The Buyer Sends an Agent: AI Will Shape 75% of B2B Purchases, and Unprovable Claims Will Lose Deals

For most of the history of B2B selling, the first person to read a proposal was a person. That is changing faster than most revenue plans assume. Buyers are putting AI agents to work on supplier discovery, comparison, negotiation and renewal analysis, and the agents are not impressed by adjectives. They look for claims they can check. A pitch that cannot be verified is not just weaker in front of an agent, it may never reach the human who signs the contract.

Three-Quarters of Purchases, One Big Caveat

Forrester's 2027 B2B go-to-market predictions, released October 21, put a number on the shift: AI agents will influence 75% of B2B purchase decisions. Forrester expects buyers to use agents across the whole process, from finding suppliers to analyzing contract renewals. The caveat matters for revenue leaders: truly autonomous buying will stay concentrated in high-volume, low-risk, repeatable purchases. Complex, high-value deals will still close with people, but those people will arrive having read whatever their agents surfaced first.

The more pointed prediction is about losing. Forrester expects two-thirds of providers to lose deals because buyers cannot verify their claims, and says go-to-market teams will need to structure proof points that AI tools can find and validate. It also predicts buyers will double their human interactions to validate AI-generated recommendations, turning to provider teams, analysts, customer references and partners because AI cannot fully judge fit, feasibility, trust or accountability. Sharyn Leaver, Forrester's chief research officer, framed the challenge as making "deliberate choices about what to automate." For sellers, the implication cuts both ways: the agent screens the claim, and a human then checks it with someone they trust.

Adoption is already broad. MarketScale, citing IDC research, reported in August that roughly 80% of B2B technology buyers use AI agents in their purchasing decisions. Most of that use assists people rather than replacing them, but it means the research phase of a deal is now partly conducted by software the seller never meets.

Procurement Is Arming Faster Than It Reorganizes

On the buying side, the ambition is ahead of the operating model. Gartner's October 8 release, drawing on a survey of 213 senior procurement leaders and a C-suite AI survey of 1,303 respondents including 134 procurement leaders, both concluded in April 2026, found that 63% of senior procurement leaders expect AI to significantly improve procurement performance over the next three years. Only 25% have undertaken job and role redesign as a direct result of AI implementation. "Successful AI implementation at the functional level is fundamentally a role redesign challenge," said Fareen Mehrzai, a senior director analyst in Gartner's Supply Chain practice.

That gap shapes what sellers will face across the table. Some of the largest buyers are already live. PYMNTS reported in July that Walmart, Maersk and Vodafone are running AI negotiators at scale, and that the agreements those agents reach are binding. The same report described governance that has not caught up: compliance methods remain retrospective, and the OECD notes that real-time oversight frameworks for agentic AI are still absent. A procurement function that buys agents before it redesigns roles is likely to hand them rigid targets and little human supervision, which is exactly when a seller's pricing discipline gets tested.

What Wins Against a Machine

The best evidence on how agents negotiate comes from MIT Sloan. In a tournament described in June, participants from more than 40 countries built AI negotiation agents that competed in over 180,000 unique negotiations, ranging from simple buyer-seller deals to multi-issue contracts. Warmth was consistently associated with better outcomes. An agent designed to be purely ruthless faced many walkouts that hurt its ability to claim value, while one that paired warmth with a focus on value claiming was effective at both reaching deals and capturing value. "To be successful in negotiations with an AI agent, you may still need to act like a human," said Jared Curhan, a professor of management at MIT Sloan.

The study also found that AI-specific tactics, including prompt injection and structured chain-of-thought preparation, were effective in agent-to-agent talks. The top agent, NegoMate, prepared with structured reasoning before each of its nearly 400 negotiations. For revenue teams, two lessons follow. Consistency now matters as much as charm, because an agent will apply the same logic to every quote it sees. And the seller's own systems are part of the attack surface: a counterparty agent probing for floor prices or discount authority will find whatever the seller's process leaves exposed.

The CRO Playbook for Agent-Mediated Buying

• Make every claim checkable. Pair each performance claim in proposals and product pages with a source an agent can verify, such as a published benchmark, a named reference or a documented outcome, since two-thirds of providers are expected to lose deals on unverifiable claims.
• Lock pricing logic before an agent probes it. Set discount floors, approval thresholds and counteroffer rules in the quoting system so a buyer agent receives the same disciplined answer every time, not whatever a rep concedes under pressure.
• Invest in the human validation moment. With buyers expected to double their human interactions to check AI recommendations, make references, solution experts and executive sponsors easy to reach late in the cycle.
• Map where the buyer's agent sits. Add a qualification question about how the customer's procurement team uses AI, and plan the deal differently when an agent runs sourcing or first-round negotiation.
• Train reps for warmth and consistency. The MIT Sloan results favor rapport plus value claiming over hard bargaining, so practice both in role play, including scenarios against an automated counterpart.
• Audit what your own agents reveal. Test customer-facing chatbots and seller agents for leaks of internal pricing, discount authority or deal strategy, since prompt injection works between agents as well as against them.

The buyer's agent is not replacing the buyer, at least not in the deals that matter most to a CRO. It is changing what reaches the buyer and in what order. Revenue teams that make their claims verifiable, their pricing consistent and their people easy to validate will be the ones the agent recommends and the human then trusts. Those still relying on persuasion alone may find the conversation ended before anyone on their side was in the room.