The B2B buying committee glossary
Nine terms, one page. Each one has a definition you can quote, the mechanism underneath it, and the test that tells you whether it is happening to you.
B2B deals are decided by a committee, not a lead. These nine terms name the parts of that committee that analytics cannot see: the state a buyer arrives in, the veto nobody voices, the handoff that snaps between roles, and the seam between the ad and the page where most spend is lost.
Behavioural state
Also called: Fogg state, audience mindset, buyer mindset.
Behavioural state is the headspace a buyer is in at the moment they encounter your page. distracted, sceptical, or actively shopping. It comes from the Fogg Behaviour Model, which holds that an action occurs only when motivation, ability and a prompt converge. The same person, reading the same page, reaches a different verdict in each state, so testing with only one state tells you almost nothing about how the page will actually perform.
The three states
- Distracted. Multitasking, scrolling, low attention. Reads headlines only and will not scroll to find what they need. Everything below the fold is close to invisible.
- Sceptical. Interested, but burned by vendors before. Wants numbers, named customers, and mechanism. Discounts adjectives entirely.
- Ideal-Desperate. Actively shopping for a solution to this exact problem right now. Reads properly, forgives a weak hook, converts on a relevant page.
Everyone recognises the third one because it is who we picture when we write. It is also the rarest.
Why the mix matters more than the states
The states are only useful with a population share attached, and that share is a property of the channel, not of your product. Derived from published CTR benchmarks and the Ehrenberg-Bass 95:5 rule, which caps buyers who are actively in-market at any moment at roughly 5%. The distribution looks approximately like this:
| Traffic source | Distracted | Sceptical | Actively shopping |
|---|---|---|---|
| Google Search, retargeted | 15% | 20% | 65% |
| Google Search, cold | 30% | 25% | 45% |
| LinkedIn lead-gen form, warm | 45% | 20% | 35% |
| Email, warm list | 40% | 25% | 35% |
| Email, cold | 78% | 17% | 5% |
| LinkedIn sponsored, cold | 82% | 15% | 3% |
| Display, cold | 85% | 12% | 3% |
The right-hand column is the single most predictive variable for how a page performs, and it swings by more than twenty-fold across channels. A page that converts search traffic can fail completely on cold social without a single word changing.
The reviewer problem
Now sort the people who approve your page. Motivation: maximal, their name is on it. Familiarity: maximal, they know the category and the roadmap. Attention: maximal, it is a calendar event with the page on a screen.
Every reviewer is in the Ideal state. On cold social, roughly 3% of the audience is. Internal review tests with the 3% and ships to the 97%, which is a coverage problem rather than a competence one and cannot be fixed by reviewing more carefully.
The same applies to a general AI assistant. Paste a page into ChatGPT and it reads carefully, patiently, all the way down, cooperatively. The most attentive reader available. Useful for drafting. Not representative of anyone you are buying traffic from.
Distraction is not a small penalty
It is tempting to treat a distracted reader as an ideal reader with less patience. The arithmetic is harsher. In WhyUser’s engine, a Distracted persona gains only 0.7× on anything that resonates, while friction penalties are amplified by dividing by the same factor. Roughly twice as hard to convince and twice as easy to lose.
Two consequences follow, and both change what you build:
- Above the fold has to close on its own. Not tease, not set up. close. For 82% of cold social traffic the fold is the entire page.
- Proof beats polish. A sceptical reader discounts adjectives to zero. A number, a named customer, or a mechanism survives the discount.
How WhyUser uses it
Every committee role is run across all three states, and the seeds are allocated by the measured distribution for the channel you are actually buying, not evenly. Running the three states one-third each would over-represent the shopping buyer by more than tenfold on cold social and produce simulated click rates in the 20–30% range, which is fiction.
Weighting to the real mix brings simulated rates into the range published benchmarks actually show. The distribution used is stamped onto every run, so you can check which one produced your result, and you can override it with your own measured numbers once you have them.
Questions people ask about behavioural state
What is a behavioural state in marketing?
It is the headspace a buyer is in when they meet your page: distracted, sceptical, or actively shopping. It comes from the Fogg Behaviour Model, and the same person reaches a different verdict about the same page in each state.
