The B2B Buyer Journey in the AI Era: From 27 Touchpoints to 3 Conversations

What does the B2B buyer journey look like in the AI era?

The B2B buyer journey has compressed from roughly 27 discrete touchpoints into three conversations: one with an AI assistant, one with peers in private channels, and one with the two or three vendors that survive to a formal evaluation. Most of the decision is made before a seller is contacted. What changed is not buyer intent but the cost of research, which has fallen close to zero and moved off the open web into synthesized answers.

In enterprise engagements we typically see 60 to 80 percent of the evaluation completed before any form fill, and first contact arriving with a vendor list already drafted. The buying committee still has 8 to 12 people, but they no longer each run their own research path. One or two members run prompts, share the output with the group, and the committee argues about a synthesized summary rather than a stack of downloaded PDFs.

This is not a smaller journey. It is the same journey with fewer observable surfaces. The touchpoints did not disappear; they were absorbed into a model answer, a private Slack thread, or a peer community post that no analytics platform will ever record. Marketing teams that measure only what they can see now report a shrinking funnel while pipeline stays flat or grows, which is one of the most common reporting conflicts we are asked to resolve.

Why did 27 touchpoints collapse into three conversations?

Three forces collapsed the journey: synthesis replaced search, private channels replaced public forums, and procurement compressed the evaluation window. Each one removes an entire class of touchpoint that marketers used to own, instrument and report on.

Synthesis is the largest factor. A buyer who once visited nine vendor pages, four review sites, two analyst summaries and a comparison article now asks one question and receives a paragraph naming three vendors. Eight to twelve page views become a single answer. The research still happened, but it happened inside a system that sends no referral traffic and publishes no record of which sources it weighed most heavily.

Private channels absorbed the second cluster. Peer validation moved out of public communities and open review threads into invitation-only Slack groups, WhatsApp circles, executive dinners and vendor-neutral peer networks. Across North America and the Gulf alike, we consistently find that the single most decisive input in an enterprise deal is a named peer at a comparable company who has already run the implementation, and that input is entirely invisible to attribution software.

Procurement compressed the third cluster. Enterprise buying committees under cost pressure have shortened evaluation windows from six or nine months to somewhere between 90 and 150 days. A shorter window forces the shortlist to be drafted earlier and with less deliberation, so fewer vendors get a hearing and the ones already familiar to the committee carry a structural advantage.

Three conversations now decide enterprise deals

Conversation one is with an AI assistant, and it decides who is considered at all. The buyer describes a problem in their own language, receives a synthesized answer naming three to six vendors, and treats that list as the starting universe. Vendors absent from the answer are not rejected on merit; they are never evaluated in the first place, which is a materially worse outcome than losing a bake-off.

Conversation two is with peers, and it decides who is trusted. Once a shortlist exists, the buyer validates it against people who have run the same implementation at similar scale. This is where category leaders lose deals to specialists, because a peer will name the vendor that solved their specific edge case rather than the vendor with the largest brand budget or the most complete feature grid.

Conversation three is with vendors, and it decides who wins. By the time a seller is in the room, the questions are commercial and technical rather than educational. The vendor that got the most useful and accurate information into conversations one and two arrives with a shorter cycle, fewer competitors on the table and measurably less discounting pressure at the close.

The practical implication is uncomfortable for most marketing organizations. Roughly two thirds of the enterprise marketing budget still targets conversation three, the one with the least remaining influence over the outcome, while conversations one and two are handled by whoever has spare capacity.

Four question types buyers put to AI assistants during evaluation

Buyers ask problem-shaped questions rather than product-shaped ones, and those questions fall into four predictable clusters. The clusters are definition, comparison, fit and risk, and each one rewards a different kind of published content.

Definition questions establish the category and its vocabulary. Comparison questions ask which vendors serve a specific segment, geography or compliance regime. Fit questions test whether a vendor works alongside a named technology stack, an existing ERP, or a data residency requirement in a regulated market. Risk questions ask what tends to go wrong, how long implementation genuinely takes, and what the total cost looks like in year two rather than year one.

