Of 24 well-known B2B SaaS homepages we opened on 18 October 2026, 23 put a demo, sales or trial form one click from the first screen. Only 5 told a first-time visitor, in the hero alone, what the product is, who it is for and one concrete thing it does. Twenty of the 24 heroes mention AI or agents.
Below are the nine tests, every measurement behind the scores and what the higher scorers do that others could copy. The rubric draws on Nielsen Norman Group, web.dev and the HTTP Archive's Web Almanac.
The rubric: nine pass or fail tests
Each test passed scores one point, so each dimension scores 0 to 3 and the total is out of 9. Binary tests let anyone repeat them.
| Dimension | Test, one point each | Basis |
|---|---|---|
| Clarity | C: names a category buyers already use; coined labels fail | NN/g homepage guideline |
| Clarity | A: names who it is for: a role, team or company type | NN/g homepage guideline |
| Clarity | S: gives one concrete specific: named functions, standards or a number | Our test |
| Speed | Time to first byte 0.8 s or less | web.dev "good" TTFB |
| Speed | HTML transfer 155 KB or less | Almanac 2025, 90th percentile |
| Speed | HTML transfer 78 KB or less | Almanac 2025, 75th percentile |
| Demo path | A demo, sales or trial form one click from the first screen | Our test |
| Demo path | Five fields or fewer on the first step | A round line, also our median |
| Demo path | Response time or call length stated | Our test |
Source: Nielsen Norman Group, 2002; web.dev, 2025; HTTP Archive Web Almanac, 2026
Clarity method. The hero means the H1 and the line under it, read in a desktop browser window of 1,024 by 768 pixels without scrolling. Navigation, logos and the title tag do not count.
Speed method. We planned to use Google's PageSpeed Insights API without a key. On 18 October it answered with a 429 "quota exceeded" error, so we used curl: one request per homepage at 14:40 UTC from a single desktop connection in Europe, recording time to first byte and the compressed size of the HTML document. It is a single run, and it measures server response and document weight, not images, JavaScript or Core Web Vitals.
Demo path method. We followed the most prominent demo, sales or trial link visible on the first screen and counted the visible text boxes, dropdowns and text areas on the first step. A group of radio buttons or tick boxes counts as one field; a lone consent box counts as none. We submitted nothing, so later steps of multi-step forms are not counted.
The scorecard: clarity and speed
Letters show which clarity tests a hero passed: C for category, A for audience, S for a concrete specific.
| Site | Category | Clarity | Passed | TTFB (s) | HTML (KB) | Speed |
|---|---|---|---|---|---|---|
| HubSpot | CRM and marketing | 3 | CAS | 0.16 | 104 | 2 |
| Pipedrive | CRM | 3 | CAS | 0.39 | 59 | 3 |
| Attio | CRM | 1 | S | 0.29 | 151 | 2 |
| Gong | Revenue intelligence | 1 | S | 0.42 | 201 | 1 |
| Intercom | Customer service | 2 | CS | 0.53 | 313 | 1 |
| Zendesk | Customer service | 2 | CS | 0.32 | 106 | 2 |
| Front | Customer service | 2 | AS | 0.40 | 1,084 | 1 |
| Amplitude | Product analytics | 3 | CAS | 1.29 | 72 | 2 |
| Mixpanel | Product analytics | 0 | none | 0.64 | 156 | 1 |
| PostHog | Product analytics | 2 | CS | 0.23 | 354 | 1 |
| Vercel | Developer platform | 0 | none | 0.18 | 86 | 2 |
| Datadog | Observability | 1 | C | 0.34 | 57 | 3 |
| Sentry | Error monitoring | 3 | CAS | 0.53 | 177 | 1 |
| Vanta | Compliance | 2 | CS | 0.25 | 73 | 3 |
| Drata | Compliance | 2 | CS | n/a | n/a | n/a |
| Rippling | HR and payroll | 2 | CS | 0.31 | 774 | 1 |
| Deel | HR and payroll | 1 | S | 0.25 | 103 | 2 |
| Ramp | Spend management | 3 | CAS | 0.53 | 138 | 2 |
| Linear | Product development | 1 | C | 0.42 | 186 | 1 |
| Asana | Work management | 0 | none | 0.28 | 173 | 1 |
| Notion | Workspace | 1 | S | 0.31 | 29 | 3 |
| Miro | Collaboration | 1 | C | 0.43 | 124 | 2 |
| Snowflake | Data platform | 1 | A | 0.41 | 126 | 2 |
| Fivetran | Data movement | 1 | S | 0.23 | 20 | 3 |
Source: Octo Agency, browser read and single curl run per homepage, 18 October 2026. Drata answered curl with a bot-challenge page (HTTP 403), so its speed is n/a. Rippling and Ramp redirected our browser to UK English pages, which were scored; Rippling's US HTML carries a different headline.
