AI Investor Cloning could remove the guesswork in IR
Recently Blackstone announced its plans to invest $500 bn in Europe over the next years. Did you see this coming and will some this capital flow into your company?
Imagine if your IR department could predict with some certainty what your investors are going to think, say and do with respect to your company.
In fact, doing just that has become a lot easier and more accurate using the AI toolset to clone investor AI programs. Most articles on how AI can help investors are about detecting general investor trends, analyzing sentiment and thus optimizing communications between IR and investors. While all these areas are valid and very helpful for IR’s effectiveness, my point here is to use an AI model to clone your investors’ AI, thus potentially helping you to know what they’re thinking and might do.
AI investor cloning
Institutional investors, in particular, increasingly rely on AI-driven models and algorithms to analyze companies and to decide whether to buy, hold or sell their equities. This makes their behavior a lot more predictable if only you knew more about how their proprietary models work.
AI can allow you to create a model which will clone your investors’ AI models over time. Machine learning (ML) and its sub segment, large language models (LLMs), can assess huge amounts of data on public investors’ assessments and behavior to detect regularities and ultimately clone each investor’s model.
LLMs are like the brains capable of understanding and generating human language – they don’t ‘think’ or ‘act’ on their own yet. An LLM would generate recommendations which are informed by a data-driven behavioral model, involving investor personas, historical patterns and contextual signals.
The result is that IR departments can think like their investors and anticipate their actions. Importantly, a customized machine learning model must be developed and deployed for each investor institution. Most investors use proprietary analytic, buy/hold/sell programs which drive their decisions on buying, holding or selling your company’s stock. But starting with models mirroring your top five investors will already make this effort worth it.
Most AI programs are designed to identify and predict general and sectoral trends only. Here we suggest to go a level deeper, using ML to think, predict and assess like each of your most important investors and asset managers.
As a result, you can predict buy-side thinking and outcomes, tailoring your IR strategy, messages and even corporate planning accordingly. Your C-level will see the growing precision and impact in gaining control of capital inflows and outflows. It will also give them time to adjust its strategies, aligning them with investor priorities and plans.
How to create your own investor LLM
There is no standardized approach for developing an LLM, but most will include the following components:
- Define your development goals and success metrics: This is the business case for your project. Do you want to predict future investor decisions and actions? How will these insights help? Will it help reduce your stock price volatility, inform IR or corporate strategy, define investor communications? How can you tell if your model is working? How well will it have to perform to be successful?
- Data, the big job: Roughly 80 percent of the work involved in developing an LLM revolves around defining, collecting and structuring data. In this case, data is on past public pronouncements and decisions of an investor’s model. Chances are that the data you have is not granular enough or not properly structured for fitting the needs or facilitating the training of an LLM model. Once you have defined your data needs, you will probably have to start at square one and collect new data. Among the types of data you will need on your investor are:
- Several years of historic data on your investors’ public assessments, opinions as well as actual buying, holding and selling behavior on your company and sector
- Macroeconomic data and trends which may drive investor analyses and behavior
- Articles on your company and sector in influential media
- Financial data and public documents on your own company’s track record, which the ML will need to correlate investor behavior to your company
- Regulatory and legal information.
- Your model: the learning brain: The heart of your model are the algorithms which process huge amounts of text and numerical data, finding regularities and correlations, generating predictive analytics on your investors.
- Deploy your LLM: Once you have everything done, your LLM still needs time to process the data, learn from it and begin generating predictive analytics on your investors. The final product will be an ‘AI investor clone’ on each key investor institution.
Considering the risks
There are numerous risks associated with developing and deploying any type of LLM. Here are just a few:
There are numerous risks associated with developing and deploying any type of ML. Here are just a few:
- Low AI readiness: Surveys claim that only 1 to 13 percent of corporations are ‘AI ready’. This often means they do not know where to start, do not have an implementation plan or the internal willingness for this kind of project.
- Bias: If the data are biased in some way, the results will reflect these biases, thus the importance of having ‘clean’ and objective data.
- Misinformation: LLMs can produce analyses which look plausible but are skewed or wrong.
- Privacy violations: An LLM can find and process internal and confidential information without you necessarily being able to catch it, exposing your company to legal risks.
- Ethical risks: Your model may prompt you to make public pronouncements which are unpopular in society, incurring reputational and financial risks.
Dr William Cox has consulted more than 50 blue chip companies over the course of decades on the financial impacts of corporate process, ESG and technology in companies, including technological upgrades and innovations. He received his PhD from the London School of Economics and other degrees from Oxford, Boston and the Harvard Kennedy School.
