Most AI business cases are 𝐨𝐧𝐥𝐲 built around upside: productivity gains, faster workflows, lower labor costs, better customer response times, more automation, higher throughput, competitive advantage. Those benefits may be real. However, if the business case quantifies the upside while leaving the downside unquantified, it is incomplete.

A CFO would not approve a major acquisition, credit exposure, hedging strategy, or capital project without considering reasonably likely downside scenarios. AI should be no different.

AI capital allocation cannot be limited to: 𝐖𝐡𝐚𝐭 𝐞𝐟𝐟𝐢𝐜𝐢𝐞𝐧𝐜𝐲 𝐰𝐢𝐥𝐥 𝐭𝐡𝐢𝐬 𝐀𝐈 𝐬𝐲𝐬𝐭𝐞𝐦 𝐜𝐫𝐞𝐚𝐭𝐞? It must include: 𝐖𝐡𝐚𝐭 𝐫𝐞𝐚𝐬𝐨𝐧𝐚𝐛𝐥𝐲 𝐥𝐢𝐤𝐞𝐥𝐲 𝐞𝐱𝐩𝐨𝐬𝐮𝐫𝐞 𝐝𝐨𝐞𝐬 𝐭𝐡𝐢𝐬 𝐀𝐈 𝐬𝐲𝐬𝐭𝐞𝐦 𝐢𝐧𝐭𝐫𝐨𝐝𝐮𝐜𝐞?

That exposure may come from incorrect outputs, customer harm, regulatory scrutiny, biased decisions, data leakage, contractual breach, IP claims, business interruption, financial misstatement, rework, brand damage, or over-reliance on automation.

Not every exposure becomes a reserve. Not every AI risk requires disclosure. However, if the use case is material to the business, management should be able to estimate the financial consequence of plausible failure scenarios.

AI ROI should include more than productivity upside. It should account for mitigation cost, insurance availability, vendor indemnity limits, contractual recovery, operational dependencies, severity, likelihood, and residual exposure. Otherwise, the company is not measuring AI return. It is measuring AI upside.

AI capital allocation becomes more disciplined when the enterprise can compare expected benefit against reasonably likely exposure.

The winners will not be the companies that slow AI adoption. They will be the companies that stop approving AI investments with upside-only math.