From my article Intelligent Automation:

What is the best starting point for leveraging Intelligent Automation? One strategy is to look first at repetitive human tasks, where users are bogged down performing tedious work, repetitive steps, or otherwise where users are shifting back and forth among different application interfaces as part of the task or process step. These scenarios offer low hanging fruit, yet plans shouldn't be limited to the easy targets. Rather, your strategy should lay the groundwork for horizontal scale, tying together discrete moments of automation within a more comprehensive, end-to-end process. To support this, develop a clear model for the separation of concern between the layers of Intelligent Automation (as illustrated in Figure 4) and specifically that between the capabilities traditionally associated with BPM and RPA. It is worth noting that BPM was never designed to fully replace the work done by human beings, but rather to facilitate that work by assigning tasks, sequencing steps, enforcing rules, and other means of work management. In contrast, RPA in fact is purpose-built specifically to replace work delivered less efficiently and effectively when performed by humans.

But, what do you think is the best starting point?