20% of support queries handled by AI.
£300K saved every month.
An AI assistant built into a fintech platform that masks personal and financial data before it reaches the model, answers routine banking queries, and grows its adoption entirely by client choice.
avoided hiring costs
voluntarily by AI
within 12 months
company average
Client
A fast-growing fintech platform for entrepreneurs, providing banking, accounting and business management tools to business clients.
Goal
Build an AI assistant as a daily business companion — handling support queries, executing banking operations and delivering personalised financial expertise — to scale client service capacity without growing headcount, while holding NPS at the level the platform was known for.
The context that made answers good was the context banking couldn’t share.
Support load had to come down without pushing a single client toward a channel they hadn’t chosen. Clients in this sector default hard to human agents, and forcing them into automation risked the NPS the platform had built its reputation on.
The technical obstacle ran deeper than reluctance. Large language models answer best with full conversational context, but in banking that context is protected personal and financial data that cannot be passed to an external model under financial services regulation. The internal knowledge base was also fragmented and had never been built for AI retrieval, so accurate answers at scale meant a structural rebuild, not a patch.
Three constraints were non-negotiable:
- Accuracy and client trust had to be proven before committing to full-scale development
- No personal or financial data could reach the LLM, even though full conversational context needed it
- Adoption had to be entirely voluntary — clients could never be routed away from human support
Mask the data, keep the context.
A functioning prototype answering questions across a single banking topic area went live in one week, at 87% accuracy. Two engineers then integrated it into the live product interface to produce a proof of concept — confirming both technical viability and that clients would actually engage with AI-assisted support.
In banking, you don’t force clients onto a bot — you earn the right to be chosen, one voluntary query at a time.
A proprietary data-masking layer strips more than 20 categories of personal and financial information — including detail clients type mid-conversation — before anything reaches the LLM, preserving full conversational context without transmitting or storing protected data. A three-stage automated quality check assesses every response for tone, relevance and correct use of context. Behind it sits a purpose-built, machine-readable knowledge system with structured fragments optimised for LLM retrieval, governed so that any product or process change is reflected within 24 hours.
The assistant was never positioned as a replacement for human support — it was offered as a voluntary alternative, surfaced through contextual entry points tied to specific moments in a client’s journey. Within 12 months, 20% of all support queries were being handled voluntarily by AI, and the assistant’s own NPS reached 8.7 against a company average of 7.5.
A fifth of the workload, gone voluntarily.
Twelve months after launch, the platform is handling a fifth of its support volume without adding headcount, saving £300K a month while its AI channel outscores the company average on NPS. The approach also won the International AI Championship in Abu Dhabi in 2025 — external recognition for a system built on strict data masking, a machine-readable knowledge layer, and adoption clients chose for themselves.
* Case studies reflect work undertaken by our Heads of AI either during their tenure with Head of AI or in prior roles before they were part of the Head of AI network; they are provided for illustrative purposes only and are based on conversations with our Heads of AI.
More case studies.
Your biggest pain point.
Fixed in 14 days. 50% off.
This started with one conversation. Book a 30 minute brainstorm call — we’ll plan your first AI project together and issue your 50% discount code. No payment today.
*Case studies reflect work undertaken by our Heads of AI either during their tenure with Head of AI or in prior roles before they were part of the Head of AI network; they are provided for illustrative purposes only and are based on conversations with our Heads of AI.