Speed is not a score. It is the part of the funnel nobody instruments

Performance conversations fail because they are held in the wrong currency. A developer brings a score out of a hundred; a business owner is deciding between this and a month of ads. The score does not translate, so the work does not get funded — and the site stays slow for reasons that have nothing to do with the engineering.

Measure the pages that carry money

• The product page, on a cold cache, from the country your buyers are in.
• The cart and the checkout — these usually bypass every cache you installed.
• Search and category results, which run the most expensive queries on the site.
• A logged-in session, if your customers have accounts.
• The same four again on a throttled mid-range Android, not a desktop browser.

The argument that works

Show the owner their own checkout loading on their own phone, on mobile data, beside a competitor's. That comparison needs no statistic and survives no argument. Then say what you will change, what you expect it to move, and how you will know — in seconds, on named pages, measured before and after.

The three effects nobody counts

1. Paid traffic. Every ad click that lands on a slow page is money already spent and not yet earned. Performance work makes the existing ad budget go further without raising it.
2. Retrieval. An assistant assembling an answer works to a deadline; a store that answers slowly is not slow in that context, it is absent from the answer entirely.
3. Support. A checkout that appears to hang produces duplicate orders, abandoned carts and messages that someone has to answer.

The second one is new and it is the reason this argument changed in the last two years. Performance used to be a conversion and a ranking question. It is now also an eligibility question: whether your page is in the set of pages an assistant could quote when a buyer asks where to buy.