Hey Clay team -- I'm coming back to a challenge that I've struggled to solve and would love feedback or direction. I have a list of addresses of businesses (commercial offices, universities, etc.). What I'm trying to do is generate a list of people who work at those addresses and meet certain title criteria. I have the title criteria dialed in, but starting from addresses and finding people based on that is where I'm stumped. Would love to hear any ideas!
No table yet! 🙂 I can easily set up a sample of addresses, though.
Hey there Andrew sorry for the delay, thanks for reaching.
In "New Table" you can use the Location filter to narrow search area down to Cities, Staes, and Countries. Unfortunately using specific address isn't an option. But from there you can enter in the rest of your criteria and see if you get any hits.
Thanks LuisArturo -- Interesting. If I narrow the results to a city (or set of cities), can I reverse enrich the contacts for address where they work?
Hmm that's an interesting idea, let me test it it to see if there's a way to make that work
Thank you!
So yes this can be done, Here is a Loom that shows how you can do this. https://www.loom.com/share/03be6291a3de4a829062374bf2f75b22
LuisArturo -- This is great! Thank you for working on this. In your example, row 8 is a common scenario for what I am running up against. The person is in San Francisco, put on the company LinkedIn enrich, the address pulled is New York. Ideally, I'd like to get the company location in the city (or closest to) the city where the person is located. In my scenario, it's common for the people I'm looking for to be employed by large, national facility services companies but work at a specific location in or near their city. Any thoughts on this?
Yes here is a follow up loom, simply because its easier to explain through video rather than typing it out. https://www.loom.com/share/f18c677263c34b1ab8f50b73acebd71b
Thanks for the help on this. I'll give this a try and see what I come up with!
Okay LuisArturo... I've had a chance to play with this. The challenge I'm running into is the variable number of locations for the company. I'm trying to do this at scale, so I don't want to scrub through each row. So, I tried using a Claygent prompt to help determine the correct address from the list. This is proving to have inconsistent results. Do you have any ideas on how I might be able to do this? Here's my table: https://app.clay.com/workspaces/10648/tables/t_KDC18qjNakIj/views/gv_YJUnSyaaNLDz In the "Addresses Found" view, the first row is a good example of what I'm working through.
Hey there Andrew sorry for the delay taking a look.
No worries! Thanks for the help.
Hey Andrew so I spent some time reworking the prompt and the one I ended up with, looked to have an accuracy around 90%. I ran it with about 25 of your rows and the only one with some questions was the first one. I also refunded you the credits used during testing
Thanks! I'll take a look!
Quick question... We're still getting text other than just the address. Any thoughts on how to address this?
Also, Row 21 is an example where there was an address available in the company enrichment that was not returned. It's almost like I need another way to process through this. I'm definitely open to other ideas.
Hey Andrew! I added a quick indication to the prompt asking it to return an address and nothing else, otherwise "NOT FOUND".Also used the Answer Formatting section with an example so it's more consistent. Appears to be understanding that logic a bit better now.
For any edge cases where it missed, I'd try re-running those separately. More often than not that should fix those.
Thanks! I'll take a look!
I still feel like the results are unreliable on this. Is there a way to map all addresses from the company enrich to their own columns so that I can process them individually?
Yes, you can use the Extract Field from Object integration to pull just that from each array, Andrew T..Example:: [https:://www.loom.com/share/8b10ccca0b794a89b93242c0086a9f5d?sid=97951552-bc6c-4639-952f-e5a27e2774c7](https://www.loom.com/share/8b10ccca0b794a89b93242c0086a9f5d?sid=97951552-bc6c-4639-952f-e5a27e2774c7)
Sorry, this is not mapping them individually to their own rows/columns but instead of excess data, it's just the address field.
That may help decrease the token count for Claygent (potentially increasing accuracy) or you could use another method by just having them nested as addresses only. There's a mapbox integration that can measure the distance between two addresses but I believe that provider charges 3 credits per result.
Yep! Definitely a bit more complicated when it's not close 100% success rate and that's what you really need. I'd probably try using chatGPT-4 with a regular chatGPT/UseAI column, but if you don't have locations and want to web scrape, then Claygent is the way. However, I know that could increase costs a little bit depending on the token count
Darn! quite tricky with LLMs 🫠
Absolutely!