Improving Claygent's B2B Customer Case Responses for Accuracy | Clay

Improving Claygent's B2B Customer Case Responses for Accuracy

hey team, im using Claygent's extract B2B customer cases and how can I ask it to NOT force respond if confidence is low or unable to identify customers? Right now it's forcing itself to product some sort of answer even if it's totally off (started giving me certifications, prospect company name, "unknown")

6 comments

Channeled
APP
·
·
To improve Claygent's accuracy and prevent forced low-confidence responses when extracting B2B customer cases: 1. Make your search queries more specific and well-defined 2. Double-check your prompts and search parameters before running tasks 3. Use our metaprompters to craft better prompts that will help Claygent identify customers more accurately
Channeled
APP
·
·
Looks like you're still looking for help. We're escalating this over to our support team and someone should be in touch soon. In the meantime, feel free to include any other information that may be helpful in solving your issue!
Bo (.
·
·
Hey! There are two ways to handle those low-confidence responses: 1. Quick fix: Use table filters to screen out low-confidence results. 2. Better approach: Check the Steps object (click on a row > enrichment details) to see exactly how it's making decisions. This helps you spot where the prompt might be leading to those off-target responses (certifications, prospects, etc.). From there you can use the Metaprompter to generate a better prompt. Pro tip: Use the free preview generator to test your updated prompts before running them at scale. You'll see immediately if your adjustments are steering clear of those forced responses.
Andi D.
·
·
Thanks Bo! So I'm using GPT 4o mini as firsts round of scraping and already asked it to only output results if confidence is high. From there I'm using the Claygent template made with Argon for round two -- "run if Round 1 is empty" to complete the gaps where 4o mini wasn't confident or couldn't find info. The pre-made Claygent Argon template seems to be fixed and there's no way to make modifications or to see the enrichment details. Should I just dip the pre-made Argon template and create a new column with Argon + custom prompt?
Bo (.
·
·
Hey! You're right - the pre-made template is rigid. Creating your own custom Argon prompt would actually give you more control and precision. Plus, you can feed it context about what didn't work in round 1, helping it avoid similar dead ends. The metaprompter will help you craft something really dialed in for your specific needs. Quick tip: When setting up the custom column, you can use conditionsto check if the GPT-4 mini result exists, similar to your current setup.
Channeled
APP
·
·
Hi Andi D.! This thread was recently closed by our Support team. If you have a moment, please share your feedback: