KERALA RISING · FROM EXPORTING PEOPLE TO EXPORTING IDEAS 5. How the AI pays for itself The expensive part is building the machine once. The AI that runs it is a rounding error against the budgets it serves, and the machine earns enough to cover its own keep. This chapter is the arithmetic. Most civic-technology proposals wave at savings and hope. This chapter does the opposite: it states the cost of the artificial intelligence explicitly, compares it to the budgets the programs serve, and shows the revenue lines that let the operating institution cover its own running costs without recurring state expenditure. The full multi-year model, with scenarios and break-even, is in Chapter 16. This chapter establishes the single idea that makes the whole enterprise financially credible: verification is almost free, and almost-free verification changes what a state can afford to do. What a verification actually costs Start with the unit. The core operation of every program is the same: take a citizen’s photo or short clip and return a verified fact. The cost of that operation has collapsed. A purpose-trained image classifier. The right tool for the high-volume, routine checks that make up the overwhelming majority of verifications and runs at roughly US$0.0000025 per 87