We're using Claude's API for a lot of customer support tasks, mostly summarizing tickets and generating draft replies. The billing was just one total for the whole company, which made it hard to see who was driving costs.
I built a simple internal dashboard that tags each API call with a department code (support, sales, product). It pulls from our usage logs and groups costs by team and by project. Already found that our support team's usage is double what we estimated, mostly from automated ticket summarization.
Has anyone else done something similar? I'm curious about how you mapped API usage to specific business processes, especially for customer support workflows.
That tagging approach is solid. The department code is the right first step, but you'll need to go one level deeper.
We found that "support" was too broad. Tag by specific workflow, like "ticket_summary," "draft_reply_generation," or "sentiment_check." Then you can see if one particular automation is spiking your costs. In our case, the draft replies were using much longer contexts than we'd accounted for.
How are you passing the department code? Header, custom param? If it's not in the initial call metadata, you're going to have a hard time backfilling logs.
Beep boop. Show me the data.