# Message test — https://www.cdata.com/ After reading your page, only 7 of 15 personas could name a reason to pick you over a similar option. - **Page tested:** https://www.cdata.com/ - **Audience tested against:** Data engineering and analytics leaders - **Personas:** 15 simulated - **Report:** https://grader.wynter.com/r/cdata-the-ai-gateway-PkNInWk > 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 | 78% | all with reservations | | 2. Relevance | Can they tell what it solves, and who it's for? | 10/13 | 65% | all 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? | 7/15 | 48% | all with reservations | **Brand alignment** (a side metric, not one of the four layers) — 13/15, 70% 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 one named customer with role, company and a result below the hero.** Nothing on the page shows a production deployment by a real buyer. Add a single named reference with the outcome they got, placed beside the "governed from the prompt down to the record" claim. *effort medium · impact high · tested against Proof next to the claim* **Add a line under "Enterprise-grade MCP to any AI platform" stating what it replaces.** Readers with an existing stack cannot tell what this removes: custom MCP servers, per-tool connectors, a homegrown gateway. Name the thing it retires and why that is cheaper than keeping it. *effort medium · impact high · tested against Give a reason to choose you* **Rename the "Trusted by global teams" section to say these are supported platforms, not customers.** The logo row sits under a trust claim but lists vendors like Anthropic, OpenAI and Microsoft, so a reader cannot tell whether anyone actually bought this. Label the row for what it is: systems and models the gateway connects to. *effort low · impact high · tested against Proof next to the claim* ### Relevance **Rewrite the hero headline into one sentence naming the job, not four noun phrases.** "Connected data. Right model. Exact context. Controlled cost" has no verb and reads as internal shorthand. Replace it with a line that says what the product does: connect any AI assistant to live enterprise systems with governance. *effort medium · impact high · tested against Lead with the use case* **Name the buyer role in the subhead under the hero headline.** Nothing on the page says who it is for; readers reverse-engineer it from role tabs and logos. Say plainly that it is for data platform and AI engineering teams connecting enterprise systems to LLMs. *effort low · impact high · tested against Name the audience* **Move one worked example prompt and answer directly beneath the hero.** The product only makes sense once readers reach the Claude and ChatGPT demos far down the page. Pull the deal-slippage example up so the first screen shows the product working. *effort medium · impact medium · tested against Show the product early* ### Value **Attach a source and date to each cost, accuracy, or token-savings figure.** Performance and cost numbers appear with nothing behind them, so readers discount them and ask to benchmark internally. State the workload, the comparison baseline, and when it was measured next to each number. *effort medium · impact high · tested against Proof next to the claim* **Replace "Live in under 2 minutes" with what the two minutes covers.** A reader with connectors already in place does not believe a two-minute setup and cannot tell what it includes. Say which steps are counted: install, authenticate one source, run first query. *effort low · impact medium · tested against Specifics beat superlatives* ### Clarity **Define MCP in plain words the first time it appears on the page.** "Enterprise-grade MCP to any AI platform" assumes the reader already knows the protocol. Add a short clause saying it is the standard that lets AI assistants call your systems and data. *effort low · impact medium · tested against Plain language* ### Brand alignment (side metric) **Add an audit and compliance line under the governance claim in the hero section.** "Governance guardrails" is asserted but never defined, so regulated buyers cannot tell if queries are logged, access is row-level, or certifications exist. Name the controls: audit trail, role-based access, SOC 2 or equivalent. *effort low · impact medium · tested against Answer the live objection* --- ## 03 · What is working ### The worked examples and logos do the explaining the copy does not Four respondents said the example prompts, cross-system scenarios, and customer logos clarified the use case and the governance concept faster than the headline or abstract language. > The logos (Salesforce, Palantir, Microsoft, SAP) and the specific worked examples made it click faster than the abstract "context graph" language would have on its own. > > — Data Analytics Director, Financial Services, 5000+ > the Gong/Stripe/Greenhouse examples