Tracking AI spend with intention is the right instinct, and the trap is tracking the invoice instead of the outcome. Most finance teams can tell you the monthly model bill to the dollar and cannot tell you which workflow it replaced or what that workflow used to cost in hours. The spend is visible, the saving is not, so the ROI conversation defaults to cutting tokens. The number worth instrumenting is cost per completed task, not cost per API call.
I couldn't agree more about the trap, Kevin. That's why it is important that teams run actual usage through their P&L, not $s invoiced. That's why I included a suggestion to add an accrued and prepaid account for tracking AI spend.
One vendor that I was introduced to recently is Fancysauce (https://fancysauce.ai/). They claim to help with some of this.
Tracking AI spend with intention is the right instinct, and the trap is tracking the invoice instead of the outcome. Most finance teams can tell you the monthly model bill to the dollar and cannot tell you which workflow it replaced or what that workflow used to cost in hours. The spend is visible, the saving is not, so the ROI conversation defaults to cutting tokens. The number worth instrumenting is cost per completed task, not cost per API call.
I couldn't agree more about the trap, Kevin. That's why it is important that teams run actual usage through their P&L, not $s invoiced. That's why I included a suggestion to add an accrued and prepaid account for tracking AI spend.
One vendor that I was introduced to recently is Fancysauce (https://fancysauce.ai/). They claim to help with some of this.