More than three years after the public launch of ChatGPT, Al adoption is surprisingly low. Roughly 84% of the world's population, or 6.8 billion people lave never used GenAl. Of the 16% who have used it, more than 98% use the free version - and often only for simple text or image prompts. Remarkably, only 1.25% use the paid, more capable tools to increase productivity.
This begs the question, is the impact of Al overblown? According to Martin Moeller, Al for Financial Services Leader at Microsoft, "the future is already here, it is just not evenly distributed yet."
"Thinking the impact of Al is overstated is the same as saying 'mobile phones are a just a passing fad' in the 1990s."
This is only the beginning. Importantly, the financial sector is on the leading edge:
Al investment in the financial sector reached $35 billion worldwide in 2023 and is projected to grow to $97 billion by 2027. As a percentage of revenue, that trails only media, entertainment, and sport. Further, while still in its early stages, Al adoption is four times faster than desktop internet.
The investment appears to be paying off, at least for some. For example, JPMorganChase is focused on using AI to increase general productivity and to deliver efficiencies in software engineering and operations. At its Investors Day 2025, the firm announced that investments in technology and AI help its employees work more efficiently and absorb further volume growth. Today, more than 230,000 employees globally use its proprietary generative AI, LLM Suite, saving three to six hours per week.
JPMorganChase accomplished this by fully embracing the opportunity AI presents:
JPMorgan is not alone. Across the finance industry, use cases continue to grow.
"We have immersed ourselves in an Al first mindset across all that we do. And Al is not just a tool, it's re-imagining workflows and it's changing the loading capacities for thousands of people on the frontline and in the back." - Mary Callahan Erdoes, CEO, Asset & Wealth Management, JPMorganChase
According to the Evident AI Outcomes Report, the first half of 2025 saw 173 new use cases from 47 of 50 tracked banks. While roughly 85% of AI use cases focus on efficiency and productivity, banks ⁹ are diversifying. "Reports of revenue uplift – including enhanced sales conversion, cross-selling opportunities and customer acquisition – surged to 16% of all use cases reported."
The question is, how can financial firms develop and deploy AI systems at scale and at speed, in a cost-effective manner?
According to OpenAI, "the fastest way to scale AI impact is to stop solving the same problems in silos. Amplifying progress means turning scattered wins into shared knowledge, documenting successful prompts, workflows and use cases so other teams can reuse, improve and build on them."
While internal team dynamics may speed deployment, a recent Harvard study suggests there is work to do at the human level.
As AI uptake increases, the most prolific users find themselves juggling and multitasking beyond the limits of their cognitive abilities. This can lead to what Harvard researchers call "AI Brain Fry" – saturated attention, mental fatigue and even burnout.
Moeller suggests "The lesson for leaders is clear: They cannot simply sprinkle AI on top of existing business processes. Instead, they need to redesign the work itself as well as their underlying operating model for the age of agents."
Meanwhile, agentic AI is outpacing many business leaders' ability to build and deploy:
Your team is likely working to develop and deploy AI agents across your workflow. According to Deloitte, "embracing agentic AI may no longer be optional for banks." Indeed, leaders in the race to deploy AI agents include Goldman Sachs, PayPal, BNY and JPMorganChase.
However, there are those in leadership roles who know they should fully embrace AI, but either:
If you or your team feels this way, you are not alone.
"We have been hearing about AI agents for three years now since the launch of ChatGPT, and every six months there is a new buzzword. In reality, most banks are still shuffling around Excel files."
Recent developments have surprised even AI insiders. In his viral essay titled Something Big Is Happening, OthersideAI CEO Matt Shumer highlighted the power of OpenAI's recently released GPT-5.3 Codex: "This is OpenAI telling you, right now, that the AI they just released was used to create itself. […] AI is now intelligent enough to meaningfully contribute to its own development."
Shumer highlighted several types of knowledge work he believes will be affected within the next 12 months, including financial analysis. He stated that frontier AI is capable of "building financial models, analyzing data, writing investment memos, [and] generating reports. AI handles these competently and is improving fast."
"The most important challenge we have right now is a question of reproducibility of the results you get with AI: Scenario analysis, sensitivity analysis, forecasting... As a regulated entity, we need to showcase to our regulator that what we do is transparent, reproducible and that there is no black box."
Of course, in highly regulated industries, there is a trade-off in both cost and potential privacy issues that hinders deployment. This is especially true when existing LLM assistants or chatbots use external application programming interface (API) calls.
Then there is the underlying data issue:
In a word, impartiality. It would seem logical that data is impartial, a nugget of truth in ones and zeroes, raw, unfiltered and ready to parse. If only it were so easy.
AI training models only serve to amplify the uncomfortable nature of data. A 2024 MIT study investigating LLM training reward models "found a persistent left-leaning political bias across nearly all these models." The authors admitted limitations: Creating datasets with zero bias is difficult, datasets are an imperfect representation of truth and falsehood and biases may be introduced both from prompts and the LLMs themselves.
