Your Client’s AI Revenue Is Probably Coming From a Rival’s Budget
Why AI growth is increasingly a substitution fight, not the market expansion most forecasts assume.There is a number in Kyle Poyar’s survey of B2B monetization that should change how any advisor talks to a client about their AI business plan, and it is not the headline number. The headline is that…
Why AI growth is increasingly a substitution fight, not the market expansion most forecasts assume.There is a number in Kyle Poyar’s survey of B2B monetization that should change how any advisor talks to a client about their AI business plan, and it is not the headline number. The headline is that hybrid pricing now dominates at 37%, up from 25% a year earlier, and that 75% of companies changed their pricing or packaging in the past year, which is interesting and expected.The number that matters is buried lower and stated almost apologetically: 70% of AI spend cannibalizes existing software budgets rather than expanding total spend. Poyar surveyed 230 B2B software and AI companies in the spring of 2026 to get it, and it quietly demolishes the business case sitting in most of your clients’ AI decks.Because the business case in those decks is an expansion case. It assumes the client’s AI product is net-new revenue, layered on top of what customers already spend, growing the pie. The vendor’s TAM slide is built on that assumption, and so is the client’s internal forecast, and so is the valuation they are hoping to justify.If 70% of the money is coming out of budget lines the customer already had, then most of that forecast is not expansion at all. It is a substitution fight: your client’s AI product versus whatever software the customer stops paying for to afford it, and substitution fights are won and priced on completely different terms than expansion. An advisor who lets a client model a knife-fight as a greenfield is setting them up to miss badly.The layoffs are the same accounting entry, seen from the vendor sideThe tech layoffs of this summer are usually filed under labor market news, and there is a real story there about workers that other people are better placed to tell. From an advisory seat, though, Oracle cutting 21,000 jobs and Microsoft cutting 4,800 are evidence for the substitution thesis, and they are evidence from the supply side of the same market Poyar surveyed from the demand side.Oracle disclosed the 21,000 figure, a 13% workforce reduction over twelve months, and acknowledged in regulatory filings that “adoption and deployment of AI technologies” contributed. Microsoft cut 4,800 on July 6, about 2.1% of its workforce, saying AI is “changing how work gets done” while insisting the roles were “not being replaced by AI,” a sentence doing a great deal of quiet work.Layoffs.fyi counts roughly 120,000 tech roles gone in 2026 so far, and the early-July additions reached well past the giants, with hospitality-software firm Mews cutting about 200 people, 15% of its staff, and CorroHealth cutting 800. This is not one or two companies making a strange call. It is a budget behavior spreading through an industry.The important feature of these cuts, the one that connects them to Poyar’s number, is that they are profitable companies cutting to fund AI infrastructure, not distressed companies cutting to survive. Oracle and Microsoft are not shrinking. They are reallocating, moving spend from headcount into AI capex.That is substitution accounting at the largest possible scale. The vendors themselves are not treating AI as net-new spending they can simply add. They are funding it by cutting something else, which is exactly what Poyar’s customers are doing when 70% of their AI spend comes out of an existing budget.When the companies building the AI are financing it through substitution inside their own P&L, it is a strange kind of optimism for a client to assume their customers will finance it through expansion. The people closest to the economics are behaving as though the pie is fixed and the game is reallocation. Your client’s forecast usually assumes the opposite, and the two cannot both be right.Substitution changes the pricing question entirelyOnce you accept that most AI revenue is substitution, the advisory question stops being “how big is the market” and becomes “whose budget line are we taking, and can we defend the price once we have it.” That is a harder question and a more useful one, and Poyar’s data gives it teeth.A substitution product competes on displacement, which means its price is anchored to the thing it displaces, not to the value it theoretically creates. This is why the pricing chaos in the survey exists. Salesforce is running four distinct AI pricing models, HubSpot introduced outcome-based pricing for its Breeze agents and halved the price of Fin, GitHub Copilot launched AI credits in June, and AI credits overall grew 126% year over year.Credit adoption sits at 29% of the surveyed companies, with another 33% planning to add them within a year; roughly half of the companies past $150 million in revenue already run them, and Clay monetizes the platform and the tokens on two separate tracks at once. That is not a market that has found its pricing. That is a market discovering, model by model, that you cannot charge expansion prices for a substitution product, and scrambling to find a structure that survives contact with a customer who is paying for your thing by canceling something else.The margin data confirms the squeeze: the median target AI gross margin in the survey is 50%, and only 12% of companies target the 80%-plus that traditional software takes for granted. Substitution products carry substitution margins, and half of standard software margin is the price of playing in a fixed pie.For the advisor, the move is to force the substitution question early, before the client has committed to a price built on an expansion assumption. Which budget line does this come out of? What is the customer canceling to pay for it? Is the value we deliver large enough to survive being anchored against the cost of the thing we are replacing, or are we going to end up, like Fin, cutting our own price in half once the customer does the math?Those questions are unwelcome in a room that came to celebrate a big TAM, and they are the ones worth the fee, because they are the difference between a client who prices for the fight they are actually in and one who prices for the fight in the slide. The cleanest version of the exercise is to draft the customer’s cancellation email for them, the note their operations lead would send to the incumbent vendor, and see whether your client’s value story survives being the reason that email exists.Substitution also changes the sales motion, not just the price, and there is large-sample evidence for how. Matt Dixon’s team collected 2.5 million recorded sales conversations and ran them through machine learning, and the finding that