An Asset Managers guide to deploying Generative AI

Generative AI can help asset managers extract value and better insights from data. Here are some pointers to adopting the right approach to adopting the right AI model.

The challenge: Consuming qualitative data

While quantitative analysis has reached new heights of automation and accessibility, synthesizing qualitative data remains a major challenge. Asset managers must keep pace with a firehose of analyst reports, news articles, expert interviews, and to form a complete view of their portfolio.

Manually consuming this qualitative data is cumbersome and time-consuming. Important signals are easily missed amidst the noise, leading to blind spots around portfolio companies. With only so many hours in the day, asset managers struggle to effectively process the breadth of qualitative inputs needed for fully informed investment decisions.

The solution: Applying Large Language Models

Large Language Models (LLMs) like Claude 2 and GPT-4 can rapidly read, comprehend, and summarise large volumes of text. This allows asset managers to efficiently synthesise relevant insights from the growing pool of qualitative data.

LLMs can be tailored to highlight key information in analyst reports, transcripts, news and more. By exponentially increasing the breadth of qualitative data processed, LLMs free up analyst time previously spent on manual review. This enables a 360-degree perspective on portfolio companies and more data-driven investment strategies.

With thoughtful application of LLMs, asset managers can realise new value from qualitative data. Time savings allow analysts to pursue more strategic initiatives, while comprehensive insights help avoid risks and identify opportunities unseen by competitors. As LLMs continue to advance, their role in asset management data workflows hold transformative potential.

Getting started with Generative AI

For asset managers, the usual strategy of waiting to see which provider ends up with the best solution and purchasing it may not be applicable. Each organisation’s AI strategy will be different, and there are several considerations that should be ironed out as a first step:

1. Do you wait for new technology or get started now to gain first mover advantage? The technology is developing rapidly so there will be a more advanced solution down the road, but how will you know when a solution is good enough, and will you have developed the necessary skills to extract value by then?

2. Do you let others work out the kinks or would you want to influence the development so that the solutions are more closely aligned with your business goals?

3. Do you prefer to learn in theory or learn by doing? Waiting buys you time to organise training and development days, while starting now allows you to directly develop the skills and comfort working with LLMs necessary to achieve your AI goals.

In either case, there are things you should be doing today to set your strategy up for success.

Begin by educating teams as to the capabilities of large language models

Many experienced the massive hype surrounding the launch of ChatGPT-4 in early 2023, but recent usage statistics have shown waning interest, suggesting most people have not found a compelling use case for daily activities.

Building a foundational level education across the organisation will provide a shared language and understanding of its importance.

Develop clear needs and use cases

Make time to identify not just where faster processing of qualitative data can add value, but exactly how you would envision it doing so.

Having clear use cases help to focus effort on high-impact applications that align to business needs rather than getting distracted by “cool” but lower value capabilities.

Start building your responsible governance framework

• Development in the AI space has far outpaced governments abilities to regulate them, but make no mistake, regulation is coming. Developing your own responsible AI governance framework has several advantages:

• Ease of maintaining legal and regulatory compliance - Updating solutions to conform to evolving regulation is far easier when starting from a somewhat developed baseline than from nothing.

• Uphold ethical AI principles - Frameworks based on ethical guidelines promote fairness, accountability, transparency that build public trust. This protects an organisation's reputation.

• Support data privacy rights - With appropriate governance, LLMs can be deployed in ways that respect individual privacy and data protection laws.

• Manage business risks - Governance provides oversight to assess risks across operations, finances, customers, staff from AI systems and take mitigating actions.

Prepare your data

Clean, structured data leads to higher quality model outputs. Asset managers spend heavily on curating the best third party data sources. The same effort needs to be applied to first party data to ensure that it is rights-cleared and suitable for processing. Restricted data likewise needs to be segmented to prevent leaking sensitive information. Overall, upfront effort to organise unstructured qualitative data pays off in better controllability, efficiency and outcomes when training large language models. It converts messy data into a strategic asset

Start testing

Now that you have laid the foundation of your strategy it is time for the fun part. Thoughtful testing is key to ensuring to long-term benefits of your AI strategy. Pick a use case and approach initial projects as validation exercises. As the landscape is shifting so rapidly, special consideration should be given to maximising flexibility and control over your AI assets – you shouldn’t be locked into any component of your tech stack.

Creating a validation framework which defines the desired inputs and outputs of the solution will enable you to fairly evaluate its capability, as well as the effects of further development.

How we help companies implement AI strategies

With our deep experience building robust data pipelines and productionising AI, we at QuantSpark are ideally positioned to help asset managers implement generative AI to maximise value. Our team understands the data challenges and use cases unique to asset management, and can design tailored data infrastructure to feed reliable, high-quality data into generative models. We view building maintainable pipelines as a prerequisite for AI success. Our modular development methodology also allows seamless integration of new AI technologies as they are released.

By leveraging our knowledge of asset management, data engineering, and hands-on AI implementation, we can partner with asset managers to turn cutting-edge generative AI into an operational reality that delivers true competitive advantage. Our end-to-end approach brings together the right components so your team can quickly realise benefits from AI-generated insights.