Outcome-Based Pricing for AI Support: What a Resolution Costs, and Who Decides It Was One

Outcome-based pricing charges you for results instead of for seats. In support that nearly always means per resolution, so you pay a fixed amount each time the software closes a ticket without a person going near it, and you pay nothing whatever for the ones it tries and fails to close. Said in that order it sounds like the fairest offer anybody has put in front of a support budget in a good long while.

It may well be, and there is a section further down that gives that side of it a proper hearing rather than a straw one. The trouble is not the idea of it. The trouble is that a per resolution quote is not a price at all, it is a price multiplied by an assumption, and the assumption belongs to whoever is doing the quoting, and it does not appear anywhere on the page they hand you across the table, mind you, nor is anybody hiding it from you exactly, it simply never got itself written down.

We sell an AI layer our own selves, so read the whole of this knowing that much about who is talking. Our bill goes against the workspace and never counts your people, which is a different shape entirely and one we plainly prefer, and there is a section down the far end where the same questions get turned back on our own pricing and we do not come out of it spotless.

What outcome-based pricing actually charges you for

Three fairly different arrangements go by the name of outcome pricing and it is worth separating them before any arithmetic happens, on account of them behaving nothing like one another once the invoices start arriving.

The common one charges per resolution. A ticket comes in, the software answers it, the ticket closes, and a fixed sum lands on the bill for that closure. The second charges per deflection, which counts the ticket that never got filed at all because somebody found their answer in a help article on the way to the contact form. The third is written per outcome in the literal sense, where the contract names the actions being paid for, a refund processed or an order traced or an address changed, and only those count.

They are not variations of the one thing, truth be told, and a quote that uses the words interchangeably is telling you something about the care that went into the rest of it. The first is countable and arguable. The second is countable too, but the counting of it is a matter of opinion rather than of record, and the customer self service page sets out what a deflection figure quietly leaves out of its self. The third is the cleanest of the lot of them and it is also far and away the rarest, on account of it requiring somebody to sit down and write out what the software is actually meant to accomplish, which is work, and a per resolution meter is the option that quietly skips that work and charges you all the same. The wider shape of the market, per agent and tiered and free and the rest of them, is set out on help desk pricing models, and this page is only the metered corner of it.

What counts as a resolution, and who gets to decide it

The pricing models page asks who decides a thing was resolved and then moves along, being a page about four models rather than about one. Here is the longer version, seeing as the answer is not a product feature at all, it is a contract term, and the whole of it goes missing from the first quote nearly every time, not because anybody meant it to go missing but because nobody in the room had a reason to raise it.

Four definitions are in circulation, mind you, and every one of them is defensible enough on its own terms. A resolution is whatever the software marked closed. Or it is a ticket the customer did not write back about inside some window. Or it is one where the customer said something that reads like thanks. Or it is any ticket that closed without a human being touching it at all, which sounds the strictest of the lot until you notice that it goes and counts the ones your customer simply gave up on, and a customer giving up looks identical from the inside of a report to a customer being helped.

The window is the whole negotiation and hardly anybody goes and negotiates it, truth be told, on account of it sitting three pages into a schedule nobody reads aloud. Set the reopen window at 24 hours and a customer who comes back on Thursday about Tuesday’s problem is a second resolution, billed again, for the identical complaint. Set it at 14 days and most of the comebacks land inside it and cost nothing. Say 12% of closures come back within a fortnight, which is a number invented for this sentence and nothing more, and very nearly the whole of that 12% either lands on your bill a second time or it does not, decided by a clause nobody read out at the meeting.

Ask for the window in writing. Ask what a reopen does to the count. Ask it before the commercial conversation warms up, because afterwards it sounds like haggling and beforehand it sounds like diligence, and it is the very same question either way, no two ways about it.

How the same quote produces three different annual bills

Here is the part that does not get printed. A per resolution rate on its own tells you nothing at all about what you will pay, since the bill is the rate times the number of tickets the software actually closes, and that second figure is a forecast somebody made about your desk before ever seeing your desk, or your customers, or the shape of it in any month of the year.