What share of B2B traffic is actively in-market?
Roughly 5% at any moment, per the Ehrenberg-Bass 95:5 rule. On cold LinkedIn or display it is nearer 3%; on retargeted search it can reach 65%. That single share is the most predictive variable for how a page will perform.
Why do internal reviewers not represent real buyers?
Because they sit at maximum motivation, maximum familiarity and maximum attention, which is the ideal behavioural state. On cold social traffic roughly 3% of the audience is in that state, so the review samples the 3% and the campaign ships to the 97%.
How much harder is a distracted buyer to convert?
Roughly twice as hard. In WhyUser’s engine a distracted persona gains 0.7 times on anything that resonates while friction penalties are amplified by the same factor, so above-the-fold content has to close on its own.
Should I test my page against every behavioural state equally?
No. Weighting the three states evenly over-represents the shopping buyer by more than tenfold on cold social and produces unrealistic conversion estimates. The distribution should match the channel you are buying.
Buying committee simulation
Also called: committee simulation, multi-persona simulation.
Buying committee simulation is a pre-launch testing method in which each role on a B2B buying committee is modelled as a separate agent that reads a page, ad, or email and decides independently whether to continue or abandon. Instead of one aggregate “does this convert” score, it returns a verdict per role, the section where each role dropped, and the specific evidence that role needed and did not find.
The five seats
A committee simulation is only as useful as the seats it models. The standard set for technical B2B software:
- Champion. wants the change, has to sell it internally, needs ammunition more than information.
- Economic buyer. signs. Needs a baseline, a number, and a defensible comparison.
- Technical decision-maker. owns the integration. Needs architecture, limits, and failure behaviour.
- End user. lives with it daily. Needs to see the actual workflow, not the value proposition.
- Security guardian. can stop the deal alone and gains nothing by approving it. Needs certifications, data handling, and residency stated plainly.
You pick the subset that matches your real deals. A three-seat committee is common at Series A; five is common at 500 employees.
Why one agent per role, not one agent playing five parts
Asking a general AI assistant to “review this page as a CFO, a CISO and an engineer” produces one voice wearing three hats. The answers correlate, because they come from one pass of one model holding all three briefs at once. Real committees do not correlate. The champion’s enthusiasm is exactly what makes the security reviewer’s silence invisible.
Running one agent per role, with no knowledge of the other roles’ verdicts, is what makes disagreement possible, and disagreement is the finding. When the champion advocates and the economic buyer abandons at the same section, you have located the deal risk. That structure is drawn as a conflict graph.
What a run produces
- A per-role decision at each scan zone: scroll, click, submit, or abandon.
- The role that vetoed, and the section where it happened.
- The specific missing evidence, sourced to something a real buyer said.
- A ranked fix list, ordered by how many roles each fix unblocks.
Every role is run across multiple behavioural states. distracted, sceptical, ideal, and many seeds, so what comes back is a distribution rather than one opinion. WhyUser fields around 300 agents across five roles per run.
What it does not replace
Being straight about the limits is what makes the rest credible.
- It does not replace talking to customers. A simulation compiled from buyer evidence is downstream of that evidence. If the evidence is thin, so is the simulation.
- It does not produce absolute conversion rates. The defensible output is rank and direction: this version beats that one, this role is the risk. Treat any magnitude as a range.
- It does not replace live testing. It runs before the spend so that live testing starts from a better asset.
Questions people ask about buying committee simulation
What is a buying committee simulation?
It is a pre-launch test in which each role on a B2B buying committee is modelled as a separate agent that reads the asset and decides independently. The output is a verdict per role, not one blended score.
How is it different from asking ChatGPT to review my landing page?
A general assistant returns one confident opinion, and returns a different one each time you ask. A committee simulation runs each role as its own agent across several behavioural states and many seeds, so disagreement between roles becomes visible and the same input returns the same verdict.
Are the simulated buyers real people?
No. They are agents compiled from real evidence: customer reviews, sales-call transcripts, community threads, and competitor comparisons. The evidence is real; the buyer is simulated.
How long does a run take?
About 30 minutes for a landing page, ad, or email, compared with roughly two weeks for an internal review round and four to six weeks for a live A/B test.
