Generative AI (GenAI) is revolutionising industries by tackling complex cognitive tasks at an unprecedented scale and cost-efficiency.
In financial services, it promises to streamline activities like crafting investment narratives, analysing customer sentiment, and enhancing decision-making – accomplishing them at record speed and near-zero cost.
Yet, despite its transformative potential, many financial institutions find themselves grappling with challenges in implementing revenue-generating GenAI use cases.
Why are so many still stuck in the ideation phase, unable to move beyond proof-of-concept?
A strategic approach to use case selection and orchestration, as proposed by global technology and innovation partner Zühlke, can help move financial institutions (FIs) beyond the proof of concept stage and deliver real value.
At the 2024 invitation-only Insights Forum, Zühlke led a dynamic discussion on AI’s impact in the financial industry and its entire value chain.
The Forum was organised by the Global Finance & Technology Network (GFTN) as part of Singapore Fintech Festival 2024.
Led by Ravi Patel, Zühlke’s Head of Financial Services for Southeast Asia, and Andrea Perl, Regional Data & AI Lead, alongside industry experts from Amazon Web Services (AWS), the session explored how FIs can overcome the PoC phase and successfully evolve from leveraging AI for operational efficiency to unlocking new avenues for revenue generation.
Driving cognitive labour cost to zero

“We frequently witness breakthroughs in foundational model capabilities.
To help navigate this dynamic landscape, Zühlke has developed a GenAI capabilities map that aligns current GenAI abilities with human cognitive traits.”
explained Andrea.
Andrea used the map to highlight GenAI’s readiness to take on more complex cognitive labour tasks, including natural language processing, creative functions like drafting investment narratives, and analysing customer sentiment through social and emotional intelligence.
As GenAI assumes these roles, the associated costs to execute them could approach zero.
Andrea cited a superapp’s experiment where GenAI reduced the time for creating app push notifications from one hundred hours to three, simultaneously increasing conversion rates.
Inspiring success stories aside, building operational use cases requires significant investment. Achieving human-level accuracy in GenAI is feasible, but challenging to achieve.
“About 80% accuracy can be reached after 20% of the project timeline. But progressing to 95%, which is the benchmark for human-level accuracy, needs a lot of engineering.
You also need ongoing input from stakeholders and experts to reach necessary accuracy and keep it up in operations.”
Andrea said.
Additionally, many applications remain task-specific. For instance, a GenAI solution that identifies relevant research papers, cannot simultaneously perform analytics on those papers.
Transitioning from POC to operationalisation
Coining the term “POC-itis” to describe a common challenge faced by many organisations, Zühlke illustrated how proofs of concept (POCs) are often being developed without successfully transitioning into scalable solutions that deliver meaningful value.
This phenomenon highlights a critical hurdle in operationalising innovative technologies like GenAI within financial services.
POC-itis occurs when AI use cases are not strategically chosen or well-orchestrated.
In many cases, the organisation lacks sufficient capabilities in terms of people, technology, processes or data quality to shift beyond the POC stage.
Addressing POC-itis is increasingly critical for financial services businesses, as rapid advancements in AI outpace the capacity of decision-makers to adapt.
A practical approach is recommended to manage the significant investment needed to scale and operationalise GenAI solutions.
“Build use cases that fit your corporate strategy. Build multiple use cases within that field so that you can reuse capabilities and components.
Single use cases often don’t provide a positive return on investment on capabilities needed to operationalise them.”
Andrea advised.
Building GenAI capabilities incrementally and aligning to specific use cases helps balance people, processes, governance, and technology effectively.
Education and subject-matter experts needed
Unlike previous iterations of AI, GenAI requires stakeholders to cope with unstructured, non-deterministic outputs at scale (i.e. GenAI models produce different outputs even when given identical input).
The award-winning AI chatbot that was co-created by Zühlke and UNIQA demonstrated exactly that and was designed to assist UNIQA’s sales staff with tariff and cost coverage inquiries.
“The chatbot can answer a policyholder’s questions about insurance coverage accurately, but verifying correctness requires expert input. In this case, we needed a legal product expert, and about half a day per two-week sprint. You need substantial time from subject matter experts, otherwise you won’t reach human-level accuracy.”
Andrea explained.
Teams across functions also need to develop a deeper understanding of how to handle and optimise the non-deterministic outputs generated by GenAI use cases.
Enhancing organisational maturity through education is crucial for effectively managing GenAI.
Without this foundational knowledge, organisations risk stalling their GenAI initiatives before they can deliver meaningful value.
From efficiency to revenue generation
While many GenAI use cases clearly demonstrate success, especially in achieving efficiency gains, advancing from these applications to revenue-generating ones represents a significant step that not many have taken yet.
It’s significant because building revenue-creating GenAI platforms comes with higher stakes, particularly in regulatory compliance and reputational risk.
A high degree of organisational maturity is needed, especially for customer-facing deployments.
“Human oversight is still necessary for critical use cases close to the customer,”
Andrea explained.
Zühlke advises a phased approach to successfully shift to build the necessary organisational maturity for the shift to revenue-generating GenAI use cases.
Start with efficiency-enhancing internal operations, then move towards customer-facing operations which promise revenue.
“Think of investment story writing for example, initial steps might be to summarise a Chief Investment Officer’s insights, gradually incorporating customer data, and ultimately producing personalised investment stories tailored to individual clients.
That is a real revenue case – because then we have a clear impact on the top-line.”
Andrea explained.
Data: the key differentiator
As models become commoditised and organisations reach higher maturity levels, proprietary data will be the main differentiator from competitors in the near future.
“Ensuring that company data is of high quality and meets security standards today is critical to leveraging GenAI effectively in the future,”
Andrea explained.
It’s just as important to use a responsible AI framework as a basis for use case implementation sustainably.
Your business can proactively meet this opportunity by starting small – identifying the use cases you can ideate and test today.
Reach out to Zühlke to explore how to ideate, create, and scale AI-augmented models, processes, and products that deliver meaningful impact.