The meeting is the formality
Begin with the deal you didn’t win.
A solicitor of fifteen years’ standing receives a query through her contact form on a Tuesday afternoon. The prospect — a manufacturing business with a contractual dispute worth, perhaps, eighteen thousand pounds in fees — has filled in the form competently and asked for a call. By Wednesday they have spoken. By Friday the engagement letter is signed. The solicitor counts this as a win for her firm’s responsiveness, and is broadly correct to do so.
What she does not see is what happened on Monday. The prospect’s commercial director, working from a list compiled by his finance colleague, opened nine law-firm websites in nine browser tabs. He spent under two minutes on most of them. On three he stayed longer, reading a single recent article and the partner biographies. By the time he reached our solicitor’s contact form he had already eliminated six firms, ranked the remaining three, and was completing what Forrester’s researchers, in their 2024 State of Business Buying, describe as “a process of confirmation, not selection.”5 The meeting on Wednesday was the ratification of a decision made on Monday afternoon.
“That 17% of purchase activity allocated to supplier interaction represents all suppliers — not each supplier.”— Brent Adamson, Harvard Business Review
This is now the structural condition of professional-services buying. Gartner’s much-quoted research, drawn from a survey of 750 B2B buyers and published by Brent Adamson in Harvard Business Review, finds that buyers spend only 17% of the entire purchase journey in direct contact with potential suppliers — and that 17% is divided across allsuppliers considered, leaving any individual firm with somewhere between 5% and 6% of the buyer’s attention.2By March 2026, Gartner’s most recent survey of 646 buyers reports that 67% now prefer a rep-free experience outright.3 McKinsey’s 2024 B2B Pulse, the ninth in an annual series spanning more than thirty thousand decision-makers, finds that 39% of B2B buyers will spend over half a million dollars through self-service channels, and one in five are willing to transact between one and ten million pounds remotely.4If buyers will commit half a million without a meeting, the sub-£2,000 decision is a foregone conclusion long before the contact form is opened.
Fifteen years ago, Google’s Jim Lecinski gave this moment a name. He called it the Zero Moment of Truth, defining it as the online research interval between stimulus and shelf where buyers now decide. In 2011 he found that shoppers consulted 10.4 information sources per decision, nearly double the previous year.1 The phrase has aged; the phenomenon has metastasised.
What the zero moment became
Lecinski’s original framing assumed a stimulus-and-search structure — an advertisement seen, a category researched, a purchase made. The model was already insufficient by the time Forrester began tracking the count of interactions per B2B purchase. Their longitudinal data records 27 interactions per buyer in 2021, up from 17 in 2019 and roughly twice the level of 2015.6 Of those 27, fifteen are digital and twelve human, but the ratio understates the asymmetry: the human interactions are short and supervised, the digital ones long and unsupervised. The buyer is alone with the website for almost all of the consideration.
Google’s own analysts, in Decoding Decisions (2020), abandoned the funnel altogether. Working with the behavioural-science consultancy The Behavioural Architects and simulating 310,000 purchase scenarios, Alistair Rennie and Jonny Protheroe described a “messy middle” — a non-linear loop of exploration and evaluation governed by six cognitive biases: category heuristics, the power of now, social proof, scarcity, authority, and the power of free. “The way people make decisions is messy,” they wrote, “and it’s only getting messier.”7 The funnel had been retired by the people who built it.
By 2024, 6sense’s research put the proportion of the buying journey completed before any seller contact at around 70%, with 80% of first contacts initiated by the buyer.8 TrustRadius’s 2024 B2B Buying Disconnect, surveying 2,164 technology buyers, found that 78% had shortlisted a product they had heard of before active research began — rising to 86% among enterprise buyers. Seventy-one per cent then bought their top-choice product.9 The shortlist was the decision; the rest was process.
What buyers’ brains are actually doing
The most useful thing to know about a website visit is that the visitor is not, in any ordinary sense, reading. A series of studies stretching back twenty years has established that the verdict on a website is reached in time spans below conscious deliberation.