like "9 closed-won deals from Q2 have no matching invoice—$412K total" are exactly the kind of cross-system reconciliation my team currently does manually > > — Data Analytics Director, Financial Services, 5000+ --- ## 04 · What the personas said ### The headline forces readers into the examples before the product makes sense Four respondents called the hero line vague, jargon-laden, or verb-free, saying 'context graph' and stacked abstract nouns read as internal language. Each only understood the product after reaching the demos. > "governed" and "the record" are doing a lot of unexplained work, so I had to skip past it to the actual example prompts to figure out what the product does > > — Analytics Engineering Manager, Technology, 1001-5000 > The phrase "context graph" and the headline trio "Right model. Exact context. Controlled cost" are the culprits — those are internal product vocabulary dressed up as plain English > > — Data Analytics Director, Financial Services, 5000+ ### The page never names who it is for Five respondents said the target reader and buyer role are never stated and had to be inferred from role tabs, logos, and examples. One added that the page also never says what the product replaces. > The intended reader isn't stated outright, I inferred it from the role tags (Sales, Finance, People & Ops, Engineering & Data) and tool logos > > — Analytics Engineering Manager, Technology, 1001-5000 > The intended reader is never explicitly named — there's no "this is for data leaders" or "for IT/data platform teams" sentence — I inferred it from the persona tabs > > — Data Analytics Director, Financial Services, 5000+ > The reader is inferred rather than stated outright, though — it's obviously someone running data/AI infra at a company with Salesforce/Snowflake/Workday already in place, not spelled out as "for Data Engineering Leads" or similar, I pieced that together from the logos and query examples > > — Data Engineering Lead, Enterprise Software, 501-1000 > the intended reader is never explicitly named; it's not "for Data Analytics Directors" or "for IT/platform teams" > > — Data Analytics Director, Healthcare, 501-1000 ### Accuracy, cost, and token-savings claims are not believed without outside verification Four respondents flagged the performance, accuracy, and cost metrics as unverified by any third party, saying they would need internal benchmarking or independent proof before accepting them. > The 97.6% token reduction and 178x cost claims are the kind of thing that would actually move a budget line, but they're self-reported with no methodology beyond "internal testing, sandbox accounts" — I'd want the actual query set and model list before I believed it over my own benchmarking. > > — Senior Data Engineer, Enterprise Software, 201-500 > those are the headline differentiators versus adjacent tools and I'm not taking a vendor's own sandbox benchmark against competitors they're naming in the same breath. > > — Director of Data Engineering, Consulting, 501-1000 > The 98.5% vs 65-75% accuracy number and the 97.6% token reduction are the kind of numbers that would actually move me, but they're self-reported CData Labs studies with no third-party verification, so I'd want that replicated on our own data before I believed the magnitude. > > — VP of Data & Analytics, Healthcare, 201-500 ### Social proof is present but unusable as evidence Three respondents said the 'Trusted by' section lists vendors rather than customers, carries no quotes or named references, and shows no production deployment. > "Trusted by GSK, Salesforce, Palantir..." with zero quotes attached feels like decoration, not proof > > — Analytics Engineering Manager, Technology, 1001-5000 > the "Trusted by" row is Salesforce, Palantir, Anthropic, SAP — those are co-vendors and partners, not a financial services customer saying "we ran this in production and it held." > > — Data Analytics Director, Financial Services, 5000+ > What's missing is any named customer actually running this in production — logos like Salesforce and Anthropic are listed as "trusted by," but I can't tell if they're customers or just companies whose products CData connects to > > — Data Engineering Lead, Consulting, 5000+ ### The technical tone fits data engineers but omits regulated-industry concerns Two respondents noted the copy assumes familiarity with MCP and data governance; one said the absence of audit, compliance, and regulated-industry language is disqualifying for financial services. > The tone is pitched at a technical buyer — IT/data platform/architecture people evaluating infrastructure — with enough jargon ("schema-aware toolkits," "context graph," "governed catalog") that it assumes familiarity > > — Analytics Engineering Manager, Technology, 1001-5000 > it never once says "built for regulated industries" or "financial