Helpdesk software: are you paying per person, or per conversation?

Helpdesk pricing looks like ordinary per-seat software until you notice that half the category is not per-seat at all. Help Scout charges by the agent: every additional person answering costs another full monthly rate, however many or few customers write in. Tidio charges by the conversation: its plans carry a number of billable conversations, and adding staff does not by itself raise the bill. Those are not variations on a theme, they are opposite meters, and which one suits you depends entirely on a ratio you can measure before you buy.

Work it through and the divergence is immediate. On a per-agent plan, a five-person support team pays five times what one person pays, regardless of whether the queue is busy. On a per-conversation plan, those same five people cost exactly what one person would, and the bill only moves when the volume does. A small team absorbing high volume is punished by the second model and rewarded by the first; a large team handling a light queue is the reverse. Neither vendor is expensive or cheap in the abstract, and any table ranking them by monthly price is answering a question that has no general answer.

The practical implication is that you should count two things before comparing anything: how many people will hold a licence, and how many conversations you handle in a typical month. Both numbers are usually known or knowable, and together they decide the category for you. It is worth doing at your expected size in a year rather than today's, because the two models diverge faster as either number grows.

A third meter has appeared on top of both, and it prices something new. One vendor charges by the AI resolution — a fee each time an automated answer resolves a customer's query, on top of the per-agent subscription. That is a usage meter attached to an outcome rather than to a seat or a message, and it inverts the usual incentive in an interesting way: the tool costs you more precisely when it works. Whether that is good value depends on what an avoided human reply is worth to you, which is a calculation only you can do, and it is not one the pricing page attempts.

The billing unit also determines what happens when things go wrong in a busy month. A per-conversation plan with a spike in volume produces a larger bill, and a per-agent plan produces the same bill with a longer queue. Neither is obviously better and they fail differently: one costs money, the other costs response time. Deciding in advance which of those you would rather absorb is more useful than comparing headline rates.

Two smaller notes, because this category has been unusually well-behaved on the things this series usually finds fault with. Both of the vendors that publish an annual discount describe it accurately: one prints a two-months-free claim that computes to a shade under sixteen and two-thirds per cent, and the other labels the identical arrangement as sixteen per cent when the arithmetic gives sixteen and two-thirds — rounding against itself rather than in its own favour. After several vendors elsewhere in these guides whose claims did not survive a division, that is worth recording.

Less helpfully, several plans in the category are labelled as starting prices rather than fixed ones, with what varies above the floor unstated. Where that is so the published figure is the least you could pay rather than the price of a defined configuration, and it cannot be compared like-for-like with a fixed rate from another vendor. One well-known vendor in the category has no captured pricing with us at all, so it is absent from our comparisons rather than represented by an estimate.

So the sequence is short: count agents, count conversations, and price both models at those numbers. Then decide separately whether an outcome-priced AI tier is worth having, and read any 'starts at' as a floor rather than a price. In a category where two vendors can be right for opposite customers, the arithmetic on your own two numbers is worth more than any ranking.