions, and most enterprise buyers aren't tracking Rovo credit consumption closely enough to know which side of that tension their organization is actually on.

A Different Model Than Per-Resolution Pricing

Atlassian's credit-pool approach is a genuinely different mechanism than the per-resolution pricing that Intercom Fin, Zendesk AI Agents, and Salesforce Agentforce have adopted in customer support AI, and the difference matters for how buyers should think about risk.

Per-resolution pricing charges for a defined outcome: a support ticket closed without human escalation. The vendor and customer both know, contract to contract, what a "resolution" costs and roughly how many will occur based on ticket volume. It is volatile in aggregate (resolution rates fluctuate) but legible at the unit level (each resolution has one price).

Rovo's credit-pool model charges for a bundle of heterogeneous actions — a quick chat question and a Deep Research query consume wildly different amounts of the same shared pool, and the organization, not the individual interaction, is the unit being metered. That makes the system harder to forecast at the team level (a single analyst running ten Deep Research queries in an afternoon can meaningfully dent a department's monthly allowance) but easier to sell at the org level, because "you get 400 credits per seat" sounds like a comprehensible allowance in the way "$0.99 per resolved support ticket" sounds like a comprehensible unit price — even though both abstract away significant variance underneath.

The credit-pool model is also closer in spirit to Stripe's usage-based billing, which research has shown creates predictable retention cliffs when customers hit usage thresholds they didn't anticipate. The difference is that Stripe's usage tiers are billed from day one; Atlassian's are not billed at all yet. That means the retention-cliff risk Stripe customers face today is a risk Rovo customers are accumulating but haven't yet experienced — which is precisely why the 90-day notice period matters so much. It is the only mechanism standing between Atlassian's current customers and an unmodeled cliff.

The Numbers Behind the Bet

Atlassian's decision to delay overage billing is backed by financial results strong enough to make patience look like the obviously correct call, at least so far. In its most recent fiscal quarter, the company reported:

Those numbers tell a clear story: giving away the AI meter is working as a growth lever. Customers who engage with Rovo expand their accounts faster, and the NRR figure — comfortably above the 110-120% range that defines "best-in-class" for B2B SaaS — suggests the free-usage period is doing exactly what it's designed to do: drive deeper product attachment that converts into expansion revenue at renewal, rather than direct AI consumption revenue today.

This is also a meaningfully different position than the one facing pure AI-native SaaS companies, whose median net revenue retention sits closer to 48% — less than half of Atlassian's. Atlassian isn't selling AI as the product; it's selling AI as a retention and expansion mechanism bolted onto an already-sticky collaboration suite. That distinction is exactly why the company can afford to defer monetizing usage directly: the AI doesn't need to pay for itself in credit revenue if it's already paying for itself in seat expansion and tier upgrades.

The Enterprise Buyer's Playbook for the Grace Period

Enterprise procurement and IT teams currently running Rovo without paying for usage have a finite, valuable window to prepare for the day the meter turns on. The playbook is straightforward but requires acting now, not when the 90-day notice arrives.

1. Build a credit-usage dashboard today, not when notified. Atlassian's admin console exposes Rovo credit consumption by user and by action type. Most IT teams aren't looking at it because there's no bill attached to the number yet. That's exactly the wrong instinct — the data is most useful before it has financial consequences, because it lets you identify which teams and workflows are driving consumption while you still have time to shape behavior without a budget fight attached.

2. Identify your Deep Research power users now. Because a single Deep Research query can cost 10x what a chat question costs, a small number of heavy users running multi-step retrieval queries can dominate an organization's credit consumption. Find out who they are and what they're using it for — if it's high-value work, that's useful information for negotiating a future tier; if it's exploratory or redundant, that's an easy place to cut consumption before it becomes a line item.

3. Push for usage-based contract language at your next renewal, before the notice period starts. Rather than waiting for Atlassian's 90-day notice to negotiate reactively, ask now for contractual protections: a cap on per-credit overage pricing, a grace tier before consumption billing applies, and guaranteed visibility into organization-wide usage trends. Vendors are far more willing to grant these terms during a renewal negotiation than during a 90-day countdown where the customer has no leverage and a hard deadline.

4. Model what 20%+ monthly growth in your own usage actually means. Atlassian's company-wide credit growth rate of 20%+ month over month is a useful proxy, but every organization's adoption curve is different. If your Rovo usage is compounding at even half that rate, a credit allowance that comfortably covers your team today may be exhausted within two or three quarters — well before any external notice period even starts the clock.

5. Treat the free period as a negotiating asset, not a free lunch. The fact that Atlassian is currently absorbing the cost of your organization's AI usage is leverage. Use the data from your usage dashboard to demonstrate adoption and value internally — to your own leadership, and eventually to Atlassian — so that when consumption pricing does arrive, your organization is negotiating from a position of demonstrated, well-documented value rather than from surprise.

What This Means for Every SaaS Vendor Building Consumption AI Pricing

Atlassian's approach offers a template that other incumbent SaaS vendors bolting AI onto existing seat-based products are likely to copy, because the economics of the approach are sound for vendors in Atlassian's position specifically: a large existing customer base, high switching costs from the core product, and an AI feature that drives expansion revenue even before it drives direct usage revenue.

The model doesn't generalize to every vendor, though. It works because Atlassian's core Jira/Confluence/JSM products already have entrenched usage and high net revenue retention — the AI feature is additive to a relationship that was already strong. A vendor without that foundation, giving away AI usage in hopes of building product attachment from scratch, is making a much riskier bet: there's no existing seat-expansion engine for the AI feature to plug into, so the free period is pure cost with a much less certain payoff.

For vendors that do have Atlassian's structural advantages, the lessons worth copying are specific: publish a credit-allowance system rather than an opaque "fair use" policy, commit publicly to an advance-notice period before turning on metering, and make sure the usage data is visible to customers in real time rather than locked in an internal dashboard customers never see. Each of those choices reduces the trust cost of eventually monetizing usage, which is the entire point of running a grace period in the first place — it only works if customers believe the eventual bill will be fair, predictable, and not retroactive.

The Risk Nobody's Pricing In

The scenario that should worry both Atlassian and its customers is not a small, manageable overage bill. It's the possibility that 20%+ monthly credit growth, sustained across a base of 300,000+ customers, eventually forces Atlassian to set an overage price calibrated to its own margin requirements rather than to what any individual customer's usage pattern can absorb. LLM inference costs do not stay flat as usage scales into Deep Research-heavy workflows that synthesize across an organization's entire content graph — and Atlassian's margin requirements on that compute are not currently visible to any customer modeling their own future bill.

That asymmetry is the real risk hiding inside the grace period: customers are accumulating usage habits based on a price of zero, while Atlassian is accumulating the cost data it will eventually use to set a price that has nothing to do with zero. The 90-day notice period guarantees customers will see that number coming. It does not guarantee they will like it, or that their workflows, built and scaled around free usage, will still make financial sense once a real price is attached to them.

Takeaway: Atlassian's Rovo credit system is a masterclass in sequencing trust before monetization — give away the meter, let the product earn its place in daily workflows, and only then decide what usage actually costs. The 90-day, opt-in overage commitment is a genuine customer protection, but it is also confirmation that a real price is coming. Enterprise buyers who treat the current free period as a permanent feature rather than a calibration window are setting themselves up for the exact kind of pricing shock that hit GitHub Copilot customers in June. The teams that come out ahead are the ones building usage visibility now, while the data has no price tag attached to it yet.