One of the conversation topics at ProcureCon East was providers who aren't passing along savings from AI because they need to recover their investment. In some cases they're asking for higher prices instead. As a startup founder with a new product and a lot to prove that felt like gross entitlement. Then I remembered getting different versions of the same request as a buyer (e.g. the post-covid resource shortage / salary spikes).
To an extent I understand why software and services providers alike are feeling the pressure to put an AI label on what you sell, the popular consensus is that you’ll be out of business by the end of the decade if you dont. Real money goes into this: software, getting it working, training people, and cleaning up after it confidently does the wrong thing.
Whether customers asked for the feature, or for the price attached to it, is a different question. You COULD settle much of this with simple math, but only if someone can or will say what the work actually produces.
This topic - outcome based pricing - is in my opinion the most important concept for a buyer to understand when it comes to contracting, vendor / operations management, risk management - i.e. EVERYTHING. If you worked with me at google you probably heard me rant about this ad-nauseum, and unfortunately you still haven’t escaped it. If you can learn how to turn your outsourcing contract / operation into an outcome based pricing commercial model you will have achieved nirvana - whether your boss cares / understands is a different question I unfortunately can’t help you with (yet…).
If you pay by the hour, you have no output price at all
If you're paying by the hour (also known as T&M aka staff aug), your price contains no information about output OR service you are buying whatsoever. A rate times a quantity of hours tells you what time cost. It tells you nothing about what that time produced, how much was useful, or whether a more competent or technology enabled team would have needed half as much of it.
And the incentive runs backwards. The longer the work takes, the more the vendor gets paid. Inefficiency, rework, slow ramp-up and over-staffing all arrive as revenue. Said in another way - the longer it takes the vendor to perform the work, the more they get paid! Are there vendors who take advantage of this? You better believe it… If you have T&M pricing in your contract your interests and the vendor's point in opposite directions on the one variable AI is supposed to move (labor). If what you are trying to buy is productivity (which as a client procurement or operations management profession should be your entire job), that is about the worst commercial structure available to you.
We have a whole vocabulary for making this sound like something else: managed service, delivery pod, flexible capacity model. But if your team assigns the work and accepts the result while the rate card pays for hours, the delivery risk is still yours. That's staff augmentation, whatever the cover page says.
Sometimes capacity is what you need: unpredictable demand, changing scope, control over how the work gets done. It took me many years to understand that the woefully underscoped contracts and operations we ran at Google had billions of dollars of new product revenue behind them…it didn’t matter. Buy it on purpose and manage it like capacity. But unless you’re selling a once in a generation product like adwords, you can’t afford to do this. You can reconstruct a historical cost per output with operating evidence outside billing-volumes delivered, quality results and effort per item. That doesn't make it an output-priced contract.
Start with the unit you are buying
My outsourcing forefathers who taught me this business had one rule for a services contract, and I've never found better: translate every rate or billing unit into a volume unit with a service level attached. This unit of work, at this service level, for this price, performing or not. AI doesn't change the question. If you hire an expensive consulting firm like the ones I used to work at to write and negotiate your contract this is what you will get, and after seeing what the opposite looks like for nearly a decade I can tell you it’s worth the money.
An hour isn't an output. Neither is a seat, a headcount commitment, a token or a bot message (and note - the way that Anthropic and Google are charging you for tokens is still an unsolved problem…unfortunately its up to you to make sure the tokens your employees are burning are worth anything at all - i’ll get back to you once i figure this one out). Those are things you can charge for. The useful output is the issue actually resolved, the invoice paid correctly and on time, the reconciliation closed to a standard your controller will sign. Count how many you got last month and what it took to produce them, including your team's review and rework. Now you can calculate useful output per human hour, and separately, total cost per useful output.
Paying per ticket can still reward the wrong thing. Close one without fixing the problem, have the customer come back twice, and you've sold three tickets instead of one RESOLUTION (or ‘closed case’). The unit needs a definition of done, a quality standard and someone who can accept it. Decide which exceptions the provider owns and which belong to you. If your approval queue holds up payment, don't pretend the provider controls that delay. Otherwise you'll negotiate a 'fixed' price and spend the rest of the contract arguing about what was included.
Once you agree what completed work means, you have something to negotiate. Ask the provider for the current volume and service level, the resources needed per unit, and what AI changes about those numbers. If the investment reduces the work required, how much of that improvement reaches you, and when? Put the answer into the commercial terms: a unit price, a volume commitment, a service level and a date. You don't need to argue about whether their AI is impressive. You need to agree what you're paying for.