“System 1 runs automatically and System 2 is normally in a comfortable low-effort mode.”— Daniel Kahneman, Thinking, Fast and Slow
Gitte Lindgaard’s group at Carleton University, publishing in Behaviour & Information Technology in 2006, demonstrated that visual appeal of websites can be reliably rated within fifty milliseconds — one twentieth of a second.10The follow-up study by Alexandre Tuch and colleagues at Basel, with Google’s Javier Bargas-Avila, extended the effect down to seventeen milliseconds. The Princeton psychologists Janine Willis and Alex Todorov, working on faces rather than pages, found that 100 milliseconds of exposure was sufficient to produce trait judgements — particularly of trustworthiness — that did not improve with longer looking; further exposure only raised the rater’s confidence. “It appears,” Todorov noted, “that we are hard-wired to draw these inferences in a fast, unreflective way.”11
This is the architecture Daniel Kahneman described in Thinking, Fast and Slow. System 1, the fast intuitive process, “continuously generates suggestions for System 2: impressions, intuitions, intentions, and feelings.”12 System 2, the slow deliberate process, ratifies. The buying decision is reached at the speed of System 1. The meeting on Wednesday is the deliberate System 2 audit of the verdict already in.
What System 1 is responding to is partly aesthetic, partly cognitive. Reber, Schwarz and Winkielman’s foundational paper on processing fluency, in Personality and Social Psychology Review, found that the more easily the brain processes a stimulus, the more positive its judgement — for beauty, but also for truth and trust.13 A cluttered homepage is not merely inconvenient; it is cognitively unpleasant, and that unpleasantness migrates to the firm. Choice architecture matters too: Iyengar and Lepper’s now-canonical jam study, conducted at Draeger’s grocery in Menlo Park, recorded conversion of 30% from a six-jam display against 3% from a twenty-four-jam display.14 Subsequent meta-analyses have shown the effect is context-dependent, but the direction is consistent: too many options paralyse. Most professional-services websites present every service line with equal prominence, and pay the conversion tax for doing so.
When the visitor finally does try to act, BJ Fogg’s behaviour model from Stanford applies: B = MAP. Behaviour happens when motivation, ability, and a prompt converge in the same moment.15The ten-field contact form is not a contact form — it is an ability suppressor. The vague “Get in touch” with no indication of what happens next is not a prompt — it is a wish.
Across all of this sits Antonio Damasio’s somatic-marker hypothesis and Read Montague’s neuroeconomics: the discovery, in the McClure et al. (2004) Neuronpaper on Coke and Pepsi, that brand knowledge changes not only expressed preference but ventromedial prefrontal cortex activity itself. Brand cues, the researchers wrote, have “a dramatic influence” on both behavioural preference and brain response, even when the product is chemically identical.16A credible website is not making a claim about competence. It is changing the buyer’s perception of every conversation that follows.
Competitive comparison: what buyers do with nine tabs open
The empirical work that comes closest to capturing the lived experience of B2B comparison is the meta-analytic literature on thin slices. Nalini Ambady and Robert Rosenthal, working at Harvard, aggregated 38 studies in their 1992 paper in Psychological Bulletinand found that judgements based on observations of under five minutes — sometimes under thirty seconds — predicted outcomes with an effect size of r = .39, with no significant improvement at longer durations.17 People are extracting signal at a startling rate, and additional information serves mainly to confirm.
This is the cognitive state in which competitive comparison occurs. The buyer opens nine tabs, and within four minutes has eliminated six. The eliminations are not principled; they are perceptual. A site loads slowly, and is closed: Google’s Need for Mobile Speed, based on aggregated analytics across 3,700 mobile sites, found that 53% of mobile visits are abandoned if a page takes more than three seconds to load.18A site looks dated, and is closed: the F-pattern reading work of Nielsen Norman Group, on a sample of 232 users, established that users scan rather than read, and that scanning patterns reward visual hierarchy and punish density. A site fails to declare its specialism, and the buyer’s category heuristic — Google’s first identified bias in the messy middle — assigns it to “general,” which on a shortlist of specialists is a death sentence.
