New Claygent Prompts for Researching and Qualifying Company Metrics

Maximilian J. · 2024-11-29T16:59:16.117Z

Hi guys important question: • I have created three Claygent Prompts that research and qualify energy related metrics for a company • Results are very good so far Problem: The prompts sometimes take up to 5 minutes each. How can this be lowered significantly???

New Claygent Prompts for Researching and Qualifying Company Metrics | Clay

8 comments

· Sorted by Oldest

Andreas W.
·
·
Hey Max! Which AI model are you using? Faster models = faster runs but they will probably break your beautiful results. I woulnd’t recommed that. Also, for me Claygents ususally time out after 2 mins or so (if they time out). You sure it takes 5? Claygent’s just can take a bit - why is time to run so important?
Maximilian J.
·
·
Hi Andreas! Thanks for reaching out 🙂 the processing time is now kind of fixed, however the timeout you have mentioned is more of an issue. Sometimes cells just seem to be stuck at "Queued". Is there a way to fix that?
Andreas W.
·
·
this is likely from the amount of material the agent looks at. If it happens too often, you may wanna break up the prompt/tasks into multiple. You could also run the same prompt with a conditional formula that it should only run if the first one failed and merge the results. Sometimes this approach helps but the first one (breaking up) will be more reliable
👍1
Maximilian J.
·
·
Thanks for the advice Andreas W.!! We have decided to continue with Clay as our tool of choice and bought the most complete plan. Are there any detailed guides on how to do more complex prompt engineering, specifically in the case of the Claygent? I know about the AI support but it would help me more if there were any in-depth guidelines on how to engineer structure, language and level of detail for the prompts to achieve the best possible output. Generally, I encounter the following issues:
I am trying to set-up two prompts that research specific metrics such as the annual revenue and a third one that combines insights with further research (on industry benchmarks and more) to conduct an energy-intensity benchmarking for companies.
However, I am struggling a little bit with data output standardization and instructions in terms of finding and returning all relevant information available on the company website. Sometimes, some of the response cells do not contain any value at all and sometimes it returns "not available" although the information is clearly available on the website.
Furthermore, I want the prompt to always prioritize any information found on the company's own website. However, sometimes it finds ambiguous information and does not return neither of it.
Andreas W.
·
·
Hey Maximilian J. First: congratulations, you have the most incredible GTM targeting tool at your fingertips now! I can only touch a few high level ideas on Claygent, but here we go:
do the manual work first. Have a set of ideally 20 companies where you know the true answer and compare
look at “steps taken” and re-write the prompt to account for where it goes wrong
Use AI in Claude to write prompts, but do it manually too, and always understand the prompt that is generated
Experiment with logic: Maybe it’s best to scrape the website first, then use Claygent on that scrape etc.
Maybe use Google first and find results there, then move on to Claygent
… This is art and science, and you will learn while you do it. A friend of mine has a paid course you may want to consider, he’s a legend in the Clay community. Grüße nach Berlin!
Maximilian J.
·
·
Thanks for the input! Will read further into these issues and try a few things 🙂
1
Jordan C.
·
·
Maximilian J. I am hosting a Clay webinar TODAY and doing this live with details from my course. Here is the summary of the course (please don’t share as this is a part of the course itself too) https://youtu.be/M6bmwj1dvvo overview of the course: https://www.linkedin.com/posts/jordancrawford_the-299-agent-7-course-is-here-youll-learn-activity-7247669670940401665-xrjq c ourse payment link: https://buy.stripe.com/dR64gj0KD5w86gU3cx Goodluck on your Clay journey
Maximilian J.
·
·
Jordan C. thanks for the message!! I have already registered as soon as I've heard claygent haha 😄 excited for later!!
❤️1

Andreas W.
·
·
Hey Max! Which AI model are you using? Faster models = faster runs but they will probably break your beautiful results. I woulnd’t recommed that. Also, for me Claygents ususally time out after 2 mins or so (if they time out). You sure it takes 5? Claygent’s just can take a bit - why is time to run so important?
Maximilian J.
·
·
Hi Andreas! Thanks for reaching out 🙂 the processing time is now kind of fixed, however the timeout you have mentioned is more of an issue. Sometimes cells just seem to be stuck at "Queued". Is there a way to fix that?
Andreas W.
·
·
this is likely from the amount of material the agent looks at. If it happens too often, you may wanna break up the prompt/tasks into multiple. You could also run the same prompt with a conditional formula that it should only run if the first one failed and merge the results. Sometimes this approach helps but the first one (breaking up) will be more reliable
👍1
Maximilian J.
·
·
Thanks for the advice Andreas W.!! We have decided to continue with Clay as our tool of choice and bought the most complete plan. Are there any detailed guides on how to do more complex prompt engineering, specifically in the case of the Claygent? I know about the AI support but it would help me more if there were any in-depth guidelines on how to engineer structure, language and level of detail for the prompts to achieve the best possible output. Generally, I encounter the following issues:
I am trying to set-up two prompts that research specific metrics such as the annual revenue and a third one that combines insights with further research (on industry benchmarks and more) to conduct an energy-intensity benchmarking for companies.
However, I am struggling a little bit with data output standardization and instructions in terms of finding and returning all relevant information available on the company website. Sometimes, some of the response cells do not contain any value at all and sometimes it returns "not available" although the information is clearly available on the website.
Furthermore, I want the prompt to always prioritize any information found on the company's own website. However, sometimes it finds ambiguous information and does not return neither of it.
Andreas W.
·
·
Hey Maximilian J. First: congratulations, you have the most incredible GTM targeting tool at your fingertips now! I can only touch a few high level ideas on Claygent, but here we go:
do the manual work first. Have a set of ideally 20 companies where you know the true answer and compare
look at “steps taken” and re-write the prompt to account for where it goes wrong
Use AI in Claude to write prompts, but do it manually too, and always understand the prompt that is generated
Experiment with logic: Maybe it’s best to scrape the website first, then use Claygent on that scrape etc.
Maybe use Google first and find results there, then move on to Claygent
… This is art and science, and you will learn while you do it. A friend of mine has a paid course you may want to consider, he’s a legend in the Clay community. Grüße nach Berlin!
Maximilian J.
·
·
Thanks for the input! Will read further into these issues and try a few things 🙂
1
Jordan C.
·
·
Maximilian J. I am hosting a Clay webinar TODAY and doing this live with details from my course. Here is the summary of the course (please don’t share as this is a part of the course itself too) https://youtu.be/M6bmwj1dvvo overview of the course: https://www.linkedin.com/posts/jordancrawford_the-299-agent-7-course-is-here-youll-learn-activity-7247669670940401665-xrjq c ourse payment link: https://buy.stripe.com/dR64gj0KD5w86gU3cx Goodluck on your Clay journey
Maximilian J.
·
·
Jordan C. thanks for the message!! I have already registered as soon as I've heard claygent haha 😄 excited for later!!
❤️1