The software version of the same question
For software, start with the work the user is trying to finish. If you sell me a reconciliations platform with an AI module, I want to know how many reconciliations my analysts can complete to the same standard, and how much review and cleanup that takes. More AI messages isn't an improvement if the team still spends Friday fixing the answers. More completed reconciliations with the same people, or the same volume with less effort, gives us something to put a price on. A new SKU doesn't tell me that either of those things happened.
The hypothetical, and the four numbers it produces
The numbers below are invented to show the method, so don't go quoting them at anybody. You buy support on an hourly basis, billed on active handling hours - precisely the contract I've just said tells you nothing - so the first move is to get output data from somewhere other than the invoice. Your ticket system says 10,000 issues were usefully resolved last month and your quality reporting says the bar was met. Hold everything else still: same work mix, same quality bar, same volume, same service.
Before: 2,000 vendor hours at $30, so $60,000 a month, and $6.00 per resolved issue. The provider deploys AI assistance and the hours fall to 1,500. The rate goes up to $36 to recover the investment, which is $54,000, plus a $6,000 monthly technology fee. The vendor's invoice is $60,000 again - not a dollar of movement. Meanwhile your own team now spends 40 hours a month checking AI-assisted responses at a loaded $50 an hour, another $2,000 that shows up in your payroll instead of on anyone's invoice. Your total cost is $62,000.
The provider has improved labor productivity. Before, 10,000 issues took 2,000 human hours: five resolved issues per hour. After, 10,000 ÷ 1,540 hours (their 1,500 plus your 40) gives about 6.49, roughly a 30% improvement. That measures output per human hour; the technology fee goes into the cost calculation below. Doing the math tells you something useful about delivery. It still doesn't change what your hourly contract pays for.
Cost is a separate calculation that happens to use the same inputs. My total cost per resolved issue goes from $6.00 to $6.20, up 3.33%. The provider's hours per issue fell from 0.200 to 0.150, down 25%, which is genuinely their achievement - but their hours per unit is not my cost per unit, because the technology fee and my 40 hours are resources too. Labor productivity up 29.87%, vendor hours per unit down 25%, vendor invoice flat, my cost per unit up 3.33%. Four true numbers out of one month of one service, and which one you hear first depends on which meeting the vendor is in.
Who gets the gain, and where it shows up
I want to be fair here, partly because I'm now one of the people asking you for money. A provider investing in AI is making a bet with their own capital, and that bet doesn't come with a claim on my budget. Under unit pricing, completing the same work in fewer hours can improve their margin. Good! They have a reason to get faster, and I know what the completed work costs. But my share of future improvements still has to be negotiated: a scheduled price reduction, an agreed share of measured savings, or more completed work for the same fee.
That benefit doesn't have to be a lower invoice. Faster service the business values, better quality, more completed work you actually need or less work for your own team can all be worth paying for. But unused capacity isn't a saving, and giving my team another queue to manage doesn't become a benefit because you've called it AI assurance.
Ask about the charge itself, too. Is this recovering an implementation, covering ongoing usage, or buying additional service? If I'm funding implementation, when does the recovery end-or did I just buy an annuity? When should the benefit land, in which number, and what happens if it doesn't?
Changing vendors means paying for a sourcing process, a transition and a new team learning the work while yours fills the gaps. Sometimes you don't have a credible alternative this year. That doesn't make the provider's AI claim untestable. Take one service with clean volumes and use your own recent delivery data to establish what you're getting now. At renewal, ask them to price that defined work at that service level, with the volume assumptions and exceptions written down. Then negotiate how improvements reach you. Their answer will tell you much more than another presentation about the AI investment.
And if you'd rather not build the spreadsheet yourself, our publicly available MaxxBot product at maxx-bot.com has a T&M to outcome based pricing workflow. You can chat with Maxx and have him do the calculations for you, using your own numbers. There's a 7 day free trial if you want to try it on one of your services.
So here's where I land. I'll pay for AI, and I'd rather the provider made money on it, because I want them in business in three years and better at the work than they are today. But show me the unit, the service level attached to it, and what happened to the resources it takes to produce one. If your hours per unit are down 25% while my cost per unit is up 3.33% and my team has picked up 40 hours a month of review, then the investment is working beautifully - just not for me. If no usable benefit reaches me, why would I pay more for your AI investment?