Neo Engage

Most banking personalisation is theatre. A customer's name in an email subject line. A product recommendation generated by a rule someone wrote three years ago. The appearance of relevance without the substance of it.

Neo Engage was designed to solve the actual problem.

The insight that drove the product came from a specific client conversation. A business banking customer expressed frustration that their bank held years of their transaction data and still communicated with them as if they were a stranger. That observation became the seed of Neo Engage: a platform that would turn the data banks already hold into genuinely intelligent, timely, and hyper-personalised engagement at scale.

The Architecture

What makes Neo Engage technically distinctive is not the personalisation itself, every platform claims personalisation, but the way the system handles data privacy while delivering it.

The data model was designed and architected by Erik, and it reflects a deliberate set of choices about how AI should interact with sensitive financial data.

Data enters through an Ingestion API that consolidates bank transaction data, client-specific information, and market intelligence gathered by an autonomous agent monitoring the web for company, industry, and competitor signals. Before any of that data touches the AI layer, it passes through a pseudonymisation step that removes personal information entirely. The AI never sees who the data belongs to.

Inside the AI layer, a Prompt Augmentation module instructs a fine-tuned private RAG model, contextualised for the specific bank's market, language, and operating environment. The output from that model does not go directly to delivery. It passes through a second, separate verification LLM that interrogates the first model's output, evaluates it for quality and appropriateness, and accepts or rejects it. This dual-model verification approach was a core architectural decision, adding a layer of quality control that most AI platforms skip entirely.

Only after verification does the system add personal information back through a re-personalisation step, before the output reaches the Delivery API and is distributed across whichever channels the bank uses: web, mobile, relationship manager, or automated communication.

The system also learns. A re-learning loop feeds verification outcomes back into the prompt augmentation layer, continuously refining the model's output based on what has and has not worked. Over time, the platform becomes more accurate and more relevant without requiring manual intervention.

The Result

The architecture delivered what it was designed to deliver: a platform that handles sensitive financial data with genuine privacy rigour, generates personalised communication that reflects what a business customer is actually doing, and improves automatically over time.

For banks, the commercial case is straightforward. Business banking customers who receive relevant, timely communication engage more, churn less, and are significantly more receptive to cross-selling. For business customers, the experience shifts from generic to genuinely useful, from a bank that sends them things to a bank that understands them.

Neo Engage was built to integrate with existing CRM, banking, and Open Banking systems, making deployment fast and the disruption to existing infrastructure minimal.