Last Updated on August 13, 2026 by Taya Ziv
Here is the story every AI founder has been telling themselves for two years. The incumbents are slow. Their software was built before any of this existed. Enterprises are about to rip out the old stack and go looking for something built natively for this era, and that something is us. The disruption window is open and we are standing in it.
I believed a version of that story too. Then I read INFUSE’s mid-year 2026 Voice of the Buyer update, released 25 June 2026, on a page INFUSE has re-stamped since, a survey of 310 B2B technology buyers across 266 companies, 18 countries and 15 industries. The picture that came back is close to the exact opposite of the story.
When B2B buyers in 2026 express a preference between an established vendor and an AI-native startup, they pick the established vendor by five to one. Twenty five percent versus five percent. Most of them, about 61 percent, say they judge case by case on the business need, which is fair. But among the ones who have made up their minds, being the new AI-native company is not the advantage. It is the thing costing you the deal.
One caveat before you rebuild your go-to-market on that number. INFUSE sells demand generation to B2B vendors, and the report’s own conclusion is that the vendors who own the categories buyers research and prove integration with the existing stack will keep winning, which happens to describe what INFUSE sells. Read it with that in mind. The direction still holds, and 310 buyers is a bigger sample than most of what gets quoted at you.
That number should reorganize how you think about your entire go-to-market, and almost nobody building right now has looked at it.
The wave is real. It is just flowing the wrong direction.
The part everyone gets right is that AI is causing a massive reshuffle of enterprise software. That is happening. In the same INFUSE survey, 61 percent of buyers are actively evaluating, planning, or already in the middle of eliminating platforms because AI alternatives now exist. One in four has already cut a platform. That is a different 25 percent from the one in the headline, from a different question, and the overlap is a coincidence rather than a finding.
Now read the other half of that finding, which is the half that never makes it into the pitch deck. Only 13 percent have added a platform because of AI.
Sit with the ratio. Twenty five percent cutting, 13 percent adding. The AI wave inside enterprises is not a buying wave. It is a pruning wave. Budgets are getting consolidated, line items are getting killed, and the money freed up is mostly going to the vendor already in the building who just shipped an AI feature and offered to absorb the workload of the tool being cut.
So the disruption is real and it is being harvested by the incumbent. Your prospect is in a room right now doing exactly the thing you predicted, questioning their stack, cutting tools, rethinking what they need. And the outcome of that meeting is usually fewer vendors, not a new one. You were right about the earthquake and wrong about which way the building would fall.
Your product is not being evaluated. It is on trial.
The other thing that changed is who is in the room, and it is worse than you think.
Forrester’s The State Of Business Buying, 2026, published 21 January 2026, puts the typical purchase decision at 13 internal stakeholders and nine external influencers, in Forrester’s own words. Twenty two people. For the big, complicated buys it climbs from there. That is not a sales conversation. That is a jury.
Forrester’s read on that room is warmer than mine, and it is worth saying so. The same release reports that 94 percent of buyers in groups of six or more say the group helps, mostly because it makes budget easier to secure and approval more likely. Fine. It still means your champion is arguing your case in a room you are not in.
And if you have seen 12 Angry Men, you know how juries work. Everybody files in already leaning one direction, the default verdict is the safe one, and the only thing standing between the defendant and a conviction is one stubborn person willing to argue for hours on their behalf. That is your champion. That is the entire job of the person who took your demo and liked it. They are not buying your product. They are going into a room with twenty one other people to defend a decision they will personally own if it goes badly.
Now ask what you have handed that person to defend you with. Usually it is a deck, some self-reported metrics, and a promise. In a room where everyone is being told to justify every dollar, that is not a defense. It is a confession.
This is why “we’re AI-native” lands so badly at that table. To you it means modern, fast, built right. To the twenty one people who have never heard of you, it means unproven, venture funded, might be acquired in a year, might change its pricing, might not exist when the contract renews. You are asking a group of people whose incentives are entirely about not being blamed to take a risk on a company they cannot Google a ten-year track record for. The incumbent does not have to be better. The incumbent only has to be defensible.
The AI research layer is not the shortcut you were promised
There is a second trap here and it is subtle, so stay with me.
