Your AI Bill Is Leaking, And Procurement Cannot See It

Decentralised AI bought on corporate cards has become a dollar-denominated cost that boards are paying for many times over.

Sign up for my Substack daily AI newsletter here.

See my AI Training course portfolio for corporate Business Leaders here.

Article link: https://open.substack.com/pub/johanosteyn/p/your-ai-bill-is-leaking-and-procurement?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/

For the first two years of enterprise AI, the rule was simple: spend fast, experiment widely, and worry about efficiency later. That era is ending, and the place it is ending first is the budget. The trouble is that most of the spend does not arrive as a single line a board can see. It arrives one card swipe at a time.

CONTEXT AND BACKGROUND

The market is turning from spending at all costs toward a sharper question of intelligence per dollar, where the metric that increasingly matters is the cost of one useful output rather than the prestige of the model that produced it. Surveys now put unapproved AI use at roughly two-thirds of office workers, who reach for the tool they prefer over the one the company sanctioned. Each of those choices is also a purchase. The result is a shadow stack of AI subscriptions that finance never approved, and procurement cannot see, growing in the gap between how fast staff adopt and how slowly policy adapts.

INSIGHT AND ANALYSIS

The leak follows a predictable path. An employee tries a free tool on live work, the results spread, peers upgrade to paid tiers on personal cards, and the tool settles into daily workflows long before anyone holds the contract to question the price. Spread across twenty teams, a firm can fund a dozen duplicate seats of the same model with no volume discount. As one illustration of the waste, Torii notes that ten separately bought plans of a single popular image tool can list around thirty percent higher than one enterprise agreement. These charges hide in plain sight, as software reimbursements on personal cards, as auto-renewals buried in a shared team card, as pay-as-you-go cloud charges logged under miscellaneous services, surfacing only when someone reconciles the quarter. By then the spend is not a pilot to approve but a habit to unwind.

For a South African board the exposure carries a second edge that Silicon Valley never feels. Most of these tools are priced in dollars while the company earns and budgets in rands, so a bill can rise even when usage is flat, purely because the currency moved. A scatter of unapproved dollar subscriptions is therefore not only duplicated spend, it is an unhedged currency position sitting outside treasury’s view, approved by no one and forecast by nobody. Multiply a modest monthly charge across departments and a weakening rand, and a rounding error becomes a variance the finance team cannot explain.

I have previously written about why trusted AI fails in the real world, where I argued that buying an AI system is itself a governance act, and that without procurement discipline an organisation is left with no clear audit trail, no clear cost ownership and no credible control when something goes wrong. Shadow AI is that failure in its most ordinary form, repeated quietly across departments until it is the norm rather than the exception.

IMPLICATIONS

This belongs on the board agenda rather than the help desk, because it is a procurement, margin and currency problem before it is a technology one. The same decentralised buying that bloats the bill also routes company data through tools no one assessed, so the cost leak and the data risk share a single cause, which means one control can address both. There is a strategic cost too, beyond the invoice. When the same capability is bought ten times over, the organisation forfeits the negotiating leverage, the volume pricing and the vendor accountability that a single consolidated contract would command. A board that cannot say how many AI tools it pays for, in which currency, and to what end, is not governing the spend. It is discovering it after the fact, in a variance report, and explaining it to an audit committee that will reasonably ask who was watching.

CLOSING TAKEAWAY

The remedy is not a blanket ban, which only pushes the behaviour further into the shadows and costs the firm the productivity it was chasing. It is to treat AI spend as the strategic cost it has become. That means a single inventory of every AI subscription and its owner, centralised chargeback so spend is tied to a use case, quarterly vendor reviews, clear procurement thresholds, sensitivity to the exchange rate at renewal, and a sanctioned, properly bought set of tools good enough that staff stop reaching for their own. None of this slows adoption; it makes adoption legible. The goal is not less AI. It is AI a board can actually see, price and stand behind.

Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net