What Tactical costs your store

Very little, and nothing on the critical path. The browser snippet loads deferred and reports after the page has rendered, and server-side detection does nothing at all for human visitors. Here is exactly why, and how to check it yourself.

01The browser snippet

• About 8 KB gzipped (22 KB uncompressed), loaded with defer, so it never blocks parsing or rendering.
• Sends off the critical path. The first report goes about a second after the script starts, as a background fetch with keepalive (sendBeacon where fetch isn’t available), and a final one when the page is hidden. Nothing waits on the response.
• No DOM writes for people. It observes; it doesn’t touch the page, so there’s no layout shift and no reflow. The one exception is an opt-in banner a store can show to declared AI agents.
• Classification runs on Tactical’s servers, not in the browser: no scoring work runs on your visitor’s device.

02Server-side detection (WooCommerce plugin / middleware)

• Matched-only firing. An event is created only when the request’s user agent matches a known AI crawler. Humans and unknown user agents cost a single in-memory lookup and nothing else: no HTTP call, no database write.
• Non-blocking delivery. The report is fire-and-forget, with a 1 second timeout in the plugin and 500 ms in the middleware; the visitor’s response never waits on it.
• Burst-guarded. A short per-(agent, URL) de-duplication absorbs crawler storms before they leave the site.

Net effect: a human page load makes no extra outbound request from your server. The only requests Tactical makes there are for the AI traffic you’re trying to measure.

03Measure it yourself

You don’t have to take our word for it. On your own store:

1. Open a product page in an incognito window.
2. Chrome DevTools → Lighthouse → Mobile → Performance. Run 3× with the Tactical snippet removed; take the median.
3. Add the snippet back. Run 3× again; take the median.
4. Compare Performance score, LCP, Total Blocking Time and CLS.

Lighthouse scores move a few points between runs on their own, so compare medians, not single runs. We haven’t published a before-and-after of our own yet; when we do, it goes here with the raw reports.