What actually goes into a help desk bill beyond the sticker price?
The advertised price on nearly every help desk pricing page is a per-agent, per-month figure for one tier of one product. Your invoice is built from several layers stacked on top of it.
The first layer is seats. Most vendors count anyone who can open the inbox and reply, which quietly includes the people you did not think of as support: the owner who answers escalations at night, the bookkeeper who handles billing questions, the warehouse lead who checks shipping threads. Every one of those is another multiple of the headline price.
The second layer is AI usage. This is the part that has changed most, and the part most teams underestimate, because it is billed by outcome or by conversation rather than by person.
The third layer is channels and capability. Voice is almost always separate and often carries telephony charges on top of the software charge. Social and messaging channels are sometimes bundled, sometimes gated behind a higher tier. Advanced routing, custom roles, reporting, sandbox environments and CRM sync frequently sit on the plan above the one you were quoted.
The fourth layer is annual commitment and minimum seat counts. Monthly billing usually carries a premium over annual, and some plans will not sell you fewer than a set number of seats, so a three-person team can end up paying for more.
Add those four layers and the gap between the sticker price and the real monthly cost is rarely small.
How do AI resolution fees actually work?
There are four common metering models, and they behave very differently on the same ticket volume.
Per resolution. You pay each time the AI closes a conversation without a human. The appeal is obvious: you only pay for work that got done. The catch is the definition of a resolution, which is set by the vendor, not by you. A customer who asks a question, gets an answer and goes quiet may count as resolved. The same customer coming back the next day with a follow-up may count again.
Per conversation or per session. You pay when the AI engages, whether or not it helped. A session often has a time window attached, so a long back-and-forth can roll into a second billable session.
Per interaction or per message. The meter ticks on volume of messages, which punishes the chatty, multi-turn conversations that AI is actually good at handling.
Credit bundles. You prepurchase units that are consumed at different rates by different actions, with overage priced above the bundle rate. This is the hardest model to forecast, because the exchange rate between a credit and a real customer question is not something you can read off your own ticket log.
Under all four, the honest answer to "what will this cost me" is a range, not a number. That is a planning problem more than a pricing problem.
Why does usage-based AI pricing get more expensive exactly when it works?
This is the structural issue worth sitting with, because it is not a billing accident.
If you pay per resolution, then every improvement to your knowledge base increases the share of tickets the AI closes, which increases your bill. Write better documentation, and you pay more. Run a promotion that triples inbound questions, and you pay more during the month when margin is already tight. Grow the business, and the line item grows with it, with no ceiling you control.
Meanwhile the thing you were trying to buy was predictability. Support volume is seasonal and spiky by nature: a shipping delay, a price change, an outage at a supplier, a product going viral. Metered AI converts each of those spikes directly into an invoice spike, at the worst possible moment.
There is a second-order effect that matters even more for small teams. When automation has a per-unit cost, people start rationing it. They narrow the bot's scope, turn it off on some channels, or avoid pointing it at the busiest question types, specifically to keep the meter down. The tool gets used less precisely where it would help most, which defeats the purpose of buying it.
How do you calculate your own real monthly cost?
Work it out before you talk to a salesperson, so you can check their quote against your own arithmetic.
- Count every seat honestly. List every person who will need to read or reply, including part-timers, the owner, and anyone in ops or finance who touches customer threads. Then add the seats you will need in your busiest quarter, not your quietest one.
- Pull your real conversation volume. Take your inbound count from your current inbox or email for the last several months, not an average month. Note the peak, because the peak is what sets your worst invoice.
- Estimate the automation share. Be generous here, not conservative. A high automation rate is the goal, and under metered pricing it is also the expensive case, so model it as the expensive case.
- Apply each vendor's unit. Multiply your automated volume by their resolution, session or credit price, then add seats, then add the channels you need that sit outside the base plan.
- Model three scenarios. A quiet month, a normal month and a crisis month. The spread between them is your real exposure, and it is the number to take into the decision.
- Check what triggers a tier jump. Routing rules, roles and permissions, CRM sync and reporting are the usual culprits. If a feature you need lives one tier up, price the tier up.
Most teams who run this exercise find that the variable AI component, not the seat count, is the part of the bill they cannot predict.
What do the major help desk vendors charge for AI?
Pricing changes often, so treat this as a map of models rather than a price list, and verify current terms on each vendor's own page.
Zendesk and Freshdesk both follow the classic structure: per-agent, per-month tiers, with AI capability sold as an add-on or reserved for higher tiers, and automated resolutions metered separately. Intercom pairs per-seat pricing with a per-resolution charge for its AI agent, which makes it one of the clearest examples of the outcome-based model. Gorgias, built for ecommerce, prices around ticket and automated-resolution volume in tiers, so your bill steps up as you cross thresholds. Help Scout prices around users and contact volume with AI features layered on. Tidio bundles a set allowance of AI interactions per plan, with more available at higher tiers. Chatwoot is open source, so the software itself can be self-hosted at no license cost, but you take on hosting, maintenance and your own model API spend, which is a real cost even though it is not an invoice from Chatwoot.
Each of these is a legitimate design with real strengths. Intercom's product depth, Gorgias's ecommerce integrations, Freshdesk's breadth, Help Scout's simplicity and Chatwoot's control all solve genuine problems. The question is not which vendor is good, it is which billing shape matches a business where a bad month for support volume is also a bad month for cash.
What does a flat-rate help desk look like instead?
The alternative structure is to put every variable inside one fixed number. Corebee is priced at a flat $99 per month with unlimited seats and unlimited AI conversations, with no per-seat charge and no per-resolution charge, so the quiet month and the crisis month cost the same. You can see the full terms on the pricing page.
What changes practically is the set of decisions you no longer have to make. You can give the night-shift person and the bookkeeper their own logins without doing mental math on seats. You can point the AI at your whole knowledge base and let it answer everything it can, because a higher automation rate does not raise the bill. You can turn on WhatsApp alongside web chat, email, voice, SMS, Slack, Telegram, Facebook and Instagram without pricing each channel separately. The AI answers from your website and knowledge base, hands off to a human when the customer asks or when it is not confident, books meetings through Cal.com or Calendly, and syncs to HubSpot, Salesforce, Pipedrive, monday or Zoho. Setup takes roughly ten minutes and does not need a developer, and the shared inbox is where your team picks up whatever the AI passes along. If you are specifically comparing against the incumbent, the Zendesk alternatives breakdown walks the same cost structure side by side.
Flat pricing is not automatically the cheaper option for everyone. It is the more forecastable one, and for a small team those are often the same decision.
Which pricing model fits a small business best?
The honest answer depends on three things about your situation.
If your ticket volume is low and genuinely stable, and only one or two people will ever need access, a metered plan can come out cheaper in absolute terms. Small, flat volume is the case usage pricing serves well.
If your volume swings with seasons, launches or supplier issues, or if more than a handful of people need to touch customer conversations, fixed pricing tends to win on both cost and sanity. You stop treating automation as a cost center to be rationed, and the budget line stops being a thing you check nervously at month end.
If you have engineering capacity and strong opinions about data residency and customization, self-hosting an open-source platform is a serious option. Just price the engineering time and the model API spend into the comparison, because those are the real costs, and they are paid in attention as much as in money.
One last test before you sign anything: ask the vendor to put in writing what counts as a billable unit, when it resets, and what happens in a month when your volume doubles. A clear answer is a good sign. A vague one tells you the variable part of the bill is going to be a surprise, and surprises in a support budget are expensive in ways a spreadsheet does not capture.