In the enterprise content libraries we audit, comparison and fit questions show the highest correlation with real pipeline, yet the library itself is usually weighted heavily toward definition content. That imbalance explains a pattern we see often: a company gets cited constantly for category education and still never appears in the shortlist answer that determines who gets invited to bid.

Risk questions are the most underserved of the four clusters and the cheapest to fix. Publishing honest implementation timelines, known failure modes, and a clear statement of where your product is a poor fit gives assistants the specific, checkable language they need to place you accurately. In practice this disproportionately improves inclusion in fit and risk answers within one or two indexing cycles.

The Three-Conversation Map: a framework for the AI-era journey

We call this the Three-Conversation Map, and it replaces the linear funnel with three influence surfaces, each with its own owner, content type and metric. Building the map takes a mid-sized enterprise team roughly 8 to 12 weeks, most of which is spent on diagnosis rather than production.

The first surface is the machine conversation. Its owner is the SEO or AEO lead. Its content type is structured, extractable, entity-consistent material published on properties a model will actually read. Its metric is citation share across a fixed prompt set. We recommend tracking 40 to 60 representative prompts on a monthly cadence rather than chasing keyword rank, because rank and inclusion have largely decoupled.

The second surface is the peer conversation. Its owner is customer marketing or community. Its content type is named-customer proof, implementation detail and third-party review depth. Its metric is the share of deals in which a reference is requested and delivered inside five business days. Peer influence is slow to build, impossible to buy outright, and the most durable asset in the entire journey.

The third surface is the vendor conversation. Its owner is demand generation working directly with sales. Its content type is commercial and technical enablement. Its metric is cycle length and competitive win rate rather than lead volume. When the first two surfaces are healthy, the third gets shorter and cheaper, and that compression is the clearest available signal that the map is working.

Which journey metrics still work, and which ones broke

Pipeline metrics survived the transition and traffic metrics did not. Sessions, keyword rank and first-touch attribution now describe a shrinking and unrepresentative slice of a journey that mostly happens on properties you do not own and cannot instrument.

The measures that hold up are pipeline sourced by self-reported channel, the share of qualified opportunities that arrive with a vendor already named, the average number of vendors present in a competitive evaluation, and cycle length from first contact to signature. These describe the outcome of the journey rather than its now-invisible middle, which is exactly what a board-level report needs.

Two newer measures are worth adding to the standard set. Citation share tells you how often you appear in synthesized answers for the prompts your buyers genuinely use. Unattributed pipeline percentage tells you how much revenue arrives with no digital trail at all. In most enterprise programs we review, unattributed pipeline has moved from around 15 percent to somewhere between 35 and 55 percent across three years, and that shift is a market signal rather than a measurement failure.

Treat the self-reported source field on your primary form as a strategic asset rather than a nuisance. One open question asking how the buyer first heard of you, read manually every month by someone senior, will tell you more about the AI-era journey than any attribution model currently sold.

How enterprise teams rebuild the journey in 90 to 180 days

Enterprise teams rebuild the journey in three phases: a 30-day diagnostic, a 60-day content and structure correction, and a 90-day measurement rebuild. Running them sequentially rather than in parallel avoids the most common failure mode, which is publishing at volume before anyone knows which of the three conversations is actually being lost.

The diagnostic phase inventories the prompts your buyers use, tests how often you appear in the resulting answers, and interviews eight to ten recent wins and losses about how the vendor list was genuinely formed. Those interviews are uncomfortable and they are the only reliable input to everything that follows. Skipping them is the reason most AEO programs stall in month four.

The correction phase rewrites the content library against the four question clusters, fixes entity consistency so a model can reliably identify what your company does and for whom, and builds the customer proof assets that peer conversations depend on. Most teams find that 20 to 30 percent of the existing library can be rewritten rather than replaced, which keeps the budget defensible.

At Lemniscate Growth we run this sequence as a pipeline-first program, and the free AEO Checkers and GEO Scorers inside The GrowthGPT are a practical way to complete the diagnostic phase before any budget is committed. The point of the exercise is not to rebuild the 27 touchpoints. It is to accept that three conversations now decide the deal and to staff each of them properly.