Take a desk handling 2,000 tickets a month at a rate of $1.50 a resolution. Now run it at three different assumptions about how many of those the AI closes on its own. At 25% you are paying for 500 closures a month, which is $750, which is $9,000 across the year. At 45% it is 900 closures, $1,350 a month, $16,200 for the year. At 65% it is 1,300 closures, $1,950 a month, and $23,400.

Same desk, same rate, same contract. The top and the bottom of that are 2.6 times apart and every one of the three is a perfectly ordinary thing for a support desk to experience. It is not a small gap. None of those figures came from anywhere except this paragraph, so do not go quoting them at a vendor, the shape they make is the only part that is real.

Now put a flat annual bill of about $16,000 beside the three of them, that being another invented figure standing in for whatever your current arrangement costs you, and watch what happens. Against the first assumption the meter wins comfortably, against the second it is roughly a draw, and against the third the flat bill was the better buy by a distance. The comparison does not merely shift as the assumption moves, it inverts, and it inverts inside a range that nobody could reasonably call extreme. That is the most of it, sat there in the one table. You were not handed a price to compare against anything. You were handed a price and a guess, printed in the very same typeface, and nothing on the page tells you which of the two is which.

Why the meter is read by the person you are paying

Every other variable bill in the business reconciles against something you hold your own self. The card processor’s fee, Stripe or whoever it is you happen to use, lands beside transactions sitting in your own ledger. The courier’s invoice lands beside parcels you can go and count in the yard, and the lot of it reconciles into Xero at month end without anybody raising their voice. The hosting bill lands beside traffic you can pull off your own logs if you have a mind to, and if the two of them disagree you get to have a conversation with evidence in it, which is the whole point of it.

A per resolution invoice has no second copy. The count comes out of the vendor’s software, using the vendor’s definition of closed, displayed in the vendor’s dashboard, and there is nowhere else in the world the number exists. Nobody else holds a copy. Nobody is suggesting anybody would go and cook it. The point is narrower and it holds even with everyone acting in perfect good faith: there is no arrangement by which you could tell.

So make one. Ask, before signing, for a ticket-level export as a CSV rather than a summary figure, carrying the ticket reference, the closure timestamp, the resolution flag, the CSAT score if you collect one, and whether a human ever replied at all. Ask as well whether it arrives monthly without being chased and whether there is an API behind it or only a person exporting a file. That export lets you recount the invoice against your own records once a quarter, which takes an afternoon of somebody’s time and then settles the question the same way every quarter after it. If the answer is that the platform reports totals and not rows, then you have learned something useful for free, and you have learned it in the month before you signed rather than in the month the bill went strange.

What a variable bill does to a fixed support budget

Support budgets are annual and they are fixed, and a metered bill is neither. Everybody knows this in the abstract and it still catches desks out, and what catches them is the way the timing of it lands.

Volume climbs when something has gone wrong, and it is rarely something you did. A release ships with a fault in it, a supplier misses a run, a payment provider has an afternoon, and 2,000 tickets a month becomes 3,200 for six weeks. On a per ticket meter you pay more in a bad month, which is annoying enough on its own. On a per resolution meter it is sharper than that, and this is the bit worth sitting with. A bad month is usually a repetitive month, hundreds of people asking the identical question about the identical fault, and repetitive tickets are precisely the ones an AI closes best. So the resolution count does not climb in step with the ticket count, it climbs faster, and your largest invoice of the year arrives in the month your desk had its worst fortnight, while the figure your finance team keyed into the ERP back in the spring sits there not moving to accommodate any of it. Nothing about that is unfair exactly. It is simply the arithmetic doing the job it was built to do, and it goes and does it hardest on the day you have the least appetite for it, and that is the most of it.

The people who handle this well ask for a collar. A monthly cap, or a floor with a ceiling around it, or a rate that steps down past an agreed volume, and any of the three turns an unbounded line into a budgetable one. Ask for it. The worst that happens is somebody says no and you have priced the risk your own self instead.

Where outcome pricing is genuinely the better shape

Fair is fair, and a page arguing one side of a thing owes the other side an honest hearing rather than a straw one.