What survives this triage is then compared, and it is here that the smaller firm’s structural disadvantage becomes acute. The Edelman-LinkedIn 2024 B2B Thought Leadership Impact Report, surveying 3,484 decision-makers in seven countries, found that 73% consider a firm’s thought leadership more trustworthy than its marketing materials, that 75% have been led by thought leadership to research a service they were not previously considering, and that 86% say strong thought leadership earns an RFP invitation. Only 15% rate what they actually read as excellent.19 The bar is low; the firms clearing it are disproportionately rewarded.
Professional services as a category
For owner-managed professional-services firms in the £1m–£5m bracket — solicitors, accountants, architects, independent financial advisers, consultancies — the situation is sharper still. Hinge Research Institute’s Inside the Buyer’s Brain(4th edition, 2022), drawing on responses from over 1,900 buyers and 3,610 sellers, recorded a 37% fall in buyer-perceived visibility of professional-services firms in just two years — from 23.1% in 2020 to 14.6% in 2022. “Technology and data issues” entered buyers’ top five challenges, rising 136%. Reliance on existing relationships as an evaluation method rose 79%.20 Translated: invisible firms are not considered, technically embarrassing firms are not believed, and the relationships that used to substitute for marketing are deepening, not multiplying.
“Buyer visibility of professional-services firms fell 37% in two years — invisible firms aren’t considered.”— Hinge Research Institute, 2022
Hinge’s 2025 High Growth Study, the tenth in the series, makes the elasticity concrete. The firms growing at 20% or more annually invest roughly twice as much in marketing as their peers and grow four times as fast, with margins up to thirty per cent higher. They prioritise educational content, differentiation, and digital maturity — precisely the levers that work via the website.21
The UK legal market makes the structural pattern visible. LexisNexis’s Bellwether 2025, in its thirteenth year, finds that 72% of small UK law firms plan organic growth, up from 40% in 2023, while 47% have no plans to invest in any technology — a contradiction that resolves only one way.22The Legal Services Board’s research with Pleasence and Balmer, summarised in the Law Society Gazette, finds that roughly one third of UK SMEs experience a legal issue but only 8% consult a law firm, and even then only 30% choose a solicitor — the rest go to accountants, HR advisers, or simply do nothing. Fewer than 15% view solicitors as cost-effective.23 Professional-services firms compete not only with each other; they compete with substitution and with inertia, both of which are silent and digital.
This sits inside a wider trust collapse. Edelman’s 25th Trust Barometer, surveying 33,000 people across 28 countries in late 2024, finds the UK at a Trust Index of 43 — among the five lowest of major economies — and 68% of respondents now believe business leaders deliberately mislead.24 In such a market, credibility cannot be assumed. It has to be demonstrated, and the website is the cheapest place to demonstrate it.
What frontier AI changes about analysing the moment
For most of the period covered by this argument, the zero moment was inaccessible to analysis. The buyer was alone with the website, by definition unobserved. Server logs counted footsteps; surveys recorded retrospective rationalisations; usability tests scaled poorly. The decision was being made in a room no one could enter.
That changed in late 2024. Anthropic’s release of Claude 3.5 Sonnet with Computer Use, on 22 October 2024, marked the first frontier model capable of operating a graphical interface in the way a human does — viewing a screen, moving a cursor, clicking, typing.25 On OSWorld, the benchmark for general computer-use agents, Claude scored 14.9% at launch, against roughly 70% for humans. By early 2026, on Stanford HAI’s reporting, successor models had reached the high-60s. OpenAI’s GPT-4o, released the same year, was the first model trained jointly across text, vision, and audio in a single network; Google DeepMind’s Gemini 2.5 added a million-token context window and native chain-of-thought reasoning. The Stanford HAI 2025 AI Index records that on benchmarks introduced in 2023, performance rose 18.8, 48.9 and 67.3 percentage points within a single year on MMMU, GPQA and SWE-bench respectively.26 Inference cost for a GPT-3.5-equivalent model fell roughly 280-fold between November 2022 and October 2024.