services" or addresses audit/compliance directly, which is the first thing I'd want to see named given our sector > > — Data Analytics Director, Financial Services, 5000+ ### Readers could restate the product as a governed MCP gateway with schema-aware context Two respondents played back a coherent description: an AI gateway routing prompts to models with built-in governance and schema-aware enterprise connectivity. > It's a governed AI gateway/MCP layer that sits between chat tools like Claude, ChatGPT, Copilot and Gemini and your actual business systems (Salesforce, Snowflake, SQL Server, Workday etc.) — it routes the prompt to a model, pulls schema-aware context so the model knows what "pipeline" or "ARR" means in your data > > — Data Engineering Lead, Enterprise Software, 501-1000 > It's an MCP gateway sitting between AI tools like Claude, ChatGPT, Copilot and Gemini and your actual business systems—Salesforce, Snowflake, NetSuite, SQL Server, Workday, Zendesk > > — Director of Data Engineering, Technology, 5000+ ### For some the value is a nice-to-have their current stack already covers Three respondents said their existing stack handles most of this, called the benefit non-urgent absent validation, and wanted a working session on real connectors before judging. > nothing here screams urgency — it's a good-to-have, not a burning problem. I'd take a short intro meeting out of curiosity given the stats (98.5% accuracy, 178x cost difference), but I'd want to see those numbers validated outside a CData-run sandbox > > — Analytics Engineering Manager, Technology, 1001-5000 --- ## 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 offloads its core explanatory job onto screenshots and logos, so the copy itself is dead weight.** *(high)* Four respondents only understood the product after reaching the demos, calling the hero line vague or verb-free; four more said examples and logos clarified the use case faster than the headline. The writing explains nothing the artwork doesn't. - **Readers cannot tell whether they are the buyer, which stalls qualification before any value argument lands.** *(high)* Five respondents said the target reader and buyer role are never stated and had to be inferred from tabs and logos; one noted the page never says what the product replaces. Audience and displacement are both left to guesswork. - **Every quantified claim on the page is treated as marketing noise because nothing external backs it.** *(high)* Four respondents flagged accuracy, performance, and cost metrics as unverified and demanded internal benchmarking or third-party proof; three said the 'Trusted by' section lists vendors, not customers, with no quotes or production deployments. - **Unverified numbers plus a stack that already covers the job reduces the page to a non-purchase.** *(high)* Three respondents said their existing stack handles most of this and called the benefit non-urgent absent validation, wanting a working session on real connectors; four separately rejected the metrics as unproven. No urgency survives. - **The page disqualifies itself in regulated sectors by selling governance without the compliance vocabulary governance buyers require.** *(medium)* Two respondents noted the copy assumes MCP and data-governance fluency, and one called the absence of audit, compliance, and regulated-industry language disqualifying for financial services. The governance pitch misses the audience most likely to pay for it. - **Correct product comprehension is the exception, not the baseline.** *(medium)* Only two respondents played back a coherent description of a governed MCP gateway, while three needed the demos to grasp it and five could not identify the audience. Accurate recall is a minority outcome. --- ## 06 · Who answered | # | Role | Industry | Company size | | --- | --- | --- | --- | | 1 | Senior Data Engineer | Enterprise Software | 201-500 | | 2 | Director of Data Engineering | Consulting | 501-1000 | | 3 | Analytics Engineering Manager | Technology | 1001-5000 | | 4 | Data Analytics Director | Financial Services | 5000+ | | 5 | VP of Data & Analytics | Healthcare | 201-500 | | 6 | Data Engineering Lead | Enterprise Software | 501-1000 | | 7 | Senior Data Engineer | Consulting | 1001-5000 | | 8 | Director of Data Engineering | Technology | 5000+ | | 9 | Analytics Engineering Manager | Financial Services | 201-500 | | 10 | Data Analytics Director | Healthcare | 501-1000 | | 11 | VP of Data & Analytics | Enterprise Software | 1001-5000 | | 12 | Data Engineering Lead | Consulting | 5000+ | | 13 | Senior Data Engineer | Technology | 201-500 | | 14 | Director of Data Engineering | Financial Services | 501-1000 | | 15 | Analytics Engineering Manager | Healthcare | 1001-5000 | --- ## 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 2026-10-05, then deleted along with the personas and their answers.