The rest of this post gives you the arithmetic, the definitions to interrogate, and the questions to put to any vendor before you sign.
What does a vendor mean when it says a conversation was resolved?
The word resolved does a lot of quiet work on an invoice, and it rarely means the same thing from one platform to the next. Some count a resolution when the AI replies and the customer does not write back inside a set window. Others count it when the customer clicks a thumbs up, or when a ticket closes without a human ever opening it. A few count any session the assistant touched at all, including ones that ended by passing the customer to a person.
Those definitions are not equally generous to you. A silence-based definition bills you when a frustrated customer gives up and leaves. A touch-based definition bills you for the handoff you were trying to avoid in the first place. So before you compare two prices, get the definition in writing and push on the edge cases. Does a conversation that ends in a human handoff still bill? If the same customer asks the same question the next morning, is that one resolution or two? Does a greeting that goes nowhere count? Do the conversations your own staff start while testing count? Does a voice call bill differently from a chat?
The party that writes the definition is the party that benefits from a broad one, so it is fair to ask for worked examples rather than a single line of marketing copy. Then check the mechanics around it: are unused resolutions pooled across the month, do they roll over, what is the overage rate once you pass the bundle, and is there an annual commitment that locks the bundle size in before you know your real volume.
How do you work out what you are paying per resolution today?
You cannot judge a quoted price without a baseline, and most teams have never calculated theirs. Take one ordinary month, not your quietest and not your peak. Add up every line of your support stack for that month: seat licenses, AI add-ons, channel or messaging fees, the apps bolted onto the help desk, and anything billed per contact. That is your tooling cost.
Now count the conversations that genuinely ended with the customer's problem handled. Not tickets opened, not sessions started. Divide tooling cost by handled conversations and you have your current cost per resolution, measured honestly.
Then run it a second time with people included. Estimate the hours your team spent inside the inbox that month and apply their loaded hourly cost. The fully loaded figure is usually the one that matters, because the real question is never whether AI is cheaper than nothing; it is whether AI plus a smaller amount of human attention costs less than human attention alone, while leaving customers at least as well served.
Two more passes are worth the effort. Break the numbers down by channel, because a voice call and a one-line chat about shipping are not the same unit of work. And ask which direction each figure moves as you grow: a per-seat cost rises in steps when you hire, a per-resolution cost rises smoothly with demand, and a flat fee does not move at all. Pricing fairness is mostly a question about slope, not about today's number.
What makes a per-resolution price fair, and what makes it unfair?
Five tests cover most of it.
Predictability. If you cannot state next month's bill within a narrow range before the month starts, the price is not fair to a business that plans cash flow weekly. Variable billing is tolerable when it tracks revenue; support volume often spikes exactly when something went wrong and revenue did not.
Proportion to value. A resolution that saves a customer relationship is worth far more than one answering a question about opening hours, yet most meters charge the same for both. A fair structure either prices well below the value of the cheapest useful resolution, or stops metering entirely.
An honest definition. Covered above, and it is the single place where a headline price can mislead most.
No penalty for doing well. If a good campaign, a seasonal rush, or a product launch triples your inbound and triples your bill, the pricing is taxing your success. Look for a cap, a declining unit rate at volume, or a flat fee.
A clean exit. Month to month billing, an export of your conversation history and your knowledge base content, and no clause that raises your rate once you depend on the tool. Portability is part of price, because a price you cannot walk away from is not really a price you agreed to.
How does usage-based pricing change the way your team works?
This is the cost that never appears on the invoice. When every answer carries a meter, teams start rationing. They tuck the chat widget onto a single page instead of the whole site. They narrow the assistant's scope so it declines more questions. They hesitate before adding WhatsApp or SMS, because each new channel means more billable conversations, even though those are the channels customers actually prefer. They avoid proactive messages that would have caught a problem early.
Each of those decisions is rational under the pricing model and bad for the business. Support deflection is not the goal; a customer who gets a correct answer quickly, on the channel they chose, is the goal. Pricing that makes you flinch at your own inbound traffic is quietly steering you away from that. It is worth asking, honestly, whether your team would turn the assistant on across every channel if the marginal cost were zero. If the answer is yes, the meter is already shaping your service.
How do the main help desk platforms approach this?
Broad strokes, because published plans change and you should read the current page before deciding.
Zendesk and Intercom both sit at the mature end of the market, with deep reporting, large app ecosystems, and enterprise features. Both combine a per-seat subscription with a separate charge tied to AI-handled conversations, so your bill grows along two axes at once. Freshdesk follows a comparable per-agent model with AI capability layered on. Gorgias is built around ecommerce workflows and prices in tiers tied to ticket volume, which suits stores with predictable seasonality. Help Scout leans toward simple, human-feeling email support and has priced around contacts handled. Tidio targets smaller sites with conversation-based tiers and an AI add-on. Chatwoot is open source, so the licence can cost nothing while you pay in servers, upgrades, and the engineering hours to run it, which is a real cost even when it never appears on a card statement.
None of these is a bad product. They are different shapes, and the shape is what you are choosing. If you want a longer comparison of the per-seat model against alternatives, see the Zendesk alternatives breakdown.
When is a flat monthly fee the better shape?
A flat fee wins when your volume is uncertain, when you want every teammate in the inbox without counting badges, and when you want to open more channels rather than fewer. It converts an unpredictable variable into a fixed line item, and it removes the internal argument about whether a given conversation was worth answering.
Corebee is priced this way: a flat $99 per month, with unlimited seats and unlimited AI conversations, and no per-seat or per-resolution fees. The practical effect is that your cost per resolved conversation falls every time volume rises, which is the opposite slope from a meter. Everything arrives in one shared inbox, the assistant answers from your website and your knowledge base, and it hands off to a person when the customer asks or when it is not confident. Web chat, email, voice with an AI receptionist that tells callers it is an AI, WhatsApp, Slack, Facebook, Instagram, SMS, and Telegram are all live, and it can book meetings through Cal.com or Calendly and sync to HubSpot, Salesforce, Pipedrive, monday, or Zoho. Setup takes about ten minutes and needs no developer. The pricing page has the full detail.
Flat pricing is not automatically right for everyone. If you have very low volume and a single part-time agent, a cheap entry tier elsewhere may cost less. If you need a specific enterprise workflow engine or a deep custom integration, buy the platform that has it. The point is to pick the shape that matches how your volume behaves.
What should you check before you sign anything?
Ask for the resolution definition with examples. Ask what happens at overage and whether there is a cap. Ask whether the bundle resets monthly or rolls over. Ask what a human handoff costs. Ask whether adding a channel adds a charge. Ask how seats are counted when a part-time person logs in twice a month. Ask what you can export if you leave and in what format. Then run your own baseline from the method above and compare the fully loaded figures, not the headline ones.
Finally, run a real trial on your real traffic rather than a scripted demo. Watch what the assistant does with your awkward questions, your refund policy, and your out-of-hours calls. A price is only fair relative to the quality of the work, and the only way to judge that is on your own inbox.