Blog & Insights

AI marketing automation in banking: tools have tripled, impact has slipped. Five shifts compared with campaign automation, and what to build first.

Next best action in banking 2026: why the propensity list is not enough, the three questions a decision keeps apart, and how it is measured.

TD survey: 69% would hand the bank's assistant a task, 38% want alerts to better products. What that means for customer dialogue and lifecycle decisions.

Visa study: 48% of banks invest in AI for efficiency, 30% for customers; Germany's cost focus is 77%. Why Lagarde's warning about sameness fits in.

Part 5/5: Three domain agents, an autonomy matrix, six jobs of orchestration and two runtimes. Why the bounded mandate is the basic unit.

Part 4/5: Personalisation is the execution of a checked decision, the same everywhere. Why contact budget and channel suppression are part of it.

Part 3/5: Predictions are inputs, not results. What a decision record contains, which four decision types exist, and what the language model should not do.

Part 2/5: Three contexts, five functions, four kinds of memory. What turns a data field into a context object, and what the agent should not see.

Part 1/5: CLM supplies the rows, CVM the objective. Why agents without an operating model optimise for clicks, and what belongs in each cell of the matrix.

Golden context, customer loop, check step: why in banks the time from signal to checked action matters, not the tool. Opening article of the series.

Banks hire CVM managers, suites promise personalisation, what they mean is often cross-sell. Why sequence matters once agents make the decisions.

Six banks publish principles for agentic commerce as Amazon blocks third-party shopping agents. What liability, identity and mandates mean for banks.