Brief intro:

At the outset of starting my job as an analyst for Gather Real Estate, one of the key tasks I tackled was researching potential leads for our business.
We would acquire a dataset of property owners from a target county where we had a property buyer lined up and ready to purchase.
We would scrub that dataset for properties that match our ideal property description from our buyer, and then we would manually use the website BeenVerified.com to research each person (lead) and try to locate contact information for the specific individual.
Sometimes those Excel lists had 40 properties, sometimes they had 8,000+ properties.
Average research time for each contact was about 6 minutes. So between myself, and another researcher, we were able to complete a large dataset (2,000 leads) in around 2-3 weeks of hard work. Unfortunately, the manual aspect of this task was particularly taxing because of how repetitive it was. Also, because Gather Real Estate is a small office with only 6 employees at the time, there were always other non-research related tasks that needed to be done and my workload was always heavily overburdened . 

My process: 

I started a several week journey in my limited spare time looking through various automation tools that would allow us to speed up this process. Eventually I settled upon Windows Power Automate Desktop (PAD). It comes included with most Windows computers, and has fairly advanced RPA (Robotic Process Automation) capabilities. The best feature being that we had access to it without incurring additional cost to the company which was especially handy for testing its functionality and proving an MVP version of this project.
So I  started the process of brainstorming action steps, learning about PAD's ability to perform the task needed, and testing performance.

Skip ahead about 6 months, and I was wrapping up my development of what had grown to be a fairly complex system.
Along the way, I had learned a lot about the simplest (and therefore best) processes for locating a contact through the means we had.
The singular project had now split into multiple development projects.
We now also had a bot that would research LLC's to track down a primary contact for a property owned by a corporation instead of an individual.

Final Outcome:

The final product has decreased our  production time for a research set substantially.
I've done a lot of evaluation on this particular system and the stats look like the following.

Each program is capable of completing a pace of one lead researched every 60 seconds, however, the number of captchas we encounter at that pace is cumbersome and creates a stop almost every 20 minutes or so. We have slowed that time to one lead researched every 2 minutes in order to make the captchas more manageable. 

To recap, our manual research efforts looked like one lead every 6-10 minutes with a 75% completion rate and an 80% accuracy rate on the lead contact information. Additionally, we were bound by typical work hours of 40 hours a week and no nights or weekends.

The automated operation obviously has no nights or weekend limitations and has the following characteristics:
1 lead every 2 minutes, with a 65% completion rate and an 80% accuracy rate.
Meaning a typical large dataset went from having 1,200 good leads and taking 2.5 weeks to complete.
Now we can provide 1,040 good leads in 22 hours of research time which can easily fit into a weekend while we're out of the office.