Conflict graph
Also called: committee physics, influence map, veto map.
A conflict graph is a map of how the members of a buying committee influence each other’s decision on one specific asset: who advocates, who hesitates, who vetoes, and which direction the influence flows. It turns a blended conversion number back into the disagreement that produced it, which is where the actionable fix always sits.
Why an average hides the answer
Take a five-role committee where three roles convert enthusiastically and two abandon. The blended number is 60%, which reads as “decent page, needs work.” That reading is wrong in a specific and expensive way.
In a committee purchase the two abandonments are not 40% of the outcome. If one of them is the security guardian, they are 100% of it. Consensus purchases are conjunctive: every seat has to clear, so the weakest role sets the ceiling. Averaging is exactly the wrong operation.
As an average: 60% positive. Ship it, iterate later.
As a conflict graph: Champion and technical decision-maker advocate. End user is neutral. Economic buyer hesitates at the pricing section. Security guardian vetoes above the fold. The deal cannot close in this state, whatever the average says.
What the graph shows
- Position per role. advocate, neutral, hesitant, or veto.
- The trigger. The section and the missing evidence that produced each position.
- Influence direction. whose position moves whose. A champion cannot carry a security veto, but a resolved security objection frequently unlocks a hesitant economic buyer.
- The unblocking fix. The single change that moves the most positions, which is rarely the change that moves the average most.
How to read one
Work the graph from the veto inward, never from the average down. Three rules:
- Fix vetoes before you optimise advocates. Making the champion more enthusiastic does nothing if security is blocking.
- A hesitation behind a veto is often not real. Resolve the veto and re-run before spending anything on the hesitation.
- Watch for the fix that helps one role and hurts another. Adding compliance detail above the fold can clear the guardian and bury the end user. The graph shows the trade before you ship it.
Questions people ask about conflict graph
What is a conflict graph in B2B marketing?
It is a map of how each buying committee role reacts to one asset, showing who advocates, who hesitates, who vetoes, and which way influence flows between them.
Why not just use an average conversion score?
Because committee purchases are conjunctive: every seat has to clear, so the weakest role sets the ceiling. A 60% average can describe a page that cannot close a single deal because the security reviewer vetoes above the fold.
How do you act on a conflict graph?
Work inward from the veto. Fix blocking roles before optimising for roles that already advocate, and re-run before spending effort on hesitations that sit behind a veto.
Dark funnel
Also called: dark social, untracked buying journey.
The dark funnel is the part of a B2B buying process that leaves no trace in your analytics: the forwarded link, the Slack thread, the peer DM, the podcast, the AI assistant that summarised your page without anyone visiting it. Most committee members never touch a tracked surface, so the majority of the evaluation happens somewhere your reporting cannot see.
Why it got darker in 2026
The dark funnel used to mean peer conversations and forwarded PDFs. A newer layer now sits on top of it: buyers ask an AI assistant about you before they visit you, and often instead of visiting you. The assistant reads your page, summarises it, names alternatives, and hands the buyer a position. all without a session in your analytics.
The practical consequence is that the first version of your pitch that a committee member sees is frequently not written by you. It is a summary, made by a model, from whatever it could extract from your site and the wider web.
What this changes about your page
Two readers now matter, and they read differently.
- The human reads the first screen, scans, and decides in seconds. Design, rhythm, and proof carry this.
- The model reads text. It cannot see an icon, a chart image, or a checkmark rendered as a glyph. If your strongest differentiation is expressed only as a visual, it does not exist to the reader who summarises you.
A page can be excellent for one and empty for the other. Testing for that gap is cheap; discovering it from a flat quarter is not.
How to work with a dark funnel instead of against it
- Stop trying to attribute it and start asking. The most reliable instrument is one question on every sales call: “how did you first come across us, and did you look us up anywhere before replying?”
- Write for the forwarded read. Assume the person who decides arrives cold, alone, through a link with no context. Every asset should stand on its own without the champion’s narration.
- Make the machine-readable version match the human one. Anything you would say out loud to a buyer should exist as text somewhere on the page.