Keep in mind that the volume of data is increasing exponentially:
Today's analysts can draw upon unprecedented quantities of data and research to drive their investment insights. But such massive resources risk becoming overwhelming without the tools to access and interrogate them effectively. In 2010 we collectively created, captured and consumed two zettabytes (ZB) of data worldwide. Fast forward to 2025 and that ballooned to an estimated 181ZB. And there is no sign of slowing down: By 2028, IDC expects data generation to skyrocket to 394ZB, fuelled by AI's massive growth.
According to data integration firm Qlik, "technically we're reaching an inflection point where ingestion and transformation of data can be done in real time and hybrid transactional and analytical data can be stored and processed in the same place." This immediacy will affect everything from operational efficiency to risk mitigation and decision making.
Like past tech revolutions, AI is a double-edged sword. Advanced AI capability is significantly increasing cyber threats. At the same time, there is concern that proprietary data could be leaked to train future AI models, effectively rendering it public domain. Tech leaders increasingly believe on-premises infrastructure and data is the best solution.
This lends itself to the way finance firms can develop and deploy AI effectively.
Using lean language models that combine retrieval augmented generation, reasoning trace fine tuning and budget forcing, finance firms can develop highly autonomous agents that "perform remarkably well, especially in settings where consistency, speed, privacy, or cost matter more than state-of-the-art generality. This makes them an ideal candidate for modern agentic AI workflows."
"Key for me is the importance of the leadership mindset banks need to embrace: Agentic AI is not about simply changing what is there but about fundamentally transforming the core of the bank." – Martin Moeller
This will also require greater agility. And that means not only embracing innovative technology and tools but also understanding how they work with legacy systems. Which brings us to one of the biggest challenges financial firms face:
"Although technology has undeniably helped facilitate the provision of financial services through electronic bookkeeping, for example, digital technology has so far failed to deliver financial integration in Europe. In fact, non-interoperable technological ecosystems in each country – shaped by diverging national regulatory regimes – have created siloed pools of asset liquidity, further entrenching fragmentation." — Pero Cipollone, Member, Executive Board of the ECB
Loosely speaking, the above is an example of tech debt. A broad definition that includes:
To address technical debt without the need to replace existing systems, finance firms can use a smart overlay deployment as a near-term solution. This essentially "wraps an AI agent around an existing, well-defined process and the underlying technology." An incremental solution might include building autonomous agents from scratch to replace legacy systems. Finally, a long-term strategic approach to unlock agentic AI would require full process redesign.
A word of caution: Amazon recently convened an all-hands 'deep dive' due to GenAI assisted software deployment gone wrong. Junior and mid-level developers, asked to produce more, faster, without adhering to standard dev protocols, let AI agents loose on the company's legacy code. The outcome: Amazon's website and app would not allow customers to check out or view account information for six hours.
As Digital Transformation continues, there is also a kind of tech debt in research. The shift to digital-first strategies for distributing research isn't always straightforward. That is partly due to outdated conventional document management systems and partly due to the way different generations consume content.
"When it comes to [research] using Chat GPT, asking my very personalised questions about developments in a sector, in an economy and so forth and telling it to use official sources, [that] gets me a long way. So when I think about our users, the question is: Why would they use us if they have ChatGPT, Copilot, Gemini, or Anthropic at their hands?"
You may have experienced the uncomfortable situation where someone from a different generation is speaking and you're not sure what they mean. Depending on the generation, it might even feel as if they're speaking a different language.
There are so many moving parts to what you're writing about in research, but also the toolkit you're using and the platforms, the indicators. AI is part of it. But it's also nerve wracking because you feel like a lot of old structures are disappearing and breaking away. You sometimes feel like you don't really know what ice float to jump onto."
Most certainly you've noticed another kind of generational drift based on technology. Whereas Baby Boomers might prefer a printed version of that report your team just created, Gen Z are digital natives. They grew up with smartphones, don't remember life without the internet and most still believe those 500+ Instagram followers are real friends.
Most importantly for research teams, Gen Zs enjoy snackable content in the form of 30-90 second videos, memes and easy-to-understand infographics. Ask them to read a 50-page PDF... Not so much.
AI will impact the whole workflow and the readership. On the one side is generating the content and pushing it to readers. But also, how should we prepare content [so] that it can be better accessed by readers?"
Then there are the generations between the Boomers and Gen Z. For example, Millennials tend to consume a mix of formats, including podcasts, webinars and detailed reports. They cross-reference research content and often rely on multiple trusted platforms. As such, they also suffer from information overload more than any other generation.
One thing is clear, however: All generations want quality research. Indeed, they have a built-in need.
Digital content including video, social and gaming, make up 76.27% of all internet data traffic. People consume content for a variety of reasons — to sate interests and desires, pursue ambitions and for entertainment. But at the most basic level, we are driven by primal force: Humans are hard-wired to seek information.
"We are so completely enthralled by information that one could, without exaggeration, say we are addicted to it. The addiction develops under the influence of another neuromodulator, this one called dopamine." – John Coates, The Hour Between Dog and Wolf
In fact, dopamine is so addictive that given a choice between self-stimulating with dopamine or eating, many animals would choose to starve.