A meter is the right shape when your volume is genuinely spiky. A desk that does almost nothing for eight months and then carries a season pays for seats it is not using all year under any per person arrangement, and a meter charges it in the months it actually has a desk. That is not a loophole at all, that is the model working exactly as intended, and there is no arguing with the shape of it.

It is also the right shape in a first year where nobody knows whether the AI does anything at all. You are not really buying resolutions there, you are buying evidence, and a bill that only appears when the thing works is a cheaper way of finding out than a year’s licence paid up front. Truth be told, that argument is stronger than most of what gets written for outcome pricing, and it is stronger because it is temporary. It is an argument for a trial rather than an argument for a contract.

And there is a real point underneath all of it, which is that a vendor charging per seat gets paid the identical money whether their software helps you or not. Nobody should pretend that is a healthy arrangement either. A meter at least puts something of theirs on the table. The argument here is not that meters are wrong, it is that this particular meter gets read in a room you are not standing in, and you are asked to take the reading on trust.

Why the worse incentive ends up sitting on your side of the table

The vendor’s bad incentive is the one everybody writes about, and it is real enough. Repair a fault at the root and their revenue goes down. Publish one good article that stops a question being asked for the next two years and their revenue goes down again, and nobody over in that building is going to send you a card about either of them. The pricing models page says as much and it is right to.

It is also, funnily enough, not the one that will cost you. That part is mostly fine. Vendors behave themselves, on account of them wanting the renewal a good deal more than they want the one good quarter, and the ones who do not behave tend to be found out by somebody else before they ever reach you.

The one that costs you is the one that arrives in your own building. Once a resolution has a price on it, the count stops being a report and becomes a budget line, and budget lines get managed. Somebody asks whether the AI should really be let near the easy categories at all, when a person could clear one of those in well under a minute and it would cost nothing. Somebody else proposes switching it off for a fortnight to see what happens to the invoice. A third person starts checking closures against the definition, which means reading them, which is a job nobody was hired to do and nobody is thanked for. Inside two quarters your Monday meeting has acquired a standing agenda item and the item is accounting.

That is the actual cost of the model and it does not appear on the quote. You imported somebody else’s unit of measure into your own desk’s daily conversation, and a unit of measure does not sit quietly in a report, it goes and turns up in what people argue about on a Monday morning. Ask any desk that has ever been managed on average handle time, or AHT as the report calls it, how that one worked out for the tickets that genuinely needed the guts of half an hour. It is the very same mechanism, only newer, and there are no two ways about it.

The place in the page where we turn into a vendor

Maxdesk runs a support desk off a single address and nothing more complicated than that. Whatever arrives at support@, and it makes no odds whether that address sits on Gmail or on Outlook, drops into a shared inbox that behaves like a queue rather than like a folder, with states and assignment and an audit trail running underneath the lot of it, and an SLA policy over the top if you want the clock watched.

The bill goes against the workspace. Not against your agents, not against your tickets, not against your resolutions. Add the weekend cover, add the person from finance who only ever handles refunds now and then, add the whole of the night shift, and the number at the bottom of the invoice does not go and move an inch. There is a free plan, and above it a paid one, and above that one an AI layer, and none of the three counts heads. The figures themselves live on the pricing page and they are what they are as of the day you read this, which on our side was August 2026.

The AI layer is an AI customer support agent that drafts and answers on the tickets it is confident about and leaves the rest for your people, and there is a separate page on what the thing actually is if that is the question you came in with. It is sold as part of a plan. It is not sold by the resolution and there are no plans here to sell it that way.

What our own bill does that we would criticise in somebody else’s

Start with the awkward one, on account of a page like this having no business burying it down at the end.

The AI layer comes with 5,000 AI responses in a month, and past that number you are buying packs. So we meter something too, and a piece that has spent seven sections picking away at other people’s meters would be writing the cheapest sort of copy altogether if it went quiet about the one sitting in its own price list. The distinction we would argue for is a narrow one and there is no point pretending otherwise: an allowance of that sort rations the software, and it does not ever ration your colleagues. Nobody loses a login over it. Whether that distinction is worth anything is for your own self to judge and not for us to insist on.