The scorecard: demo path and totals
| Site | Form | Clicks | Fields | Timing stated | Demo path | Total |
|---|---|---|---|---|---|---|
| HubSpot | Demo, email first | 1 | 1 | No | 2 | 7 |
| Pipedrive | Sales | 1 | 7 | No | 1 | 7 |
| Attio | Sales | 1 | 6 | No | 1 | 4 |
| Gong | Demo, email only | 1 | 1 | No | 2 | 4 |
| Intercom | Sales | 1 | 5 | No | 2 | 5 |
| Zendesk | Contact us | 1 | 5 | No | 2 | 6 |
| Front | Demo pop-up | 1 | 2 | No | 2 | 5 |
| Amplitude | Sales | 1 | 5 | No | 2 | 7 |
| Mixpanel | Demo | 1 | 5 | No | 2 | 3 |
| PostHog | Sales | 1 | 6 | No | 1 | 4 |
| Vercel | Demo | 1 | 5 | No | 2 | 4 |
| Datadog | Trial pop-up | 1 | 5 | n/a | 2 | 6 |
| Sentry | Demo | 1 | 7 | No | 1 | 5 |
| Vanta | Demo | 1 | 9 | No | 1 | 6 |
| Drata | Demo | 1 | 6 | No | 1 | 3 of 6 |
| Rippling | Demo | 1 | 7 | No | 1 | 4 |
| Deel | Demo pop-up | 1 | 3 | No | 2 | 5 |
| Ramp | Demo, email first | 1 | 1 | Yes, 30 min | 3 | 8 |
| Linear | Contact, Sales | 2 | 4 | No | 1 | 3 |
| Asana | Sales | 1 | 8 | Yes, 1 day | 2 | 3 |
| Notion | Demo, email first | 1 | 1 | No | 2 | 6 |
| Miro | Sales | 1 | 6 | No | 1 | 4 |
| Snowflake | Sales | 1 | 9 | Yes, 1 day | 2 | 5 |
| Fivetran | Sales | 1 | 8 | No | 1 | 5 |
Source: Octo Agency, browser read of each form page, 18 October 2026; no form was submitted. Fields counts the first step only; timing means a stated response time or call length
Clarity: the category word is what most heroes drop
Nielsen Norman Group's first homepage guideline, from 2002, asks for a tagline that says what the company does, and applies it to well-known companies too. It still separates this sample cleanly.
The five sites that passed all three clarity tests did it the same way: a plain category word, a named reader and one hard specific. Sentry pairs "application monitoring software" with developers and a light proof line about millions of them. Ramp's UK hero lists corporate cards, expenses, supplier payments and accounting automation, then names UK finance teams. HubSpot names marketing, sales and service in its subheading; Pipedrive says CRM in its headline and describes the pipeline work in the next line; Amplitude names product teams in the headline and analytics in the subheading.
The three zero scores are not unclear brands. Mixpanel's meta description lists analytics, session replay and experiments, but its hero carries a slogan about building with customers beside you. Vercel's hero reads "Agentic Infrastructure" with no subheading. Asana's calls itself an operating system for human-agent teams, while its title tag says work and project management. Each of these sites knows its category; the hero simply leaves it to the navigation and the search snippet.