“Generative agents replicated real participants’ responses 85% as accurately as those people replicate themselves.”— Park et al., Stanford HAI, 2025
What this makes possible — and did not, two years ago — is the simulation of buyer perception at scale. The Park et al. Generative Agent Simulations of 1,000 People, published as a Stanford HAI policy brief in 2025, conditioned LLM agents on two-hour interviews with a thousand real participants and found that the agents replicated those participants’ General Social Survey responses with 85% of the consistency the participants themselves achieved two weeks later.27Synthetic personas are, on this evidence, not a marketing convenience but a methodologically defensible instrument. Subsequent work — Li, Chen, Namkoong and Peng’s 2025 paper “LLM Generated Persona is a Promise with a Catch” — is appropriately honest about the limits, particularly the tendency of ungrounded personas to confabulate, but the direction of travel is unmistakable.
The relevance to professional-services analysis is precise. The buyer’s first-impression verdict, formed in fifty milliseconds, is now reconstructable: a frontier model with computer-use can be instructed to approach a firm’s website as a specific buyer archetype would, register what is processed fluently and what is not, identify which credibility cues are present and which are absent, and produce a report not dissimilar in form to what an honest customer would say if pressed. The LLM-as-judge literature — Zheng et al. (2023), the MLLM-as-a-Judge work of Chen et al. (2024) — shows that frontier models match human evaluators most reliably when comparing two options pairwise. Which is, of course, exactly what buyers do.
This is not a substitute for talking to clients. It is a substitute for the assumption that one already knows what they see.
What the analysis actually surfaces
In practice, the patterns that emerge from synthetic-persona analysis of professional-services websites are unsurprising in aggregate and decisive in particular. The dominant failure mode is not aesthetic — most firms have presentable websites — but cognitive. The site says everything and therefore signals nothing. The biographies are written for peers, not prospects. The contact form, in Fogg’s terms, suppresses ability rather than enables it. The thought leadership, where it exists, is dated and undifferentiated, exactly the 85% that Edelman’s respondents describe as falling short. The credibility cues — case studies with named clients, demonstrable industry depth, evidence of having thought about the buyer’s specific problem — are absent or buried.
The 17% Gartner figure should be read in this light. It is not a complaint about buyer behaviour. It is a description of the room in which decisions are now made — a room that the seller has voluntarily left, but in which their website continues to speak on their behalf, often less well than they would speak themselves.
What follows from this
Three things follow. The first is that for owner-managed professional-services firms, the website is not a brochure. It is the principal sales conversation, conducted in the seller’s absence, against eight or nine competitors, at the speed of System 1. Treating it as a brochure is the most expensive mistake the £1m–£5m firm can make, and it is the modal mistake.
The second is that buyer-side perception, which has historically been inaccessible to small firms because primary research is prohibitively expensive, is now within reach. The combination of computer-use models, multimodal reasoning, and validated synthetic personas means that for the first time, a firm without a research budget can ask an empirically defensible version of the question: what does our shortlister actually see?
The third is that the gap between what buyers experience and what sellers believe they experience — TrustRadius’s “buying disconnect” — is wide and stable. The eighth annual edition of their report makes the point in numbers: vendors believed 34% of buyers spoke to a current user before purchase; in fact, 56% did, and 71% among enterprise. This gap is not closed by trying harder. It is closed by looking through the buyer’s eyes.
Fifteen years after Jim Lecinski named the zero moment of truth, the moment has not loosened its grip on the outcome. It has tightened it. The instruments for understanding it have, finally, caught up.