- Test the roles you cannot see. A buying committee simulation is a way of putting the untracked reader back into the review, because the roles that hide in the dark funnel are exactly the ones that cast the silent veto.
Questions people ask about dark funnel
What is the dark funnel in B2B marketing?
It is the part of the buying process that leaves no trace in analytics: forwarded links, Slack threads, peer conversations, and AI assistants that summarise your page without anyone visiting it.
Why does the dark funnel matter more now?
Because buyers increasingly ask an AI assistant about a vendor before visiting the site, and sometimes instead of visiting it. The first version of your pitch a committee member sees is often a model-written summary rather than your page.
How do you measure the dark funnel?
Mostly you do not measure it, you ask about it. A self-reported question on every sales call about how the buyer first heard of you, and whether they looked you up anywhere, is more reliable than any attribution model.
Evidence ledger
Also called: sealed claim ledger, prediction track record.
An evidence ledger is a time-locked record of predictions. Every finding a simulation makes is written down and sealed before the campaign runs, then graded HIT or MISS against what actually happened. Because the claim is sealed before the outcome exists, the resulting accuracy rate cannot be reconstructed after the fact, which is the only thing that makes a vendor’s accuracy claim worth anything.
The problem it solves
Any vendor can claim an accuracy number. Almost none can show you the losing predictions. The standard move is to collect outcomes first, then describe which ones the model “would have” called, a claim that is unfalsifiable and therefore worthless to a sceptical CFO.
Sealing fixes this. The prediction is written, timestamped, and locked at the moment the simulation runs, before the campaign is live and before any outcome exists. When the result arrives, the only remaining action is grading. Nothing can be edited, re-scoped, or quietly dropped.
How grading works
- The customer grades, not the vendor. Each sealed claim is marked HIT, MISS, or CAN’T-GRADE by the customer’s team against their own data.
- Misses stay in the denominator. A ledger you can remove entries from is not a ledger.
- Can’t-grade is a real outcome. Some predictions cannot be checked against available evidence, and pretending otherwise inflates the rate.
- Grades feed back. Confirmed findings raise the weight of that pattern for that account; misses lower it. The model gets sharper for that customer specifically.
WhyUser’s current published position: 473 claims graded, 84% accuracy across live accounts, with the method and the known blind spots on the same page.
Why this is the part that compounds
The simulation itself is reproducible. Anyone with a capable model and a few weeks can build a version of it. What is not reproducible is a record of predictions sealed at specific times and graded against real outcomes, because that record can only accumulate at the speed of real campaigns. A competitor starting today cannot buy last year’s sealed claims.
For the buyer, this matters for a plainer reason: it is the artefact you take to your CFO. Not “the tool says the page is broken,” but “here is a record of what it predicted before the campaign ran, and here is how often it was right, graded by us.”
Questions people ask about evidence ledger
What is an evidence ledger?
It is a record of predictions sealed before an outcome exists, then graded HIT or MISS against what actually happened. Sealing before the fact is what makes the resulting accuracy rate verifiable.
Who grades the claims?
The customer, against their own data. A vendor grading its own predictions is not evidence, and misses stay in the denominator.
Why does sealing matter?
Because a prediction described after the outcome is unfalsifiable. Sealing at the moment of the run means the claim cannot be edited, re-scoped, or quietly dropped once the result is known.
Pre-launch stress test
Also called: pre-flight test, campaign staging, marketing staging environment.
A pre-launch stress test is a review step that runs a campaign asset against simulated buyers before any budget is spent, in the same way engineering runs code against a staging environment before production. It answers “what will break, and for whom” at the one moment when changing the asset still costs nothing.
Every other function has one. Marketing does not.
Engineering has staging and CI. Finance has audits. Sales runs role plays before the big meeting. B2B marketing still tests in production: launch the campaign, wait six weeks, read the damage in a dashboard, and by then the budget is spent and the quarter is over.
The asymmetry is not about rigour. It is about the cost of being wrong arriving after the money leaves.
Where it sits in the workflow
A stress test is one new step, not a new process. It goes between review and launch:
- Plan. brief and ICP
- Build. copy, design, ad creative
- Review. internal alignment
- Stress test. The asset meets the committee, roughly 30 minutes
- Launch. push live, allocate spend
- Measure. live performance and A/B
Internal review catches typos, brand drift, and legal risk. It does not catch a silent veto, because everyone in the room already believes in the product.
