SIGNAL
Runaway AI bills push large organizations into two reflexes: hard walls that throttle experimentation, or procurement gates that turn the bill into a line item to be killed. Standard Bank's CIO Jörg Fischer, at AWS Summit Johannesburg, publicly modeled a third reflex: ask where the tokens actually flow before deciding how to discipline them.
Fischer's three-bucket split is the surface. The load-bearing move is the isolation: software engineering comes out of the general staff bucket and gets its own accounting, because the heaviest token consumption runs through developer tools like Cursor, GitHub Copilot and Claude Code, which carry the largest token bills in the software engineering bucket. Most finance teams read total spend and reach for a single throttle. Fischer reads the spend by where it lands. That is the reusable category: instrument the bill by consumption shape, not by total.
The philosophy follows: every developer carries a registered individual account with a soft cap, and is trusted to justify usage rather than blocked from access. The named falsifier is whether isolating developer tooling surfaces where metered spend actually lives; if it does not, the buckets were wrong, not the discipline.
Fischer also rejects the framing that high AI usage equals high capability, a position worth preserving against procurement-led scoreboards that will reward activity over output. The stakes are straightforward. CIOs who copy the model get a comparison test against the alternatives. Engineers inside those organizations get a seat at the justification table, which is rarer than it sounds.
Reported by Sky for Type0, from Standard Bank CIO talks about that hack and managing AI token budgets. Read the original: dailymaverick.co.za