Lab scores are your site on its best day, on a machine nobody owns, on a network nobody uses. Field data is what your actual visitors endured. Only one of these predicts revenue, and it is not the one in the deck.
Every web team has a performance number it likes. It usually comes from a lab tool, it is usually green, and it usually appears in a slide right before the word "optimized." The number is not a lie. It is a controlled experiment: one synthetic visit, from one clean machine, on one fast connection, to one page, measured under conditions your customers will never experience.
Field data is the other number. It comes from real visitors on real phones on real networks, aggregated by the browser itself, and it describes the site your buyers actually used. When the two numbers disagree, and they disagree constantly, the field is telling the truth.
The field metrics that matter map to moments a visitor can feel. Largest Contentful Paint is how long they stared at nothing before the page appeared. Interaction to Next Paint is how long the page ignored them after they clicked. Cumulative Layout Shift is whether the content held still or jumped as they tried to read it.
Each one has a threshold, but the thresholds matter less than the pattern. A site can pass every lab audit and still make its visitors wait six seconds in the field, usually because of the same repeat offender.
SEEN IN THE FIELD [REPRESENTATIVE]: LAB SCORE 96 / FIELD LCP 6.4 SECONDS / CAUSE: AUTOPLAY HERO VIDEO, 14MB, APPROVED BY COMMITTEE
The video nobody watches loads before the headline everybody needs. That is not a performance problem. That is a decision problem wearing a performance costume.
Lab scores survive in decks because they are flattering and repeatable. Field data is neither. It moves with your traffic mix, it punishes the marketing tags nobody audits, and it cannot be improved by testing on a newer laptop. Improving it means removing things, and removing things means telling somebody their thing is the problem.
Report the field numbers, at the 75th percentile, for the templates your revenue flows through. Track them over time, not as a launch-week ritual. When they degrade, treat it as a regression, because that is what it is. And when someone presents a lab score, ask the only question that matters: is that what our visitors got, or what our test machine got?
The lab tells you what the site could be. The field tells you what it is. Budgets should follow the second one.
One email when something worth reading exists. No cadence, no drip, unsubscribe stays one click.