What gets stress tested
- Landing pages, which role abandons, at which section, and what proof was missing.
- Ad campaigns. whether the page delivers the promise the ad made. See scent trail.
- Email campaigns, which subject and CTA win for each role, and whether the destination holds the promise.
- Content and technical assets. whether the reader the piece was written for is the reader it actually fits.
The honest limitation
A stress test predicts rank and direction, not absolute performance. It will tell you that version B beats version A and that the economic buyer is your risk. It will not tell you your conversion rate will be 4.2%. Anyone selling you the second number is selling you a precision they cannot have, and one wrong magnitude destroys the credibility of everything else in the report.
WhyUser publishes its own hit rate for exactly this reason. See the accuracy and track record page: 473 claims, each sealed before the outcome existed, graded by the customer against their own data.
Questions people ask about pre-launch stress test
What is a pre-launch stress test in marketing?
It is a step that runs a campaign asset against simulated buyers before any money is spent, so problems surface while the asset can still be changed for free. It is the marketing equivalent of a staging environment.
How is it different from an internal review?
Internal review catches typos, brand drift, and legal risk, and everyone in the room already believes in the product. A stress test introduces adversarial readers who have no stake in approving it.
Does a stress test replace A/B testing?
No. A/B testing measures real traffic and is the only thing that proves live performance. A stress test runs before the spend so the version you put into the A/B test is already better.
How long does a pre-launch stress test take?
Around 30 minutes per asset, against roughly two weeks for an internal review round and four to six weeks for a live A/B test to reach significance.
Scent trail
Also called: information scent, message match, ad-to-page continuity.
Scent trail is the continuity between the promise a buyer clicks and the content they land on. When an ad, subject line, or link makes a specific promise and the destination leads with something else, the scent breaks and the visitor abandons within seconds. even when the ad and the page are each individually good. The failure lives in the seam, which is why testing either asset alone never finds it.
Where the term comes from
Scent trail comes from information foraging theory, which models a person searching for information the way an animal forages for food: following the strongest available cue, and abandoning a trail the moment the scent weakens. It has been a standard concept in usability research for over two decades. What is new is how expensive it has become, now that a click costs real money in B2B paid media.
Why this failure survives review
Ads and landing pages are usually reviewed separately, often by different people, sometimes by different teams. Each is approved on its own merits. Nobody is accountable for the seam between them.
The result is the most common and least diagnosed form of wasted spend in B2B: a well-targeted ad pouring qualified clicks into a page that was never going to convert those clicks. Cost per click looks fine. Cost per opportunity does not. The dashboard blames the page, the page team blames targeting, and the actual defect. The mismatch. is nobody’s line item.
Ad: “Cut your incident triage time in half.”
Page H1: “The unified observability platform for modern engineering teams.”
Both are competent. The buyer arrived holding a specific promise about triage time and landed on a category statement. Nothing on the first screen confirms they are in the right place, so they leave, and the page never gets to make its case.
How to test the seam
Test the pair, not the parts. A simulation loads the ad or email first, forms the expectation a real buyer would form from it, then loads the destination and checks whether that expectation is confirmed on the first screen. What comes back is a per-role continuity read: the technical decision-maker may find the scent intact while the economic buyer does not, because they arrived holding different promises from the same creative.
This is what WhyUser’s ad campaign simulation and email campaign simulation measure.
The cheapest fix in demand gen
Scent repair is usually a headline and a subhead, not a redesign. Echo the ad’s exact promise in the page’s first screen, in the buyer’s words rather than the category’s. It is close to free, and it is the fix with the shortest distance between change and result.
Questions people ask about scent trail
What is a scent trail in marketing?
It is the continuity between the promise a buyer clicks and what they land on. If the destination leads with something other than the promise, the buyer abandons within seconds even though both assets are individually good.
What is the difference between scent trail and message match?
They describe the same failure. Message match usually refers to matching the ad headline to the page headline. Scent trail is broader: it covers the whole expectation a buyer forms before clicking, including who they think the product is for.
