Hey all ✋ We are facing some inconsistencies with the salesforce lookup functionality. Some companies cant be found even if there is an exact match. For instance SHD AG from row 113 is exactly called SHD AG in our Salesforce
Hi Felix, thanks for getting in touch! can you try to match per domain instead of name?
I marginally got better matches but thats not really solving our problem. It also seems quite arbitrary. Sometimes it works sometimes not.
Maybe this is the issue
click autoupdate here please
hm, the auto update is off on purpose. I ran the matching tests manually and it did not improve the results.
even if the URL match would be slightly improved it would still kill the usecase because we dont have exact URLs for all our companies in salforce. I absolutely dont understand how the “contains match” can be so flawed
Hey Felix, sorry for the delay. Jumping in here!
Yeah, it's unclear if this is on our end not processing the call correctly or if it's on SFDC's end i.e. not working/matching properly for some reason. Looping in Colin P. to help us out in case we're missing something :)To confirm, Felix B., that row 113, SHD AG should exist in your SFDC records which should be a match with Company Name, correct?
Felix B. looks like the table was pushing in the "original url" exactly. most of these original urls have a leading http://, https://, or www, and a trailing / or subfolder. i assume your entries in salesforce mostly have urls like shd.de, not https://www.shd.de/. since the lookup record column is looking to see if there's a salesforce record containing the string, searching for https://www.shd.de/ won't return any results, because that string technically isn't contained in shd.de. anyways, i added in a new formula column to remove those http / https protocols, www domains, and trailing slashes / subfolders from Original URL. i then passed this column as the search term in (Salesforce) Lookup Record. this resolved the issue for the given example (in row 113, SHD AG is now found). also gave you some extra credits for the trouble. let me know if you have more missing rows with salesforce lookup record and we can troubleshoot those ✅
Colin P. fair point, works already way better now! Problem though is that for most companies in our CRM we do not have a URL and run Clay to get URLs, but obviously need to know if the company is in the CRM already to avoid duplicates. If you check the last row within the table (Clay Test - Salesforce Lookup) it does not get any matches based on the company name, although we disabled the exact match. This clearly feels like a bug, doesnt it? Weirdly, this seems to be quite arbitrary. 2 weeks ago the lookup based on the company name worked very well.
Okay I figured out that the results are way better if I normalize the name with GPT4. You think thats as good as it gets or is there another way to keep improving?
depends on how you're getting this initial list of names. looks like they were dropped in like a CSV or manually entered, not pulled from any datasource in clay - are there any other unique datapoints you can pull in from wherever this dataset was generated in the first place? similarly, what other uniquely identifying datapoints are you collecting about companies in salesforce? a few possible ones that come to mind: email, lead id, import id, company phone, maybe even street if most of these companies aren't on the same street. i don't know how accurate these are, because i can't look inside your sfdc account, but any of these that are actually collected for companies in your sfdc would be good fallbacks. to improve your current approach, you could also run a google search for the company's name in clay. the first result will probably have the company's actual normalized name in the title, so that might improve accuracy compared to just asking gpt-4. another thing you could do (which would be less accurate) would be to use gpt to generate a keyword or two about the company description, then run the "contains" filter to see if that's included in the company description in salesforce. you'll probably get back a ton of entries, especially if these companies are in the same industry / have similar descriptions. you could then ask chatgpt to rank the salesforce descriptions by which one is a most likely match with the company description, and assume that's the matching company in salesforce. you could even supercharge that above approach by using claygent to do research on the company, and see if that provides additional context to help you match. one more thing that'll be possible in the next few weeks - we'll be launching a gpt-vision integration. if you collect company logos rigorously in the "photo url" field, you could pass gpt-vision a salesforce record's logo and a linkedin enriched company logo and ask it if they're the same. the issue is, these are still all workarounds. you're going to be limited with a salesforce lookup by what you can use to identify these entries. if most entries in sfdc don't have a domain, or have a name completely different from the name included, or don't have any other data points that can be inferred from the list given, i don't think anyone - clay, other software, a smart human being - could match the entries in this table perfectly with salesforce. instead, it seems like both a smart human and clay would fall back to googling the company, reading the description, checking it against salesforce entries, etc until they were reasonably confident they had a match. hopefully the above approaches give you something analogous to that process; the caveat is, just like with a human manually doing this research, you can't be certain the matches will be perfect!
okay, fair points. Super helpful! I’ll try out a few approaches! 🫶
.png){kind=logo}