survived all that data is that the biggest deal killer is not the competition but customer indecision, the fear of messing up rather than the fear of missing out. Put that finding inside a substitution fight.An expansion sale asks the buyer to make one decision: whether the new thing is worth new money. A substitution sale asks for two, because the buyer must also choose what to cancel, and the cancellation has an owner, a renewal date, and an internal champion who will notice. Every added decision is another place for the deal to die of hesitancy, which is why Dixon’s prescription, de-risk the decision rather than sell harder, is close to mandatory here.The client who makes the cancellation easy, who maps the switch and absorbs the migration and names the budget line, is doing more for the win rate than another feature demo ever will.CAMP is the tool the moment calls for, not the TAM slidePoyar offers a specific instrument for the pricing question, the CAMP test, and it is the right thing to reach for here because outcome-based pricing is where every AI vendor wants to end up and where most of them will hurt themselves. CAMP asks four things about whether you can actually charge for an outcome: consistency, attribution, measurability, and predictability.Can the product deliver the outcome consistently, can you attribute the outcome to the product rather than to everything else the customer did, can you measure it in a way both sides accept, and can the customer predict what they will owe?Outcome pricing works when all four hold and becomes a liability when they do not, because you have promised to be paid on a result you cannot reliably produce, attribute, measure, or forecast.The reason CAMP is the tool for a substitution market specifically is that substitution makes attribution brutal. When your client’s AI product is net-new, the customer has a clean before-and-after, and the value is easier to attribute.When it is displacing an incumbent tool, the customer already had a version of the outcome from the thing being replaced, so the marginal value your client can claim, and price against, is only the improvement over the incumbent, not the whole outcome. That is a far smaller and far more contested number, and a client who prices as though they own the full outcome when they only own the delta will lose the CAMP test in front of a customer who runs the comparison. HubSpot halving Fin’s price is what losing that argument looks like in public.So the two documents an advisor should be suspicious of are the vendor’s TAM slide and the client’s own expansion forecast, because both encode the assumption Poyar’s number contradicts. The TAM slide sizes a pie that 70% of the actual spending says is mostly already claimed by someone else. The forecast projects growth that is really transfer.CAMP is the corrective, because it prices the product against the specific outcome it can defensibly own in a substitution fight, which is the delta over the incumbent, measured in a way the customer will accept.One more feature of the 2026 market raises the stakes on getting that math defensible, which is that the entity running the comparison is increasingly not a person. G2’s data this spring had 51% of B2B software buyers beginning their research in AI chatbots, up from 29% a year earlier, and IDC projects that 70% of business buyers will use AI to find and choose tools by 2028, before ever visiting a website.Tom Orbach has been documenting the vendor response: companies like Buffer and Resend now ship machine-readable pricing files so that agents can parse their plans without a demo call. An agent doing procurement does not get charmed by the TAM story and does not get tired in the third hour of negotiation.It lines your client’s price up against the incumbent’s, computes the delta, and reports it. In a market where the comparison is automated, a price that only survives as long as the customer never runs the numbers is a price with a short life expectancy.The part I would flag before running with itI do not want to overstate the 70% figure into a law of nature, because it is one survey of 230 companies at one moment, and there are expansionary AI categories where the new spending is real, where the AI does something no incumbent budget line ever covered.Some of your clients are in those categories, and for them the substitution frame is too pessimistic. The advisory skill is telling which is which, and the tell is usually whether the customer had a budget for this before. If they did, you are in a knife-fight and should price like it.If they did not, you may have expansion, and you should still test it hard, because “no one was spending on this” is exactly what every category says right up until it turns out they were spending on a clumsy manual version of it all along.It is also fair to ask how the substitution picture squares with the venture tape, because the money looks like the opposite of a fixed pie. Crunchbase counted $392 billion into North American startups in the first half of 2026, an all-time record, with roughly 80% of second-quarter dollars going to AI companies. But the same wrap notes that deal counts stayed well below prior peaks.The capital is concentrating in a small elite while the long tail fights for scraps, and that is what investors do when they believe a few winners will capture the reallocated spend, not what they do when they believe the pie is growing evenly for everybody. Record funding and substitution economics are compatible. The investors are betting on who wins the knife-fight, and a client should notice that nobody writing those checks is pricing the round as if there were no knife.The steadier point underneath the number is the one I would actually build the engagement on. The companies closest to the money, the vendors funding AI by cutting their own staff and the customers funding AI by cutting their own tools, are all behaving as though the pie is fixed.When both sides of a market are financing the new thing through substitution, an advisor who lets a client plan for expansion is the last optimist in a room full of people voting with their budgets. Price the substitution. The expansion, if it is real, will survive the more conservative model. The substitution masquerading as expansion will not, and it is better to find that out in the planning than in the renewal.This story is published on Generative AI. Connect with us on LinkedIn and follow Zeniteq to stay in the loop with the latest AI stories.Subscribe to our newsletter and YouTube channel to stay updated with the latest news and updates on generative AI. Let’s shape the future of AI together!Your Client’s AI Revenue Is Probably Coming From a Rival’s Budget was originally published in Generative AI on Medium, where people are continuing the conversation by highlighting and responding to this story.Source: Generative AI Pub — Published — Category: Image AI