Why does a broken scent trail not show up as a problem?
Because the ad and the page are reviewed separately and each passes on its own merits. The defect lives in the seam between them, which no single review owns.
How do you fix a broken scent trail?
Echo the ad or subject line promise in the first screen of the destination, in the buyer’s words rather than the category’s. It is usually a headline and subhead change, not a redesign.
Silent veto
Also called: quiet no, unspoken objection, committee veto.
A silent veto is a B2B purchase that dies because one member of the buying committee decides against it and never says so out loud. The champion stays enthusiastic. The deal quietly stops progressing. Because no objection is ever spoken or logged, the marketing team never learns which piece of missing proof killed it, and the same gap stays on the page for the next campaign.
Why silent vetoes happen
Enterprise software is bought by a committee and killed by an individual. The person who clicked your ad is rarely the person who has to sign, secure, or operate the thing. Three structural facts make the veto silent:
- The veto holder never met you. They evaluate from a forwarded link, alone, in four minutes, with no context and no relationship to protect.
- Saying nothing is cheaper than saying no. Raising an objection means owning it. Letting the thread go quiet costs nothing and carries no risk.
- Your page was written for the champion. It answers enthusiasm. It does not answer “what happens when this fails in production” or “where does the data live.”
How to spot one after the fact
A silent veto leaves a distinctive residue. Look for:
- A strong first call, then a stall with no stated reason.
- “We’re going to revisit this next quarter” arriving without a single objection on record.
- Healthy engagement metrics next to a conversion rate that will not move.
- The champion going quiet after saying they would “socialise it internally.”
The single most useful question to ask a champion is not what do you think. It is “who else has seen this, and what did they say?” The answer is usually a name and a shrug. That shrug is the veto.
Why analytics cannot see it
This is the expensive part. A silent veto does not appear in any funnel report, because from the analytics point of view nothing went wrong. Bounce rate is normal. Time on page is normal. The visitor who clicked was satisfied. The person who killed the deal either never visited a tracked surface or arrived through a forwarded link that carries no campaign parameters. The dark funnel.
So the loss is real and the record is empty. That is why teams keep re-running campaigns with the same structural gap in them, quarter after quarter.
How to test for one before launch
You cannot A/B test a silent veto. An A/B test measures the visitor who clicked, and the veto holder is not that person. You need to put the asset in front of every seat on the committee before spending the budget.
That is what a buying committee simulation does. WhyUser runs one agent per committee role. champion, economic buyer, technical decision-maker, end user, security guardian. across several behavioural states, and reports which role abandons, at which section, and which specific piece of evidence was missing when they did. Every finding is sourced to something a real buyer in that role actually said in a review, a call, or a community thread.
Economic Buyer. ABANDON at the pricing section. No stated cost of the current alternative, so no baseline to compare against. Blocking evidence missing: a before-and-after number tied to a named workload.
Security Guardian. ABANDON above the fold. No SOC 2 signal, no data-residency statement anywhere on the page. Left before reading the value proposition.
Questions people ask about silent veto
What is a silent veto in B2B sales?
A silent veto is when one member of the buying committee decides against a purchase and never voices the objection. The deal stalls without an explanation, so the vendor never learns which missing proof point caused it.
Who usually casts the silent veto?
Most often the economic buyer or the security reviewer. Both evaluate late, evaluate alone, and have no relationship with the vendor to protect, so declining to engage costs them nothing.
Why does a silent veto not show up in analytics?
Because the person who casts it usually never touches a tracked surface. They read a forwarded link with no campaign parameters, or a summary someone pasted into Slack. From the analytics view, nothing went wrong.
Can you test for a silent veto before launch?
Yes, by simulating each committee role separately against the asset before it ships. A buying committee simulation reports which role abandons and what evidence was missing, at a point where changing the page is still free.
Synthetic persona
Also called: simulated buyer, AI persona, agent persona.
A synthetic persona is a simulated buyer, generated by a model and used in place of a human respondent. The term says nothing about quality, because it says nothing about sourcing. Two products can both sell synthetic personas while one invents its buyers from a prompt and the other builds them from a corpus of real buyer language. Sourcing is the whole distinction, and it is the question most vendors are vaguest about.