It should be said plainly as well that we benefit from you distrusting per resolution pricing. We do not sell it, we sell the flat shape, and the flat shape is the one this page happens to conclude in favour of. That is not an accident and you should weigh the whole of it knowing that much. The arithmetic above works the same in your spreadsheet as in ours, which is rather the point of showing it.

And flat pricing carries an unfairness of its own that we would sooner not skip past. A quiet month costs exactly what a hammering month costs. If your desk is genuinely tiny, or your volume is seasonal in the way described further up, then a meter may well cost you less than we do over a year, and there is no version of this page where we go and tell you otherwise just because it suits us. The honest test is what a ticket costs you either way, worked out on your own numbers.

The last of it is the channel. Email is the only one we handle at all. There is no chat widget on offer, no phone line at this end of it, no portal a customer could log into, and not one of those three is sitting on a roadmap waiting its turn either. If what is genuinely wrong at your end is that your customers would far rather be chatting than emailing, then our pricing model is not the thing standing in your way, and this is not the page you needed, and fair is fair about saying so plainly.

How to check this before you sign anything

You do not need the whole estate for this. One year of tickets and an afternoon of somebody’s time will do the job.

Pull last year’s tickets out as a CSV. Count the closures three separate ways: the ones no human ever replied to, the ones that were not reopened within 7 days, and the ones not reopened within 30 days. Three counts off the same year, and the gap between the largest and the smallest is the width of the argument you are about to have with a vendor. If that gap comes in under about a fifth, the definition barely matters and you can get on with the commercial conversation. If it comes in wider than that, and on most desks it does, then the definition matters a great deal more to you than the rate does, and you should be negotiating the clause rather than the number, whatever the person across the table would rather talk about.

Then multiply each of your three counts by whatever rate you were quoted, and put the three answers beside your current annual bill. That is the comparison. Anything shorter than that is not really a comparison, it is a brochure with a bit of arithmetic printed on it, and what you would be signing off is the arithmetic rather than the deal.

Five things to settle in writing before anybody signs:

1. What event marks a ticket resolved, stated as a rule a person could apply by hand.
2. How long the reopen window runs, and whether a reopen inside it is billed again.
3. Whether a deflection counts as a resolution, and what evidence stands behind the count.
4. Whether a ticket-level export is available to you, monthly, with the resolution flag on it.
5. What the cap or the collar is, and what the rate does when volume goes past it.

If you want the flat version of this instead, ours is the one where the bill does not move when the desk does. And if you would rather test that on your own year of tickets before believing a word of it, that is by far the more sensible order to do it in.

Common questions about outcome-based and per-resolution pricing

What is outcome-based pricing in customer support?
It charges for results rather than for seats or licences. In practice that nearly always means per resolution, where a fixed amount is billed each time the software closes a ticket without a person handling it, and nothing is billed for the attempts that fail.

Is per-resolution pricing cheaper than per-agent pricing?
It depends entirely on how many tickets the software actually closes, and that figure is an assumption at the point you sign. On a desk of 2,000 tickets a month at $1.50 a resolution, an assumed 25% closure rate costs $9,000 a year and an assumed 65% costs $23,400. Those are invented figures, but the 2.6x spread between them is the reason a rate on its own tells you nothing.

What counts as a resolution?
There is no standard, which is why it belongs in the contract rather than in the demo. Four definitions are in common use: the software marked it closed, the customer did not write back inside a set window, the customer confirmed it, or no human ever replied. Each one produces a different bill from the identical year of tickets. None of the four is wrong.

Does a reopened ticket get billed twice?
Under most per-resolution contracts, yes, if the reopen falls outside the window. That is why the length of the window matters as much as the rate. A 24-hour window bills the Thursday comeback as a fresh resolution; a 14-day window does not.

Can you audit a vendor’s resolution count?
Only if they give you the rows. Ask for a monthly ticket-level export carrying the closure timestamp, the resolution flag and whether a human replied, so the invoice can be recounted against your own records. A summary total cannot be checked against anything.

Should a small team choose per-resolution pricing?
It can suit a genuinely spiky or seasonal desk, and it suits a first year where you are really buying evidence that the AI works at all. It suits a steady desk less well, because the bill peaks in the months the desk is already struggling. Settle the definition and the cap before the rate either way.