Issues with Claygent Prompts: Empty Cells and Undefined Results

Hello, yesterday I ran 3 subsequent claygent prompts for 450 rows. However I have encountered several critical issues:

Some cells being completely empty, neither "queued" or "retrying" or "time out"
Some cells with the prompt result "not defined" and no further explanation on the issue

Especially the latter is worrying as I have been charged for each of these runs, despite not delivering any results, nor insights on what went wrong. To put this into perspective, this happened to about 1/20 prompt runs which may not sound much but it requires

a lot of manual time to monitor the whole process and re-run faulty cells individually and
causes additional costs of about 5% which is especially relevant for larger scale operations as I am currently trying to reduce costs by making the whole process more lean and efficient

Thanks for the help in advance!!!

10 comments

Channeled
APP
·
·
I apologize for the issues you've encountered with your Claygent prompts. Let me address your concerns and provide some steps to improve your results:1. Empty cells and "not defined" results: • These issues can occur due to various reasons, such as API timeouts or processing errors. • We're continuously working to improve our system's reliability and reduce such occurrences. 2. Charges for unsuccessful runs: • I understand your frustration regarding charges for runs that didn't produce results. I'll make sure to forward this feedback to our billing team for review.To improve your Claygent results and minimize these issues, here are several steps you can take:1. Optimize your search queries: • Make your queries specific and well-defined • Break large data requests into smaller batches • Monitor your usage limits and quotas 2. Follow these best practices for your prompts: • Start with a clear system prompt that provides context • Include step-by-step instructions • Be explicit about your desired output format 3. Structure your prompts effectively: • Set proper context for the task • Provide clear instructions to avoid incorrect assumptions • Specify exactly how you want the data formatted and returnedFor better accuracy, use our metaprompter feature to help craft more effective prompts. This tool will help generate AI-approved prompts that are more likely to give you better outcomes.If you continue to experience issues or need further assistance, please don't hesitate to request human assistance by clicking on "Talk to a person". Our support team will be happy to help you troubleshoot and optimize your Claygent usage. Sources: Prompt Engineering Crash Course [Clay 101] - Clay University Prompt Engineering Crash Course [Limitless Research] - Clay University
Maximilian J.
·
·
How do I achieve this:
Break large data requests into smaller batches
Monitor your usage limits and quotas
Channeled
APP
·
·
When working with large data requests in Claygent, here are the key points to manage your processing:For batch processing:** Break down your large data requests into smaller, more manageable segments ** Process these smaller batches sequentiallyFor monitoring usage:** Keep track of your usage limits and API quotas ** Note that processing is limited to 10 records per second per workspace ** Be aware of the maximum burst capacity of 20 records ** Ensure each payload doesn't exceed 100KBTo optimize performance:** Use specific and well-defined search queries ** Verify your network connection is stable * Double-check your prompts and search parameters before running tasks Source: Import Data to Clay Using Webhooks
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!
Maximilian J.
·
·
Yea but how exactly do I monitor usage and all the mentioned metrics?
Maximilian J.
·
·
Also are there any really specific guidelines on how to build prompts that are optimized the way clay works technically?
Stefano
·
·
Hi Maximilian, thanks for reaching out. Our team is currently working on an update that will allow you to monitor your usage limits and quotas, but this feature is not available yet. In the meantime, if you're using an external API, you can track your usage directly within the API. For example, if you have a ChatGPT key, you can monitor your usage on OpenAI's website. When it comes to best practices with AI prompting, I can offer three main recommendations to help you get started: 1) Try using ChatGPT directly in your browser (not through the API on Clay) and ask it to help you structure your prompt. Provide it with some context, and it should guide you in crafting a more effective prompt. 2) Under the "Examples" section of your OpenAI enrichment, you can use "good" answers as examples for ChatGPT to follow. To do this, simply select "Add Examples" and then choose "Add from previewing rows." This will help ChatGPT better understand the kind of responses you're looking for. https://downloads.intercomcdn.com/i/o/w28k1kwz/1275264676/e47b43f51e3874293f03fcd70899/CleanShot%2B2024-12-03%2Bat%2B13_30_51-402x.png?expires=1733409000&signature=49aa9ffb696991b8023081da0ccc474f8e79db1d71110410473f60746b749e10&req=dSIgE8t4mYdYX%2FMW1HO4zds5Pf9nzZnMn9%2Fnvk5ThvXpog%3D%3D%0A Then, a menu will appear where you can give a thumbs up or thumbs down to the answers you think are good or bad. This feedback will help improve the prompt when it’s re-run. 3) Be as specific as possible about what you want or don’t want to see in the output. For example, if you're looking for a specific text length, make sure to specify that clearly in your prompt. Let me know if this helps!
Channeled
APP
·
·
Hey there - just wanted to check in here to see if you needed anything else! Feel free to reply back here if you do.
Channeled
APP
·
·
We haven't heard back from you in a bit, so we're going to go ahead and close things out here - feel free to let us know if you still need something!
Channeled
APP
·
·
Hi Maximilian J.! This thread was recently closed by our Support team. If you have a moment, please share your feedback: