# Message test — https://www.resemble.ai/ After reading your page, 11 of 15 personas could name a reason to pick you over a similar option. - **Page tested:** https://www.resemble.ai/ - **Audience tested against:** Security and fraud leads at telecom carriers, banks and insurers, government agencies, and large platforms who are seeing deepfake fraud in calls, meetings, or media and need to catch it as it happens. They champion the deal but rarely own the budget, so they're building a case with a committee of 4 to 5 people, often including compliance, HR, or a channel partner, and they usually need on-prem or data-residency options to get through procurement. They can put a number on the fraud and have a buying timeline, typically for deals of $50K and up. - **Personas:** 15 simulated - **Report:** https://grader.wynter.com/r/multimodal-deepfake-detection-and-watermarking-SOlkAaY > These answers are generated by AI, scored on Wynter's B2B Message > Layers framework using behaviorally-diverse simulated personas. The > methodology is real and the critique is directional. What a simulated > persona cannot have is a live budget, a renewal coming up, or a boss > asking about this quarter. --- ## 01 · The scores Every persona answered all four questions. These are four independent proportions of the same panel, not stages of a funnel. | Layer | Question | Cleared the bar | Strength | Of those who passed | | --- | --- | --- | --- | --- | | 1. Clarity | Do they understand what you do? | 15/15 | 87% | 6 without hesitation, 9 with reservations | | 2. Relevance | Can they tell what it solves, and who it's for? | 15/15 | 84% | 4 without hesitation, 11 with reservations | | 3. Value | Do they actually want it? | 12/15 | 67% | all with reservations | | 4. Differentiation | Is there a reason to pick you over the alternatives? | 11/15 | 61% | all with reservations | **Brand alignment** (a side metric, not one of the four layers) — 12/15, 67% strength (all with reservations). Does the page read like the company you actually are? **Fix first: Differentiation.** Earliest failing layer, walking the sequence in order — not simply the lowest score. --- ## 02 · What to change, layer by layer Ordered worst-first. Specific edits, not a restatement of the score. ### Differentiation **Add a short section giving two specific reasons to choose this over existing fraud tools.** Nothing on the page says why this beats the voice fraud tooling a buyer already runs. Name the concrete edge, detection latency, languages covered, integration time, rather than general strength claims. *effort medium · impact high · tested against Give a reason to choose you* **Collapse the model variant names into one line stating what a customer buys.** Overlapping model and product names make the purchase unit unclear. Lead with the single product name and what it does, then list variants once beneath it. *effort medium · impact medium · tested against Plain language* **Define "real-time" in the hero with a latency figure in milliseconds.** Real-time reads as anything from sub-second to a few minutes, so the claim cannot be compared with any rival. Give the measured detection latency and the audio length it needs. *effort low · impact medium · tested against Specifics beat superlatives* ### Value **Replace the logo wall with one customer case study naming the fraud losses avoided.** Logos show who bought, not what happened after. Swap in a short reference from a named security or fraud team with a before and after number and a timeframe. *effort high · impact high · tested against Proof next to the claim* **Add the benchmark dataset and evaluation method directly beneath each accuracy percentage.** The accuracy numbers float with no dataset, test set, or method behind them, so buyers treat them as marketing. Put the named dataset, sample size, and who ran the test on the same line as the number. *effort medium · impact high · tested against Proof next to the claim* **Add a deployment line stating cloud, private cloud, or on-premises availability.** Security buyers cannot tell whether audio leaves their environment, and silence on deployment ends the evaluation. State the deployment options and data residency in one line near the pricing or product section. *effort low · impact high · tested against Answer the live objection* ### Brand alignment (side metric) **Add a line stating detection accuracy on noisy contact center audio.** Contact center buyers assume lab accuracy collapses on compressed phone calls. State how the model performs on that audio, with the number. *effort medium · impact medium · tested against Answer the live objection* **Add a named-audience line under the H1 for fraud, security, and trust-and-safety teams.** The buyer has to work themselves out from the use-case tiles. Say who the product is for in one sentence above those tiles. *effort low · impact medium · tested against Name the audience* --- ## 03 · What is working ### The use-case grid is the one element that makes the buyer and problem obvious Five respondents said the use-case grid or list immediately signalled the intended security, fraud, and trust-and-safety buyer, even though no job title or audience is named explicitly. > I saw "KYC + account onboarding" and "Contact center fraud" and immediately recognized my own workflows. > > — Head of Security, Insurance, 5000+ > the row of use cases right below — "Contact center fraud," "KYC + account onboarding," "Law enforcement forensics," "Executive + brand protection" — makes it obvious who this is for > > — Head of Security, Telecommunications, 5000+ > the use-case grid right below it — "Contact center fraud," "KYC + account onboarding," "Law enforcement forensics," "CSAM + sexualized deepfakes" — spells out the buyer segments without me having to infer much. > > — Security Lead, Government, 1001-5000 --- ## 04 · What the personas said ### Key performance claims lack defined parameters and are hard to parse Four respondents said claims lack tolerance, audience, or definition — 'real-time' is undefined as sub-second versus human speed — and that page clutter, repetition, and product naming across model variants forced re-reading. > The phrase "explainable results" and "deterministic score and verdict" get thrown around without definition — deterministic to what tolerance, and explainable to whom, an analyst or an auditor? > > — Chief Information Security Officer, Government, 1001-5000 > Same with "real time": is that sub-second, or just "faster than a human reviewer," because those imply very different infrastructure > > — Chief Information Security Officer, Government, 1001-5000 > a wall of cookie/consent text (PostHog, HubSpot, Clarity, LinkedIn Insight Tag) sits before you even get to the product, and then the page repeats itself (two 'contact center' blocks, duplicated testimonials verbatim) > > — Fraud Operations Manager, Banking, 1001-5000 > the real friction was the model-name soup like "DETECT-World," "DETECT-3B Omni," "PerTh Multimodal," and "Signal" thrown at me without a plain sentence saying which one I'd actually buy or start with > > — Fraud Prevention Manager, Government, 5000+ ### The buyer has to be reverse-engineered because no audience is named Two respondents said the target buyer must be inferred from use-case tiles rather than stated, with no explicit job-title targeting anywhere on the page. > they never just say "built for fraud/compliance teams at banks or call centers" outright — I had to piece it together from the eight use-case cards > > — Fraud Operations Manager, Banking, 1001-5000 > it's clearly cast as a horizontal tool for several buyer types (trust & safety, KYC, forensics) rather than speaking to me specifically — I had to pick my lane out of eight listed > > — Fraud Prevention Manager, Financial Services, 5000+ ### Accuracy claims are not believed because no source, dataset, or methodology is named Six respondents flagged the accuracy numbers as unverified, asking for named datasets, leaderboard sources, independent benchmark methodology, or a live pilot before trusting them. One also questioned performance on noisy contact center audio. > the page gestures at "third-party validated benchmarks" and "public leaderboards" without naming which leaderboard, what dataset, or what adversarial conditions were tested > > — Chief Information Security Officer, Government, 1001-5000 > I'd want them on a call with our fraud analytics lead to walk through false-positive rates on real call center audio (compressed, noisy, accented — not clean studio samples) > > — Fraud Operations Manager, Banking, 1001-5000 > 99.5% audio accuracy is just a number on a page, I'd need to see it validated against my own call traffic, not a benchmark I can't audit > > — Security Lead, Telecommunications, 1001-5000 ### Logos are not enough; respondents want a documented outcome case study Three respondents asked for a specific reference — documented fraud loss reduction, a telecom SOC case study, or a named insurance client — instead of logo walls. > none of the named logos (Telnyx, Okta, Deutsche Telekom) are insurers > > — Fraud Operations Manager, Insurance, 1001-5000 > A named telecom SOC or carrier fraud team case study above the fold — not a logo, an actual sentence like 'X carrier cut call fraud losses by Y% using real-time detection integrated with their contact center stack' > > — Head of Security, Telecommunications, 5000+ ### Missing on-premises deployment is an outright evaluation blocker Two respondents flagged that on-prem deployment is never mentioned, one calling it a blocker to evaluation and one citing it as a competitor advantage. > "on-premises" isn't mentioned anywhere here, and that's a hard requirement for us, so before I take a meeting I need someone to confirm deployment model > > — Senior Security Director, Financial Services, 5000+ > Resemble wins if they can confirm on-prem support in writing and show one more customer of Telnyx's caliber; right now Pindrop has the track record and the deployment answer already sorted > > — Senior Security Director, Financial Services, 5000+ ### No