AI is not the problem; replacing the category with it is. Twenty of 24 heroes mention AI or agents. Three of the five top clarity scores do too: HubSpot, Pipedrive and Amplitude mention AI in a line that also names what the product is. All three zero scores mention AI or agents, and none names a category.
Audience was the rarest pass. Only 7 of 24 heroes said who the product is for. "Your team" and "teams and agents" appeared often, and they describe every buyer. The B2B SaaS Website Playbook covers the full message hierarchy; here, 17 sites were one named role away from another point.
Rotating headlines. Gong, Notion and Fivetran rotate part of the H1; we scored what rendered.
Speed: the server is fast everywhere, the HTML is not
Time to first byte barely separated these sites. web.dev calls 0.8 seconds or less good and over 1.8 seconds poor, and notes TTFB is not a Core Web Vitals metric in itself. Twenty-two of the 23 measured homepages answered within 0.65 seconds from our connection; Amplitude, at 1.29 seconds, sat between good and poor on this single run. The median was 0.34 seconds.
Document weight is where the spread is. The Web Almanac's 2025 page weight chapter puts mobile home page HTML at 33 KB at the median, 78 KB at the 75th percentile and 155 KB at the 90th. Six of our 23 came in under 78 KB, with Fivetran at 20 KB and Notion at 29 KB. Nine sent more than 155 KB, and Front sent 1,084 KB.
We looked inside the heaviest documents. On Front, 84% of the uncompressed HTML is inline script; on Gong it is 76% and on Rippling 61%, which is the pattern of a JavaScript framework shipping its page data inside the document. PostHog's weight is mostly inline SVG and style, 45% and 42%; Sentry's is 44% inline SVG. A heavy document has to arrive and be parsed before the browser can do much else, so on a slow phone connection it delays everything after it.
Demo path: one click is universal, the form is where sites differ
Every site except Linear, whose Contact link opens a choice between sales and support, had a form one click from the first screen. Our B2B SaaS Demo Request and Lead Form Statistics 2026 post counts which call to action leads on 30 homepages; this sample adds what happens after the click.
HockeyStack's December 2025 study of 80 B2B SaaS companies, vendor data from an analytics company, puts demo pages at 0.91% of site traffic with a 70% bounce rate and a 5.5% conversion rate. Few people arrive, and most of those leave, so the first screen of the form carries a lot.
First-step length ranged from one field to nine. HubSpot, Gong, Ramp and Notion ask for an email address only; Front and Deel show two or three fields and a Continue button. Vanta and Snowflake ask nine fields on one screen. The Demo Request Conversion Rates 2026 post has the evidence on field count and completion.
Almost nobody says what happens next. Ramp's demo page promises an expert-run, 30-minute tour and says booking takes two minutes. Asana says its sales team replies within a business day; Snowflake says responses typically take one business day. Zendesk and Miro say "shortly". The other 18 forms say nothing about timing. It costs one sentence.
Some labels lead somewhere unexpected. Zendesk's "View demo" opens an interactive product tour with no form; its contact form sits behind "Contact us". PostHog labels its sales form "Talk to a human". Datadog's first screen offers only a free trial, a five-field sign-up that includes a password, so we scored that path.
How to use this
The tests take about 30 minutes per site with a browser and a terminal. Run them on your homepage and two competitors.
1. Read your hero cold. Cover the navigation and logo, then ask someone outside the company to name the category, the buyer and one thing the product does. A miss on any of the three is a one-line fix.
2. Weigh the HTML. Run curl against your homepage and note time to first byte and compressed size. Over 155 KB is worth a look inside: inline data, styles or SVG usually explain it.
3. Walk the demo path. Count clicks and first-step fields, then check whether the page says when you will hear back. Add the response time if you can keep the promise.
For the design decisions behind these tests, read B2B Website Design in 2026: The 9 Decisions That Decide Whether It Books Demos. When we build marketing sites in Webflow we would run these same nine tests on the staging site before launch, and our marketing website service explains how we scope that work.