Three grades of sourcing
Every synthetic persona is built from something. That something decides whether it is useful, and the three options are not close in quality.
| Prompt persona | CRM / call-data persona | Evidence-compiled persona | |
|---|---|---|---|
| Built from | The sentence you just typed | Your own records, deals and call recordings | Those, plus sources outside your walls |
| Who it represents | Nobody in particular | People who already replied to you | Also people who never replied |
| Knows why you lost | No | Only where a loss was logged, and logged honestly | Yes. Competitors’ reviews say it in public |
| Sees people who never contacted you | No | No. They are not in your CRM | Yes. Community threads |
| Cites a source per finding | Nothing to cite | Sometimes | Yes, on every finding |
| Main failure mode | Invention | Survivorship | Corpus bias, reduced but still present |
Note the last row. The third column is the widest sample available, not a clean one. Any vendor claiming otherwise is overselling.
Why CRM-built personas feel rigorous and mislead
This is the trap, and it catches careful teams rather than lazy ones.
“We built our personas from our own call transcripts” sounds far more rigorous than a prompt, and it is. Real people said those words. The problem is which real people.
Everyone in your CRM replied to you. Everyone on a recorded call agreed to take it. Every closed-won record is a person you persuaded. Your first-party data is a census of people you already reached, which makes it the wrong sample for the job marketing actually has: reaching the people who did not respond.
So a CRM-built persona models your best-case buyer with real fidelity, then tells you your page works. It usually does. For them.
Where the non-responders leave a trace
They are not silent. They are just not talking to you.
- Competitors’ reviews. Written by people who evaluated your category and chose somebody else, explaining exactly why, in public. The closest thing to a transcript of the deal you lost.
- Community threads. People with the problem, comparing options, who have contacted no vendor at all and will not until they are ready.
- How AI assistants describe your category. Increasingly the first version of your pitch a buyer sees, and not written by you. See dark funnel.
WhyUser compiles from these alongside your own transcripts. The transcripts still matter. They are just not sufficient alone, and a vendor who stops there has built a very good model of your existing customers.
Four questions that separate the grades
Ask these of any vendor selling synthetic personas, including us.
- Name your sources. “Trained on B2B data” is not an answer. Which corpora, refreshed how often, and can I see the source on an individual finding?
- What share comes from outside my own systems? If the answer is none, you have a well-modelled version of the people you already reached.
- Run the same page twice. If the verdict moves, you cannot test an edit, because you can never separate your change from the model.
- What does it refuse to claim? A vendor who cannot name a limitation has not looked for one.
What no synthetic persona can do
- Replace talking to customers. It is built from customer evidence, so it sits downstream of it. Thin evidence in, confident nothing out.
- Give you an absolute conversion rate. Rank and direction hold up. A precise number does not, and one wrong magnitude sinks every correct finding beside it.
- Escape its corpus. Public reviews skew to the delighted and the furious. The quiet middle is under-represented, and a good system says so.
- Surprise you from outside its evidence. A real person occasionally says the thing nobody modelled. Worth paying a panel for.
Questions people ask about synthetic personas
What is a synthetic persona?
A simulated buyer generated by a model and used in place of a human respondent. The term says nothing about quality on its own, because it says nothing about what the persona was built from.
What is the difference between synthetic persona products?
Sourcing. Some invent the buyer from a prompt at the moment of asking. Some build it from your CRM and call recordings. Some compile it from those plus sources outside your walls, such as competitors reviews and community threads. The three are not close in quality.
Are personas built from my CRM and call data good enough?
They are rigorous about the wrong population. Everyone in your CRM already replied to you and everyone on a recorded call agreed to take it, so first-party data is a census of people you already reached. Marketing needs the people who did not respond, and they are not in there.
Where can I find the buyers who never responded?
In competitors reviews, written by people who evaluated your category and chose somebody else, and in community threads where people compare options without contacting any vendor.
Are synthetic personas accurate?
They are useful for rank and direction rather than absolute numbers, and only as good as what they were built from. The way to judge a vendor is to ask whether they publish a graded track record of predictions sealed before the outcome existed.