.svg){kind=logo}
Every post on agentic buying recently opens with the same number. By 2028, Gartner says, 90% of B2B buying will be AI agent intermediated, pushing more than $15 trillion of spend through AI agent exchanges. It is an alarming figure, and it is behind the selling of a great deal of procurement software this year, but it has also been widely misunderstood.
AI purchasing agents are genuinely changing how B2B buying works, but not in the way that headline implies, and the difference matters enormously if you are the person accountable for pipeline.
By the end of this you will know what that forecast actually counts, where agentic buying is really biting in 2026, and the three things worth fixing in your paid media and measurement setup before the year is out.
The 90% figure does not mean machines are buying your software
Read the wording rather than the headline. "Intermediated" is carrying almost all the weight in that sentence. It counts any purchase where a buyer ran a chatbot query during research, where a procurement platform re-ranked a supplier list, or where an internal copilot summarised an RFP. A buyer asking ChatGPT to compare four vendors is intermediated buying under that definition. So is a category manager pasting a longlist into Copilot. Neither is an autonomous agent signing a contract.
Now put the largest buyer study of the year next to it. Forrester's 2026 Buyers' Journey Survey covered nearly 18,000 global business buyers and found that 94% used AI somewhere in their purchase process. The next finding is the one nobody quotes: buyers do not trust what it gives them. AI search tools deliver incomplete or unreliable information often enough that buyers compensate by validating outputs against peers, product experts and industry analysts. Roughly one in five said they felt less confident in their decision after using generative AI.
The rest of that research points the same direction. The typical B2B buying decision now involves 13 internal stakeholders and nine external influencers. Procurement professionals are decision-makers in 53% of buying cycles and get involved early. More than 60% of buyers now run a trial before committing, rising to 78% on purchases above $10 million. That is not a market drifting toward autonomous machine procurement. It is a market that has become slower, more collaborative and considerably more sceptical.
The agent is not signing your contract. It is deciding whether you make the list of contracts worth considering.
What agentic buying actually looks like in 2026
There is a real autonomous layer, and it is worth being precise about where it lives. Tail spend, replenishment and routine reorders are genuinely being handed to agents, because the supplier, the pricing and the terms were already negotiated. The agent executes inside rules a human set. Forrester expects around a fifth of B2B sellers to face agent-led quote negotiations during 2026, which is a real operational shift for anyone selling components, consumables or anything with a catalogue and an API.
Considered purchases behave completely differently. When there is an annual contract, a security review and thirteen stakeholders involved, the agent does not take over the transaction. It takes over the front end. Forrester found 55% of buyers now compare vendors inside AI tools, 54% research products there, and 47% build their internal business case there, all before any vendor contact at all.
Picture how that plays out. A procurement analyst asks an AI assistant for the leading vendors in a category, gets a scored shortlist against their stated requirements, drops it into a slide, and circulates it internally. Nobody visited your site. Nobody filled in a form. Your sales team has no idea the evaluation is happening. G2's research this year found roughly half of B2B software buyers now start with an AI chatbot, that 69% ended up choosing a different vendor from the one they had originally planned on, and that a third bought from a vendor they had never previously heard of. Those last two numbers are the interesting ones, because they cut both ways. Incumbency is worth less than it was. So is obscurity.
Where this actually hits your pipeline
The first hit is positional. If the shortlist forms inside an AI tool during a research phase you cannot see, then your presence in that tool's answer is a pipeline input, not a branding nicety. There is no follow-up call to recover a miss you never knew about.
The second hit is measurement, and it is the one costing agencies and in-house teams money right now. AI-referred traffic is growing fast and arriving qualified. Demandbase's platform data showed monthly ChatGPT-referred visits to the B2B properties it measures rising from around 645,000 in June 2025 to 2.6 million in June 2026, a 303% increase, with a sharp inflection in May 2026. The problem is that most of that demand does not present itself honestly in your analytics. Mobile assistant apps strip referrer data, so those sessions land in direct. GA4's default AI Assistant channel grouping helps, but it does not cover every engine, and Perplexity traffic still routes to generic referral. A buyer who was recommended by an assistant and then searched your brand name shows up as branded organic, and your paid brand campaign quietly takes credit for demand it did not create.
The third hit follows from the second. If AI assistants are generating branded demand you cannot see, and your brand campaigns are capturing that demand at a flattering cost per acquisition, you will systematically over-invest in capture and under-invest in the thing that created the demand. That is a familiar failure mode dressed in new clothes, and it gets expensive at five and six-figure monthly budgets across several markets.
What to fix before the end of the year
Start with AI visibility and GEO, because you cannot argue about budget without it. Build custom channel groupings that separate AI-referred sessions from organic and direct, cover the engines GA4 misses, and use server-side tagging to catch sessions before referrer data is lost. Most B2B accounts see a meaningful slice of traffic reclassify out of direct the moment they do this, and that reclassification is usually the highest-intent slice.
Then connect it to revenue rather than sessions. An AI-referred visit that never becomes pipeline is a vanity metric, and the conversion rates being reported for this channel are high enough that people are getting careless with them. Push the source through to your CRM, import closed-won outcomes back into the ad platforms as offline conversions, and report AI-influenced demand on the same CAC and ROAS basis as everything else. This is exactly the discipline that separates multi-market paid media measurement from channel-level reporting, and the same plumbing that makes smart bidding trustworthy on Google's newer campaign types makes AI-influenced demand legible.
Finally, defend the ground where the agent hands the buyer back to you. When someone leaves an AI assistant having been told about four vendors, their next action is usually a branded or comparison search. If a competitor owns that moment in Germany or Singapore while you only cover the UK, you funded the research and someone else banked the deal. Comparison and alternative-to terms deserve real budget and real coverage across every market you sell into, not the leftovers.
None of this requires believing that autonomous agents will be signing enterprise contracts by 2028. It only requires accepting what is already measurable: that a large share of your evaluation now happens somewhere your analytics cannot follow, and that the vendors who close that gap first will be the ones who can prove which spend is actually building pipeline.
If you want to see what that gap looks like inside your own account, book a discovery call today and we will walk through it with you.
