Conduct real user testing to improve conversion funnels
All the metrics in the world will never substitute for real user testing. To improve conversion rates, user testing should be sought from the start with real transactions. Test transactions are not enough to equate to the experience of both personally conducting and personally-witnessing the funnel via a real user.
It's easy to fall into making assumptions over how visitors and customers will react to the given options. It's less clear that there may be problems without taking the time to view from another perspective.
What is obvious to someone who is already familiar with the product/brand, may pose problems for those without prior knowledge. Witnessing how visitors use a site can transform the framework of the site, in a way that split testing would never make possible.
User testing may be conducted via literal first hand witnessing of a likely customer going through the site, from heatmap tracking, or more detailed tracking via tools such as HotJar. Overall, the goal is to replicate the physical checkout process in the digital world in manners that allow for feedback and improvements.
Data is extremely important to improving conversion rates. No amount of data can entirely obfuscate user testing. There's untold insights available when looking deeper than data and into real user testing.