Modernizing Your Core With AI? The Metric to Own Is Cost Per Decision

Key Takeaways from This Blog:

• AI costs will rise as adoption succeeds: Even as the cost per AI-powered decision falls, banks will spend more overall because cheaper intelligence gets applied to more decisions across the institution.
• “Cost per decision” should become a core banking metric: Banks need clear ownership and measurement of what each completed AI-assisted decision costs and the value it produces, or successful programs may look like uncontrolled spending.
• AI pricing could become the next form of vendor lock-in: As traditional core switching costs decline, banks should control their intelligence layer and closely scrutinize providers’ AI pricing models to avoid rebuilding dependency through usage-based fees.

The next wave of core banking modernization won't come from replacing your ledger. It will come from the intelligence layer you build on top of it, the layer that reads your core's data and turns it into decisions. Your core is becoming a commodity, and over time it should cost you less. But the layer above it follows a rule most banks haven't planned for: the cheaper it gets, the more you're going to spend on it.

It's counterintuitive, and it's the trap I'd watch most closely as you modernize. The institutions that execute best use the most intelligence, which makes them the most exposed to what it costs. Yet almost none can tell you what a single decision costs to produce.

We've seen this before. Storage got cheaper, so companies stored more and spent more on it than ever. Online banking made each interaction far cheaper than a branch visit, but customers didn't bank less. They banked more. Lower unit prices didn't shrink the bill. They raised it.

Intelligence behaves the same way. Every time the cost of running a model drops, another class of banking decisions becomes worth automating. As that cost falls, you won't run today's work for less. You'll point intelligence at more of it: fraud, pricing, lending, and thousands of smaller decisions that never justified the effort before. Those decisions also grow more sophisticated. One might pull data from the core, check policy, and draft a recommendation before anyone looks at it. Each step gets cheaper. Better decisions use more of them.

So two things become true at once. Your cost per decision falls while your total intelligence spend climbs. The technology doesn't stay expensive. It becomes cheap enough to use everywhere, and everywhere is what you pay for.

The better you use AI, the more it costs you

Most technology risks in banking punish the banks that hesitate. This one punishes the banks that succeed. Picture an institution that modernizes well. Strong contracts, sound architecture, an AI-powered fraud capability that delivers. Word travels the way it does inside a bank when something works. Lending wants it, then deposits, then operations, and eighteen months later one initiative runs across the institution. Then budget season arrives, the intelligence line has grown faster than forecast, and the board asks a fair question: what did we get for the extra money?

Most won't be able to answer it. Not because the technology failed, but because nobody built the discipline to measure what those decisions cost, or what they returned. Without that, healthy adoption and runaway spending look identical. Boards don't keep funding costs they can't explain, so good programs get cut because the governance failed, not the capability.

Which brings me to a number community banks are about to start managing. For decades we've run our institutions on the efficiency ratio, cost of funds, and cost per loan. The next one is cost per decision. And a decision isn't a single model call. It may run through several models and policy checks before it resolves, so govern the completed decision, because that's where the value is created.

Not every decision deserves the same level of intelligence. You don't put your most senior commercial lender on every consumer loan, and you shouldn't run a routine, low-risk process on your most expensive models every time. The institutions that handle this well won't avoid usage-based pricing. They'll govern it, matching the cost of intelligence to the value of the decision.

But governing anything requires someone to own it. This needs a name, one person accountable for what your intelligence costs, where it's consumed, and whether the value justifies the spend. Reviewing invoices is accounting. Managing consumption is strategy. The institution that builds this discipline walks into that same budget meeting with a real answer: here is what we spent improving fraud decisions, and here is what happened to fraud losses.

As core switching costs fade, providers rebuild lock-in through AI pricing

There's a second shift underway. For years, a provider's advantage lived in switching costs, and leaving was painful enough that most institutions stayed. That friction is easing, and the revenue it protected won't be given up quietly. As the cost of leaving falls, providers will rebuild that lock-in in what they charge for intelligence. So watch how they price it. When provider revenue starts flowing through AI subscriptions and per-decision fees, you're watching the moat move. Control your intelligence layer and you can compare models and adapt as the market shifts. Run every decision through one provider's meter and you've rebuilt the dependence you spent years trying to escape, only now it's variable.

Cost, ownership, and the provider's pricing all point back to the same place. Three things belong in your next board conversation. First, treat a rising AI bill as ambiguous. It can mean adoption is working or spending is running away, and the number alone can't tell you which. Only cost per decision can, so don't let a big number end the conversation before you know what it bought. Second, decide who owns cost per decision before your first production model goes live, not after the first surprising invoice. Third, watch your providers' pricing as closely as their products. Products change every quarter. Pricing models change rarely, so when one moves, that's their real strategy showing itself.

None of this is a reason to wait. The banks that lead over the next decade will be the ones building now. It's a reason to decide, before you turn the meter on, who owns cost per decision. Because once intelligence is cheap enough to use everywhere, the hard part isn't using it. It's knowing which decisions are worth what they cost.