AI cost isn’t a pricing problem. It’s a consumption problem — and consumption compounds silently in the gap between a pilot and a rollout, long before anyone’s watching.
Somshankar Bandyopadhyay’s reporting for Khaleej Times is one of the few in this cycle to name that mechanism instead of comparing price tags, in an interview with Sherif Tawfik, Chief Business Officer at Core42, quoted in the piece: “The biggest hidden cost is not the model itself, it is the unmanaged consumption... the real challenge is that consumption compounds before the controls to manage it are in place.” Core42 shipped Compass, a metering and governance platform, on the strength of that diagnosis — a company willing to put a problem on the record and then build a product around it is corroborating evidence the problem is real, not a reason to discount it.
Doctors diagnose you and prescribe the treatment in the same visit, and nobody calls that a conflict of interest.
Here’s why that works, set against what happens when an AI vendor does the same thing today:
Structural Comparison
Diagnosis, prescription, and accountability
| Dimension | A doctor’s diagnosis-and-prescription | A vendor’s diagnosis-and-product |
|---|---|---|
| Who sets the standard | A licensing board defines the standard of care before any diagnosis is trusted | No license, no board — the vendor sets its own standard |
| What happens if the fix doesn't match the problem | Malpractice liability, peer review, professional sanction | Next quarter's invoice, discovered after the decision is made |
| Who can get a second opinion | Any patient, at any time, structurally expected | No equivalent role exists in most AI procurement |
| Who reviews the fit over time | Ongoing standard-of-care obligations, not just at time of treatment | Nobody, once the contract is signed, unless someone is explicitly assigned to |
The gap isn’t that a vendor pairs a diagnosis with its own remedy — that’s the normal shape of expert relationships. The gap is the missing right-hand column: nothing external checks whether the prescribed platform actually fits the scale of the problem, before or after it ships.
Consumption Dynamics
Running the numbers the interview skips
200 req / day
50,000 req / day
A metering platform will show that gap once it exists. It won’t tell you who was supposed to be watching for it, because nothing external requires anyone to be.
This is the same gap our Problem Framing piece names from the buyer’s side: a problem statement borrowed wholesale from whoever named it first, vendor category and all, is a problem statement nobody on your team actually owns. And it’s the same audit our ROI Definition piece asks you to run before signing any AI business case — trace the assumption to its source, or it’s a guess wearing a decimal point.
Closing Test
Before buying a tool to manage AI spend:
Don’t ask whether the vendor has a conflict — that question misses the point, and most experts you already trust have the same structure. Ask instead: who plays the board’s role here?
Name the person accountable for reviewing the routing rules against actual volume, on a set schedule, not just at rollout. If no one can answer, you haven’t found a conflict of interest. You’ve found the licensing board that was never appointed.