Comparing prices when nobody sells the same unit

A monthly figure is only a price once you know what quantity it buys, and in this category the quantity is defined differently by every vendor. That makes the usual comparison, lining up three plans side by side and looking at the numbers, close to meaningless. You are comparing the price of a kilo with the price of a litre.

This is not necessarily anybody being sneaky. The meters correspond to different real costs and different views of what the product is. A vendor who thinks of this as a support tool meters conversations. One who thinks of it as a knowledge product meters documents. One who came from a helpdesk background meters seats. Each is internally coherent and none of them convert into each other without your own numbers.

So the work is conversion. Take one set of volumes, your own, and price every candidate against it. That is an hour with a spreadsheet and it is the only comparison that means anything.

Per conversation or per session is the most common and the easiest to reason about, but the definition varies. Ask how a conversation ends: after a fixed period of silence, at the end of the browser session, or never. Ask whether a visitor who returns the next day starts a second one. Ask whether a session where somebody opened the panel and typed nothing counts at all.

Per message or per reply scales with how talkative the assistant is, which is a design decision the vendor makes rather than one you do. A product that asks a clarifying question before answering will produce more billable events for the same underlying demand. That is not automatically bad, since clarifying questions can be good, but you are paying for a behaviour you do not control.

Per resolution or per answered question sounds the most aligned with value and is the most defined by the vendor. Ask precisely what counts as one, who decides, and whether a refusal counts. If the definition is a conversation that did not escalate, you are paying per event that includes people who gave up.

Per seat or per agent is inherited from helpdesk pricing and often fits badly here. If nobody on your team logs in day to day, you are paying for chairs nobody sits in. If everybody has to have a login to read a conversation, the cost grows with your headcount rather than with anything the product does.

Per indexed page, per document or per source moves with your content rather than your traffic. It is stable and predictable, and it charges you for writing more, which is the opposite of the incentive you want if the whole plan is to document more. Per stored word or per chunk is the finer grained version of the same thing, and it makes a large collection of documents expensive to keep even if nobody ever asks about them.

Per crawl or per refresh charges for keeping the index current, which rewards leaving it stale. Per site or per domain matters exactly when you have more than one, which is more often than people expect. And a flat rate with a fair use clause has a unit too; you simply do not get to see it until you cross it.

Anything metered by conversation or message counts events, not people, and a surprising share of events are not customers. Automated traffic opens widgets. Your own team tests things, repeatedly, especially in the first month. A colleague shows it to a friend. Somebody links to the page from a forum.

Placement matters more than traffic does. The same site with the widget opening automatically on every page produces a different event count from one where a small button waits in the corner, and the difference is much larger than any change in visitor numbers. That means an estimate built from your traffic without knowing the placement is not an estimate, it is a guess with arithmetic on top.

Three questions worth asking before you sign: does an interaction with no message count, are refusals billable, and can you exclude your own addresses from the meter. The third one sounds petty and matters most in the month when you are testing hardest.

There are three axes a business grows on and each meter tracks a different one. Traffic growth drives conversation and message meters. Content growth drives page, document and storage meters. Team growth drives seat meters. Nothing tracks revenue, which is the axis you actually care about.

So the useful question is not what this costs now, it is which axis you expect to move next. A business about to triple its documentation should be wary of anything priced per page. A business about to run a campaign, or add a widget to a high traffic landing page, should be wary of anything priced per conversation with an uncapped overage. A business about to hire a support team should look hard at seat pricing.

Model the month at several times your current volume as well as at your current volume. That second column is where the products separate, and it is the column that will be true within a year if anything you are doing works.

What happens when you reach the limit matters more than where the limit is, and it is usually one of two things. A hard cap means the assistant stops answering, which will happen on your busiest day, because that is what busiest means. An uncapped overage rate means it keeps working and the bill is unbounded, which is fine until something automated finds your widget.

The sensible arrangement is an overage rate with a ceiling you set, and it is worth asking for explicitly. Ask for the overage rate as a number, not as a description, and ask what notice you get before it starts. A product that will tell you at eighty per cent of the allowance is meaningfully different from one that tells you on the invoice.

Then the extras, which are where the difference between two similar looking plans usually lives. Removing the vendor's branding. Additional sites or domains. Exceeding a storage allowance. The price of the plan that includes the one feature you need, which is often several steps up. Annual commitment against monthly, and what the notice period is. And whether you can export your material and your conversation history, which is not a price but is what determines whether the first decision is reversible.

Start from what you can measure rather than what you can guess. Your analytics already know how many sessions the pages carrying the widget receive in a month. That is the denominator.

The engagement fraction, the share of those sessions that produce a conversation, is the number you cannot look up, because it depends on placement, prominence, the kind of site, and whether the panel opens by itself. Do not take an industry figure for this; there is no such thing that survives contact with your specific page. Measure it, by running something for a month, which is the argument for a short paid pilot over a long negotiation.

Then count what the content meters would charge for. Not the pages you plan to write, the pages that exist: indexable pages on the site, plus documents you actually intend to upload, plus their rough size. Count the people who genuinely need to log in, which is usually fewer than the people who ask for a login.

With those four figures, price every candidate twice, once at today's volume and once at several times it, and put the two columns next to the headline prices you started with. The ranking almost always changes, and the product that looked cheapest on the pricing page is frequently not the one that is cheapest for you.

The annual discount is real and it is small compared with the cost of being wrong about the unit. For the first quarter, pay monthly even at a worse rate, and treat that quarter as the measurement exercise it is.

During it, collect your own usage in the vendor's own meter and ask for it in a form you can export. The second month should be a calculation rather than an estimate, and by the third you should be able to say what a year costs at your real engagement rate, with your real content, at whatever growth you are actually seeing.

This is also the period to find out how the meter behaves in practice rather than in the documentation. Whether test conversations were billed. Whether automated traffic showed up. Whether re indexing after a content update counted as anything. None of these are on the pricing page and all of them are on the invoice.

What is one unit, defined precisely enough that you could count them yourself. What happens to a conversation that gets no reply, or that is refused, or where nobody typed anything. Whether re indexing is metered separately from storage.

What happens at the limit, in both directions: is there a hard stop, is there an overage rate, is there a ceiling, and how much warning do you get. Whether you can exclude internal traffic. How many sites the plan covers and what an additional one costs, which matters more than it sounds if you have a shop on a second domain.

And two that are not about price at all. Can you export your indexed material and your conversation history, in a usable form, without asking. And what is the notice period, because that is what determines whether the answers to all the previous questions are findings or regrets.