Web vitals impact
- DataMilk monitor's Web Vitals for each page to make sure it doesn't downgrade the performance of the website. Signals collected: LCP, FID, CLS, FCP, TTFB
- Smart Components are planned to not impact user experience by avoiding the shift of big elements on the page.
Average impact across customers:
- FCP (First Contentful Paint) was, on average, 2585.0 ms for treatment vs 2584.4 for control.
- For LCP: 3668.7ms for treatment vs 3670.2 ms for control
- No significant difference for median percentiles for these values:
- FCP: 1953.2 (Treatment) vs 1948.0 (Control)
- LCP: 2828.9 (Treatment) vs 2832.1 (Control)
- No significant difference for median percentiles for these values:
Performance/Latency
DataMilk monitors web-vitals and stores them on the logs. To monitor web-vitals for the treatment arm (when experiments are applied) DataMilk has to store this information and process it to compare the performance in respect to the control arm (when no experiments applied). This involves intensive data storage and processing costs. Refer to web vitals impact above for more information.
Using sophisticated machines, caching and other infrastructure techniques, our AI models are lightning fast. 95th percentile is at 25ms.
The size of the DataMilk JavaScript is also very small to ensure that we don't negatively impact the users bandwidth. Typical downloaded package size is under 40 kilobytes.
DataMilk suggest customers use async when loading the script so that it can be loaded in parallel with the page. We also serve the script from data-centers around the world to reduce latency.