Zendesk Switched to Per-Resolution Billing. Here's What That Actually Costs.
Something changed in January 2026 that most support teams didn't notice until the invoice came in. Zendesk quietly shifted its AI Agents from an included feature to a metered line item: you now pay per automated resolution, and overages bill automatically with no cap warning.
I work across both Zendesk and Help Scout depending on the team I'm supporting, and this pricing shift has come up in more conversations over the past few months than almost anything else in the support tooling space. So let me break down what it actually costs, where the model gets slippery, and how Help Scout is playing a different game.
What "Outcome-Based Pricing" Actually Means
The model is simple in theory. When Zendesk's AI Agent resolves a customer inquiry without escalating to a human, you pay for that resolution. The committed rate sits at roughly $1.50 per automated resolution. If you're on pay-as-you-go rather than a committed tier, it's closer to $2.00.
On top of that, Zendesk's Advanced AI features — the ones that make the AI Agent actually useful, like intelligent triage by intent and sentiment, admin copilot, and the voice AI agents that rolled out earlier this year — are sold as an add-on at around $50 per agent seat per month, on top of whichever Suite plan you're already paying for.
The per-resolution fee doesn't replace the seat fee. It layers on top of it. That's the part that catches teams off guard.
The Math at Real Ticket Volumes
Let me put some numbers around this, using estimates from teams in the mid-market range where Zendesk is most common.
Take a support team with 15 agents handling roughly 4,000 tickets per month. The AI deflects, say, 30% of those without human involvement — which is realistic for a well-trained bot on a common product with clear FAQs. That's about 1,200 automated resolutions per month.
- Advanced AI add-on: 15 agents × $50 = $750/month
- Automated resolutions: 1,200 × $1.50 = $1,800/month
- That's roughly $2,500/month before your base Suite plan
Now scale that. If your bot gets better and deflects 50%, you've just doubled the resolution line item to $3,600/month. In the old model, better bot performance meant you saved money on human agents. In this model, better bot performance means a higher bill — unless you've negotiated volume discounts upfront, which requires a sales conversation most teams don't have before they're already on the plan.
The pricing model's logic is that AI is doing real work and should be paid for doing it. The problem is that "doing real work" and "resolving the customer's actual issue" aren't always the same thing — and only one of those definitions benefits Zendesk.
Who Defines "Resolved"?
This is the question every Zendesk customer should be asking before committing to this model, and most aren't asking it loudly enough.
Zendesk's outcome detection logic defines a resolved conversation, not your CSAT score, not your first-contact resolution rate, not whether the customer came back with the same problem two hours later. If the bot closes the ticket under Zendesk's criteria, the resolution counts and the charge fires.
That matters because deflection rates and actual resolution quality diverge more than most teams expect. A bot can close a conversation by sending a help article link and waiting for the customer to go quiet — and that can count as a resolution under automated detection. Whether the customer was actually helped is a different question.
I'm not saying Zendesk's detection is bad, only that it's their detection, not yours. Before signing a committed tier, I'd want clarity in writing on exactly how a resolution is counted, especially for conversations that reopen, escalate late, or get low CSAT ratings after closure.
How Help Scout Is Playing This Differently
Help Scout has taken the opposite position on AI economics. Their AI Drafts feature — which generates suggested reply drafts for agents — is bundled into Plus and Pro plans at no extra charge. AI Assist, which helps agents refine and rewrite copy mid-ticket, ships with every paid plan. AI Answers, their customer-facing chatbot, is included in Beacon deployments.
The tradeoff is scope. Help Scout doesn't have workforce management tools, dedicated quality assurance scoring, or voice AI. If you're running a large operation that needs automated QA across 100% of interactions, staffing forecasting, or phone deflection, Help Scout's AI toolkit won't get you there — and Zendesk's will, at a price.
But if you're a team of 5 to 30 agents running email and chat support, Help Scout's model means your AI costs are fixed and predictable. You pay for seats, and the AI is part of what you're buying. That's a fundamentally different relationship with the product's economics.
Where This Leaves Support Leaders
The honest answer is that Zendesk's model makes sense for large teams with genuinely complex support needs and the volume to negotiate favorable per-resolution rates. If you're running thousands of automated resolutions per month and the cost-per-resolution is displacing agent hours you'd otherwise be paying for, the math can work in your favour.
The model starts to look more hostile when:
- Your bot performance improves and your bill scales with it rather than shrinking
- Your team is too small to qualify for volume discounts
- You didn't negotiate the resolution definition before signing
- You're relying on CSAT or reopens to measure quality but the billing system uses different signals
What's changed since January 2026 isn't just the pricing — it's the incentive structure. When AI is a fixed cost, you want it to be smart. When AI is a usage charge, you need to want it to be smart in exactly the ways that align with the vendor's measurement, not just your customers' experience.
What I'd Do Before Renewing
If you're on Zendesk and your contract is coming up for renewal in the next six months, three things are worth doing before you sign:
First, pull your last 90 days of AI resolution data and apply the committed rate to it. See what you'd have paid under this model versus what you actually paid. That gap tells you whether you're over or under your current allocation.
Second, ask your account manager to define, in writing, how a resolution is counted. Specifically: what happens to conversations that reopen within 24 hours, conversations that escalate after bot contact, and conversations with a CSAT of 1 or 2.
Third, set billing alerts. The January 2026 change introduced automatic overage billing with no in-app notification before the charge fires. You want to know when you're approaching your committed volume, not after you've exceeded it.
None of this is to say Zendesk is a bad product — it's powerful, mature, and if you're at scale it's often the right call. But the pricing shift this year changed the economics enough that running the numbers again isn't optional. The AI features that used to be a differentiator are now a cost centre that needs managing with the same attention you'd give any other line item.
If you're working through a support stack decision or trying to figure out whether to stay, switch, or renegotiate, I'm happy to think through it. The contact form is the best place to reach me.