Stage 02 · Conversion: The Page Is a Decision, Not a Design

Ariadne's second stage is described as turning a visit into a decision: "right service, clear next step." I want to defend the word decision, because it changes what you build.

What the visitor's brain is doing

The response-time constants that govern interface design are older than the web and have not moved. Miller established them in 1968; Card, Robertson and Mackinlay reinforced them; Nielsen's restatement is the version most people know Nielsen 1993:

• 0.1 second is about the limit for the user to feel the system reacted instantaneously.
• 1.0 second is about the limit for their flow of thought to stay uninterrupted, though they will notice the delay.
• 10 seconds is about the limit for keeping their attention on the task at all.

Google's Core Web Vitals thresholds were chosen with explicit reference to this literature, and the thresholds themselves are published: at the 75th percentile of page loads, Largest Contentful Paint is "good" at ≤ 2500 ms and poor above 4000 ms; Interaction to Next Paint is good at ≤ 200 ms and poor above 500 ms; Cumulative Layout Shift is good at ≤ 0.1 and poor above 0.25 McQuade & Pollard 2020. Interaction to Next Paint replaced First Input Delay as a stable Core Web Vital on 12 March 2024 Viscomi 2024.

Note the 75th percentile choice. It is deliberate: it guarantees that three of four visits met the target, rather than letting a good median hide a bad tail.

Four gates in series, so page yield is their product:

Four gates passed at 80% each gives $0.8^4 \approx 0.41$ — a page where nothing is wrong and well over half the qualified traffic still leaves. A visitor who fails any one gate is gone, and the page owner usually cannot tell which one. This is the core reason conversion work without measurement degenerates into taste.

The statistic that should change how you spend

Here is the finding I put in front of every founder who wants a redesign.

Kohavi, Tang, Xu, Hemkens and Ioannidis reported, in a peer-reviewed open-access paper, that across Google, LinkedIn and Microsoft about two-thirds of experiments show that promising ideas that were implemented in products failed to improve the metrics they targeted — and in highly optimised domains such as search engines, "failures were in the range of 80–90%" Kohavi et al. 2020.

These are organisations with world-class product teams, enormous data, and every incentive to be right. If two-thirds of their good ideas fail, the prior on your redesign is not favourable.

What to change when you cannot test

Most owner-led service businesses do not have the traffic to run a well-powered A/B test on a landing page. I will treat that problem properly in a later post. For now, the practical position:

When you cannot measure a change, prefer changes whose value does not depend on measurement — the ones that are true regardless of the experiment you cannot run.

Say who it is for, in their words. Positioning is the first gate and the cheapest to fix. It costs nothing and fails silently.

Name the specific service. "Digital transformation" fails gate two for everyone. "Fixed-fee CRM build for firms of 3–25 staff" fails it for almost nobody who matters.

Show evidence, quantified. This is the same lever that raised citation rates in the generative-engine study: specific numbers with attribution. Humans and retrievers both reward it.

State the next step concretely. Not "get in touch" — what happens, who replies, how long it takes, and what it costs. Gate four is where vagueness is fatal, because the visitor is being asked to accept an unbounded commitment.

Make it fast. Not because speed converts by itself, but because the response-time limits are hard constraints on whether the other four things get read at all.

The conversion event is not the click

One definitional point that quietly determines everything downstream. If you define conversion as form submissions, you will optimise for form submissions, and you will get them: more, worse, cheaper. The measurement stage exists to prevent exactly this, and Goodhart's law will get its own post because it deserves one.

For a service business the conversion worth optimising is a qualified conversation — an inquiry that matched your written criteria and reached a human who could help. Anything else is a number that goes up while the business does not.

References

1. Nielsen, J. (1993). Response times: The 3 important limits. Nielsen Norman Group (originally Chapter 5 of Usability Engineering, Morgan Kaufmann). https://www.nngroup.com/articles/response-times-3-important-limits/Practitioner source; derives from Miller (1968) and Card, Robertson & Mackinlay (1991).
2. McQuade, B., & Pollard, B. (2020). How the Core Web Vitals metrics thresholds were defined. web.dev / Chrome for Developers (published 21 May 2020, updated 2025). https://web.dev/articles/defining-core-web-vitals-thresholds
3. Viscomi, R. (2024). Interaction to Next Paint is officially a Core Web Vital. web.dev blog, Google Chrome team, 12 March 2024. https://web.dev/blog/inp-cwv-launch
4. Kohavi, R., Tang, D., Xu, Y., Hemkens, L. G., & Ioannidis, J. P. A. (2020). Online randomized controlled experiments at scale: Lessons and extensions to medicine. Trials, 21, 150. https://doi.org/10.1186/s13063-020-4084-y