Setting Billing Limits for OpenAI Key in B2B Enterprise Context | Clay

Setting Billing Limits for OpenAI Key in B2B Enterprise Context

Hi All 👋, We connected our Open AI Key and we received the following alert: we need to set billing limits in OpenAI. We are currently using this to list build and for enrichment. What limit should we set? For context we are b2b and our ICP are enterprises.

23 comments

Gabe E.
·
·
Hi Maria, Are you all sorted here? Please let me know if you're still having issues and how else I can help here! Gabe
Maria T.
·
·
Hi Gabe E. Our usage has cost $1,884 in 2 days - something is clearly misconfigured . Any thoughts on what could be happening?
LuisArturo
·
·
Hey there Maria thanks for reaching out, to confirm this is for the "Rasa" workspace.
Maria T.
·
·
Correct LuisArturo
LuisArturo
·
·
Let me take a look here, my initial feeling here is that there may be a table scheduled to run here, but need to check which one it may be.
1
Maria T.
·
·
LuisArturo Gabe E. were you able to look into this?
Gabe E.
·
·
Hi Maria, is your OAI account linked to other accounts or only Clay ? Do you mind sending along a breakdown of the spending for us to look into? If you provide me with an associated table I can also start digging around in there! Gabe
Maria T.
·
·
Gabe E. sending the clay table that broke our API limits let me know if you are able to identify what happened. thank you
Gabe E.
·
·
Are you able to send along the Open AI spend breakdown so I can compare? Are there any other accounts linked to your Open AI key that could be responsible for the charges?
Maria T.
·
·
the key we use is in an isolated project just for clay.com so afraid no other usage in engineering that could cause this
Maria T.
·
·
Maria T.
·
·
Maria T.
·
·
Here’s the models that it used and the spend Gabe E.
Gabe E.
·
·
Thanks for all this Maria. I'm routing it to the team and will circle as soon as I have an answer.
1
1
Maria T.
·
·
Gabe E. Were you able to look into this?
LuisArturo
·
·
Hey there Maria jumping back in here for Gabe, jumping in here for Gabe, One thing we did notice was that there was no "Max cost" value entered in here. What this does is it will go and limit the amount of money used row in the AI column and if there's no value set it will default to 50 cents being the max value. Now won't always hit 50 cents, this is more of an upper limit for how many how much money it will use. But when we do set it to a lower value it does a good job of respecting that limit and it could help bring the cost down.
(14 kB)
Maria T.
·
·
Thanks, Luis. I will pass this along. Do we know why we went over the limit?
Gabe E.
·
·
Hi Maria, I'm looking closer here now. Here are some general recommendations based on the logs you passed along. Please let me know how this looks and what else would be helpful here! 1. Massive Request Volume ** 553,242 requests in one day (July 25th) is extremely high ** This suggests batch processing or retry loops ** Your Claygent likely processed thousands of LinkedIn profiles 2. Token-Heavy Operations ** LinkedIn profile analysis generates large input tokens ** Complex prompts multiply token usage per request ** Profile scraping includes lots of text data 3. Model Selection Impact ** GPT-4.1 Mini used 924M tokens in one day ** Even "mini" models add up with high volume ** Input tokens (875M cached, 40M uncached) dominated costs How to Reduce Costs: 1. Batch Size Control ** Limit Claygent to 50-100 rows at a time ** Add delays between batches 2. Prompt Optimization ** Use shorter, more focused prompts ** Reduce LinkedIn profile analysis depth 3. Model Selection ** Switch to GPT-4o Nano for simpler tasks ** Use Claude or Gemini (often cheaper) 4. Data Preprocessing ** Pre-filter obvious non-matches before AI analysis ** Use *existing Clay data instead of live profile scraping
Maria T.
·
·
thnaks, Gabe E. this is really helpful :_
Maria T.
·
·
Ihave passed this along
Gabe E.
·
·
So glad! Please let me know how this goes.
Maria T.
·
·
Will do
Gabe E.
·
·
Happy building!