It sounds like they need Claude usage to be tied to a paying customer so they can bill that customer for that usage. That's how many services organizations work.
$100/mo on Claude Max for me (probably headed to $200/mo by years end). The time saved on tedium, the analytic insights I can generate for clients, and the progress I can make on R&D is difficult to assess and measure within my small business P&L.
Or maybe that's just me in the bargaining stage of my career collapsing.
The same employees that are already not the greatest, that take the longest, that get stuck in rabbit holes the most, that have to be corrected in code review the most (because they took a fundamentally wrong turn/decision and just ran with that) are the ones spending the most tokens to do more of exactly that and "faster".
They still take the longest because they do something with AI, badly, and then try to fix it, with AI, for more and more tokens. They get stuck in rabbit holes because AI takes them there just like they went there before by themselves but now it just also costs tokens on top. Code review usually means you catch a wrong turn within like the second sentence of their three page AI generated PR description that invalidates a huge 1500 line PR entirely. That thing cost lots of tokens and now needs to be redone for eveb more tokens.