competitive comparison is offered, and a named competitor is assumed to be ahead Two respondents noted the absence of any comparison against current fraud tools, with one stating Pindrop holds an advantage on on-prem support and track record. > Rules out: no head-to-head vs. what I already run. > > — Fraud Prevention Manager, Telecommunications, 5000+ > Resemble wins if they can confirm on-prem support in writing and show one more customer of Telnyx's caliber; right now Pindrop has the track record and the deployment answer already sorted > > — Senior Security Director, Financial Services, 5000+ ### Model and product boundaries blur even where the category is clear Two respondents said the product category is legible but the edges between underlying model variants are not, leaving it unclear what is actually being purchased. > the named model architecture — DETECT-World, Signal, PerTh, Identity — because it signals they've built distinct tools for distinct jobs (deterministic scoring vs. watermarking vs. identity matching) rather than one black-box classifier rebranded four ways > > — Chief Information Security Officer, Government, 1001-5000 > they list five different model products (DETECT-World, Signal, PerTh, Identity, plus watermarking) and it's not obvious which of those is the actual telecom fraud product versus which is bolted on for content moderation > > — Head of Security, Telecommunications, 5000+ --- ## 05 · The hardest read An adversarial pass over the findings. Every claim below was checked against the panel's own answers; unsupported ones were dropped. - **The page's core performance numbers are dead weight because nothing substantiates them** *(high)* Six respondents rejected the accuracy claims as unverified, demanding named datasets, leaderboard sources, or independent methodology, and four more said claims lack tolerance, audience, or definition — 'real-time' undefined. The headline proof point is the… - **Logo walls actively backfire as proof** *(high)* Three respondents dismissed logos and asked instead for documented fraud loss reduction, a telecom SOC case study, or a named insurance client. Named brands without outcomes read as decoration next to unverified accuracy numbers flagged by six. - **A buyer cannot determine what they would actually be purchasing** *(high)* Two respondents said model and product boundaries blur even where the category is legible, and three cited product naming across model variants as forcing re-reading. Unclear SKUs stall procurement regardless of interest. - **The page loses deals on silence rather than on argument** *(high)* On-prem deployment is never mentioned and was called an outright blocker to evaluation, while two respondents noted no comparison against current fraud tools and named Pindrop as ahead on on-prem and track record. Omissions are being filled in by competitors. - **The use-case grid is carrying the entire page alone** *(medium)* Five respondents credited the grid as the one element making buyer and problem obvious — and it is the only positive theme against six negatives. Relevance is established by a single tile block, then unsupported everywhere else. - **Naming the audience explicitly is free upside the page refuses to take** *(medium)* Two respondents had to reverse-engineer the buyer from tiles with no job-title targeting anywhere, and five who correctly identified the security and fraud buyer did so despite no audience being stated. Inference works until it doesn't. --- ## 06 · Who answered | # | Role | Industry | Company size | | --- | --- | --- | --- | | 1 | Fraud Prevention Manager | Telecommunications | 5000+ | | 2 | Chief Information Security Officer | Government | 1001-5000 | | 3 | Senior Security Director | Financial Services | 5000+ | | 4 | Fraud Operations Manager | Banking | 1001-5000 | | 5 | Head of Security | Insurance | 5000+ | | 6 | Security Lead | Telecommunications | 1001-5000 | | 7 | Fraud Prevention Manager | Government | 5000+ | | 8 | Chief Information Security Officer | Financial Services | 1001-5000 | | 9 | Senior Security Director | Banking | 5000+ | | 10 | Fraud Operations Manager | Insurance | 1001-5000 | | 11 | Head of Security | Telecommunications | 5000+ | | 12 | Security Lead | Government | 1001-5000 | | 13 | Fraud Prevention Manager | Financial Services | 5000+ | | 14 | Chief Information Security Officer | Banking | 1001-5000 | | 15 | Senior Security Director | Insurance | 5000+ | --- ## 07 · Before you act on this The methodology is real, and the critique is directional. What a simulated persona cannot have is a live budget, a renewal coming up, or a boss asking about this quarter. **Validate anything you're betting on with real ICPs who are actually in-market.** Being wrong is more expensive than you think. Finding out is cheaper than you'd guess. Wynter runs message testing with verified B2B professionals — trusted by HubSpot, RingCentral, Shopify, Cognism, Paddle, Veeam, Rippling and Miro. This report is kept for 60 days from 2026-09-29, then deleted along with the personas and their answers.