Software buyers will self-serve a $25,000 purchase, and only 17% of vendors can take it

More than half of B2B software buyers say they would be comfortable completing a routine purchase worth $25,000 or more without speaking to a salesperson, and only 17% of software sellers have a self-service path in operation that could take the money. The figures come from Cleverbridge research reported by MarTech on Friday, the same day Search Engine Journal made a structurally identical argument about machine buyers: surfacing a product inside an AI answer is the easy half, and completing the transaction is where it breaks.

Software marketing has spent two years optimising for discovery — AI visibility, citations, shortlist presence. Both reports say the constraint has moved downstream of all of it.

Getting shortlisted matters more. Getting paid takes longer.

G2's 2026 Buyer Behavior Report, cited by MarTech, found more than 80% of buyers used AI chatbots for software recommendations over the past two years, with nearly half saying AI had its greatest influence during shortlisting and evaluation. The shortlist has become stickier: 80% of AI users bought from their initial shortlist in at least three of their last five purchases, against 65% of those who did not use AI.

That is where the good news stops. Nearly three-quarters of buyers in the Cleverbridge study say purchases are significantly delayed or abandoned because of internal approvals or back-and-forth with the vendor. Waiting for pricing or a quote delays 36%; negotiating terms delays 30%. Eighty-four percent would prefer a self-service digital path, and 96% of sellers agree routine transactions are moving that way — while 17% actually operate one.

The G2 figures give the delay a shape. Forty percent of buyers now say evaluation is the longest stage of the journey, up from 36% a year earlier. Once a vendor is chosen, security review holds up 39%, budget approval 32% and implementation planning 25%. Finance is arriving earlier too: its participation in buying committees rose from 31% to 46% in a year, and 49% of buyers said their CFO had reversed a purchase the buying team had already approved. Three-quarters expect positive ROI within six months of signing.

None of that is a procurement statistic. It is a brief for a pricing page. If a finance reviewer can veto a deal your champion already won, the material that survives the veto has to exist publicly, before the deal, in a form a stranger can forward.

The second buyer cannot wait for a quote at all

The SEJ piece is a contributor column by Greg Jarboe built on an email exchange with Konstantin Klyagin, who founded the QA agency QAwerk in 2015. Its subject is retail rather than SaaS, but the mechanism transfers. Jarboe notes that Shopify president Harley Finkelstein told the company's February 2026 earnings call that orders arriving through AI-powered search had grown 15 times since January 2025, routing through three protocols built inside a year: Google's Universal Commerce Protocol, OpenAI's Agentic Commerce Protocol and Salesforce's Agentforce Commerce.

Klyagin's argument is that an agent buys nothing like a person. It fires rapid structured API calls and decides in seconds — precisely the behaviour rate limiting and bot detection were built to stop. On one client funnel his team found two components each keeping their own copy of the same plan data, with small differences in price and attributes: invisible to a human, who would shrug and refresh, and fatal to an agent that selects on one source and is validated against the other. Jarboe's verdict on the sequencing is blunt: teams have "solved the part of the problem that was never actually the hard part."

A "contact sales" button is a hard stop for a machine buyer. It cannot negotiate, and it will not wait.

Put the two reports side by side and the same gate appears twice. The human buyer waits at the quote request and slips a quarter. The agent buyer cannot pass it at all — nothing on the other side is parseable, and there is no salesperson to email. Every fix that shortens the human path is also the fix that makes the path machine-legible.

Read the sources with their interests in view

Cleverbridge is a merchant of record for software and SaaS companies, selling checkouts, subscription management and quoting, so its research argues for the category it sells; read the self-service numbers as advocacy backed by data, not neutral measurement. MarTech is owned by Semrush. The SEJ column rests on a single vendor interview and reports no verified production incident of an agent breaking a checkout — a restraint worth noting, because this is a risk arriving, not a wave that has landed. What holds up across both is the direction, not the magnitude.

What to change this quarter

• Instrument the gap between "requested pricing" and "quote delivered". If nobody owns that number, the 36% delay is invisible inside your funnel.
• Move ROI proof, security documentation and implementation scope onto public pages. The CFO who reverses the deal is not on your demo call.
• Name the self-service ceiling deliberately. If buyers will transact at $25,000 and your checkout stops at $2,000, that is a packaging decision, not a payments constraint.
• Audit whether plan names, prices and entitlements match across your pricing page, your billing system and your API. One source of truth is a marketing requirement now.
• Test a purchase path with scripts, not just with humans, before adding another AI surface to the visibility roadmap.