Brief intro:

After building out many different bot automations for our office, we eventually realized a smaller gap that needed to be filled in our capabilities.
You could say, we had become so efficient in other areas, we needed to start aiming at higher hanging fruit.
We use a lot of different tools for finding ownership of a given asset, some proprietary and some not, many times these tools would be able to identify the owning company of an asset or the name of a person, but would be inspite of lots of different methods of searching, would still be missing emails, or updated contact information.
We needed the ability to go to a companies website and locate key individuals such as Chief Asset officers, acquisition managers, or company owners.

My process: 

I had spent several months delving into  the documentation of OpenAI's API and various capabilities of different AI tools. I started brainstorming how these could be used for tasks around the office and decided that the ability to use artificial intelligence as though it weree a person moderating a given program matched perfectly with our need to dynamically search different company websites for information.
I then built several test concepts to prove that this capability was doable and help work out my own understanding of the processes we would need to implement. The strategy for utilizing the AI features with a Windows Power Automate "bot" looks like the following: 

Pull the target website url off of the Excel list along with any contact names provided from our data source. 

Navigate a web browser to the target url and scrape the HTML from the website. 

Send the HTML to the GPT 4.0 Turbo API in a prompt that asks for direction about where to search for contacts on the website. 

Navigate to the section of the website based on the direction given from the GPT API. 

Search that page for Contact information.

If nothing is found, request direction from the GPT API and continue navigation and search features one more time before moving on to the next company website URL and starting the search over.

If the contact name (and hopefully their email) are found on first or second attempt, catalog the information on the Excel spreadsheet and continue to the next business URL.

Final Outcome:

Again, this project has proven that advancing technologies within AI and RPA domains has incredible potential that didn't exist 12 months ago (from the time of me writing this of course). This specific bot has given us the ability to generate contact lists for any given domain, even off of a google search, with minimal modification and should allow us to connect with target real estate buyers and sellers with much greater ease and accuracy and more importantly, replace the large number of man hours needed to do a similar task.

 I estimate that this system saves around 4.5 - 5.0 working hours per list of 100 B2B contacts.
Utilizing our CoStar bot, talked about on another project page, we can generate a list of 100 B2B contacts for targeted real estate assets in about 30 minutes, so it's possible for us to go from no contacts to around 100 highly targeted, up-to-date contacts (presuming 100% success rate) in around 2 hours total.  A feat that would've taken us weeks of manual search time without this process.
This solution is also scalable, so we could run multiple sets of bots in order to multiply those numbers to any degree we want.