Humans are also addicted to novelty – even suffering without it. According to Coates, we "are built to ignore the world unless something important happens." Of course, 'something important' is highly subjective. But to witness this phenomenon, one need only turn on a 24-hour news station, with attention-grabbing headlines and an ever-scrolling chyron.
The entire concept of Digital Transformation has made two things clear: There will always be 'new' content and people will consume it.
The 'always on' aspect of Digital Transformation is not lost on financial firms. Constantly shifting markets means new and updated research is essential to serving clients.
However, there is an important argument to be made for too much content leading to overwhelm and information fatigue:
Overall, our competitive and technologically enabled attention economy presents less than ideal conditions for well-reasoned decision-making. Information providers prioritize eye-catching appeal over veracity, diversity, and informativity, and human decision-makers are overwhelmed by information curated with little regard to its orientation to truth.
How, then, can financial firms best distribute research to clients? One option is to utilise existing platforms such as Bloomberg. Yet, if you've devoted time, money and energy into intellectual property, should you pass it on to intermediary platforms for distribution? Not only does this type of relationship often cede rights to your IP, but it can also create an unintended barrier between you and your clients.
Instead, what if your clients could 'pull' your research from a rapid-access portal without having to share your IP with others? This level of personalisation is possible. And it can be done in a way that gets to the underlying and very human, need for novelty.
"We are considering licence-based models [...] more like a data licensing approach. For example, clients might pay for access to a data feed or API and then pay additional fees based on usage or consumption."
The key to competing with the Financial Times print version at one end of the generational spectrum and TikTok at the other, is to create high-quality content in formats that tickle the information and novelty addictions. The good news is, this is one area where AI truly shines.
Imagine creating a single, high-quality research report and then handing it off to an AI with the instruction to make it useful for any medium or platform whilst meeting all compliance standards. Thanks to componentisation and AI driven automation, it is no longer science fiction.
For example, your research team might create an in-depth report intended for your top clients. Using a Create Once, Publish Everywhere (COPE) strategy, the team would re-purpose the report into a series of blog posts or short reports. Next it would create infographics for social media, a video script and a podcast episode. This keeps the core message consistent, while tailoring the delivery to the specific audience. This is the power of COPE. It makes connecting with clients of different generations much easier.
"[AI tools] are quite useful for polishing language [...] and they can also provide preliminary background information. In many ways they function like very eager interns who support more senior people in the early stages of research. I have also found that these tools are quite good at tailoring a narrative depending on the intended audience, whether it is written, visual, or verbal communication."
Even as AI accelerates, it is not yet the "everything solving rocket ship" some would have you believe. But it does make accomplishing these tasks faster. Since timely, accurate information is crucial, this approach can save resources and increase efficiency.
One way this is achieved is through componentisation, a type of granular content management system. Imagine stripping a research report down to modular components, tagged and linked in the cloud or onsite, ready to be called up to create customised reports.
Now imagine the Baby Boomer family patriarch reading an old-school printed report, whilst his Gen X son and Gen Z granddaughter consumes the same information, but in snackable chunks via mobile-friendly video, soundbites and infographics.
Through personalised profiles, your clients of different generations (or learning styles) get the content they need, as they want it. Better yet, they're more likely to consume it and make better decisions. Win-win.
Another use-case for AI-enhanced COPE methodology is regulatory updates across all platforms and touch points.
"One of our challenges is more and more regulations, which absorb time and increase the costs for the bank."
Global finance companies already face the challenge of keeping abreast of different regulators. Using a tech-based approach, e.g., agentic AI, API calls and COPE can streamline the process.
AI adoption, particularly agentic AI, is accelerating in the financial industry. It is a natural progression from machine learning and traditional AI models. However, there remains a need for clean data, a hybrid human-AI approach and a deep understanding of the end user. This is true both for internal processes and for clients, including how research is presented. The ever-increasing glut of data makes content creation easy. Innovative technologies and tools make it fast. But nuanced research still requires human experience, insight and clarity. Once the research is collected and parsed, however, we are living in a new age. More than ever before, it is possible to provide clients with the research they want in a format they love to consume, without compromising quality or running afoul of regulatory compliance.
It is a program designed to give senior executives in investment research the opportunity to step back from day-to-day operations and reflect on strategy. Each event combines a dinner with a structured discussion group led by an expert guest speaker.
The program is now in its third year, having been established to address the need for deeper strategic reflection among professionals managing complex research operations during a period of rapid change.
Discussion topics span a range of forward-looking themes relevant to investment research, including the integration of AI technology and regulatory solutions. Each event is led by a guest speaker with expert knowledge in the field.
A dedicated case study is circulated to all participants ahead of each dinner. This document outlines the key themes to be discussed, allowing attendees to arrive informed and ready to engage in meaningful dialogue.
Events are held in selected European business capitals: Zurich, Copenhagen, London, Stockholm, and Frankfurt. These locations are chosen to provide a focused and professional setting for deeper reflection and forward-looking discussion.
The next event in the series will take place in Frankfurt, Germany.