Yes, buyers now start inside a chat window. They type the problem, they get a synthesized answer, and a shortlist forms before anyone visits a website. Everybody in marketing has figured that part out and there is a whole industry selling you visibility inside those answers. That industry is right that the shift happened, and I have argued before that the traffic that survives the AI layer converts far better than what it replaced. Being in the answer matters.
But here is what the buyer research adds, and it changes the plan. Buyers do not trust the answer. Forrester found that the speed of AI search comes with incomplete and unreliable output, and buyers know it, so they compensate by going and finding humans to validate what the machine told them. My read on why the buying group is this big: the machine made research faster and made buyers more suspicious at the same time, and the suspicion is doing more work than the speed. Forrester does not make that causal claim. It reports the mistrust, and separately it reports the group size. The line I am drawing between them is mine, not theirs.
So the assistant may put you on the list. It cannot get you through the room. Between the shortlist and the signature there are now 22 people asking their networks whether anyone has actually used this thing, and the answer they get about you, from a peer, in a Slack channel you will never see, is the real deciding moment of your sales cycle.
Which means visibility inside AI answers is a top-of-funnel fix for a bottom-of-funnel problem. Necessary. Nowhere near sufficient.
What I would actually do about it
None of this is a reason to stop building. It is a reason to stop selling like it is 2021. Five things I would change on Monday.
Sell a replacement, not an addition. The budget is coming from a cut, not from growth. So do the work of naming the line item you are replacing, in dollars, out loud, in the first meeting. “We do this job and here is the thing you can stop paying for” is a fundable sentence in this market. “We’re a new category” is a sentence that gets you deferred to next fiscal year. If you cannot point at what dies when you arrive, you are asking for net new spend in a year of net cuts.
Ship a trial, not a demo. More than 60 percent of buyers now run a trial before they commit, and among buyers making purchases of ten million dollars or more it is 78 percent (Forrester, January 2026). This is the single most important shift in the data and it is very good news for a startup, because a trial is the one arena where being new does not hurt you. The incumbent’s brand advantage evaporates the moment the buyer gets to touch both things. Make your trial embarrassingly easy to start, make it use their real data, and stop treating it as the last step of the sale. It is the sale. Everything before it is scheduling. This is also the honest test of whether your thing survives contact with reality, which is where the vast majority of perfect AI agent demos quietly die.
Arm your champion for a fight you will not attend. Your buyer is going into a room without you. So build the thing they carry in there. A one page internal business case with the numbers already run. The three objections their CFO will raise, answered. A named reference at a company their board would recognize. Most founders spend all their energy on the meeting they are in and none on the meeting that actually decides it.
Talk to procurement early, on purpose. Forrester puts procurement people as decision makers in 53 percent of business buying cycles, engaging from the start of the process rather than showing up at the end to haggle. Founders treat them as the final boss who arrives to cut 20 percent off the price. They are not. They hold a real veto, and they are going to ask about security, data handling, contract terms, and what happens to their data if you get acquired. Have those answers in writing before you are asked. The startups that lose here mostly lose on paperwork, not on product.
Retire “AI-native” as a headline. Keep it in the engineering blog. In the sales conversation it is a risk signal wearing a compliment. Lead with the category the buyer already has a budget line for and the outcome they already get measured on. Let the technology be the reason you deliver it better, not the pitch itself. This is the same discipline behind the fact that building the thing was never the part that decides whether anyone shows up.
The window is open. It just has a bouncer.
I want to be careful not to oversell the gloom here, because 61 percent of buyers say they evaluate startups and incumbents equally on the merits. That is the majority, and it means the game is winnable. The five to one gap is real but it lives among the buyers who arrive with a bias, and bias is beatable with proof.
That is the whole thing, actually. This is not a market that rewards the best product or the newest architecture. It is a market that rewards the vendor who makes it safest to say yes. Incumbents get that safety for free, from time served. You have to manufacture it, deliberately, with trials and references and business cases and boring contract answers, none of which are the work you enjoy.
The founders who win through 2028 will not be the ones with the most impressive model. They will be the ones who understood that in a room of 22 nervous people, the strongest feature you can ship is the absence of risk.
