Sustainable investors should focus on making their investment processes and strategies ‘AI-ready’, says James Phare, CEO of Neural Alpha.
After two years of intense experimentation, it is becoming clear that many Generative AI pilots in sustainable investing are failing. This is not because the technology is immature, but because the architecture and ownership model are wrong. Initial hype and inflated expectations are giving way to a more sober reality: in-house pilots often stall, fail to scale, or struggle to demonstrate durable business value. As a result, confidence, sponsorship and budgets are increasingly under pressure.
This pattern is not unique to sustainable investing, but the sector faces particular challenges. Geopolitical uncertainty, regulatory complexity and relentless margin pressure have created strong incentives for asset managers and their institutional investor clients to automate and reduce cost. At the same time, ESG data, sustainability frameworks and regulatory interpretations are messy, subjective and constantly evolving. This combination has made sustainable investors both highly motivated adopters of AI and frequent victims of over-ambitious or poorly designed pilots.
The problem is rarely the model. Foundation models are improving rapidly and are increasingly capable across language, reasoning and synthesis tasks. The real issue lies in how organisations choose to build, own and govern their AI capabilities. Too many teams are focused on building out commodity infrastructure while under-investing in the one area that actually creates competitive advantage: their investment methodology.
A clear pattern is now emerging among the organisations seeing real value from AI in sustainable investing. They are redesigning their methodologies to be AI-ready, owning the intellectual property that matters, and deliberately licencing the infrastructure that does not.
The jury is still out
AI pilot investment can yield revolutionary results with many successful examples of companies traversing the pilot-to-production chasm across industries. Companies like BMW, Walmart, Wayfair and Colgate-Palmolive have all been able to demonstrate substantial speed benefits, cost savings, enhanced employee and customer engagement amongst other benefits. In finance, innovative fintech leaders are driving the strongest results with companies like Klarna, which has doubled its revenues and cut its workforce by half over the last three years. Most investors now expect portfolio companies to have a credible AI strategy.
Yet the results are mixed according to studies. Perhaps the most pessimistic study is a2025 Massachusetts Institute of Technology report, which found 95% of Generative AI pilots fail to deliver measurable business impact. Similar studies conclude high failure rates of between 70% and 95%.
GenAI pilots fail for a variety of reasons. Lack of a clear business problem is a significant one with broad, ill-defined, even ‘cool’ use cases championed by executives lacking grounding in operational revenue, risk or cost outcomes. Related to this is misunderstanding the core strengths of large language models (LLMs), a key technology powering most GenAI tools, and overestimating model capabilities – leading to unrealistic expectations particularly for advanced reasoning, accuracy or agentic workflows. Scalability challenges with prototypes that work for ten users but fail to scale for 500+ is also a common problem.
AI success stories
LLM adoption in sustainable investing is disrupting many areas traditionally serviced by data vendors and consultants. Reporting, due diligence, investment research, corporate benchmarking and screening are amongst those areas with successful adoption delivering days of work in hours. Indeed AI innovation is even commoditising some traditionally data-intensive areas such as controversy detection and adverse media monitoring with significant business model challenges for incumbent vendors.
ESG data acquisition is a strong use case for AI in sustainable investing. Parsing sustainability reports and filings to extract ESG metrics, targets and policies is a common use case, again much to the detriment of incumbent data vendors with large analyst teams. Corporate sustainability scoring and benchmarking takes this a step further by comparing peer group performance and calculating intensities to distinguish leaders from laggards. Similarly, stewardship and proxy voting workflows are also seeing extensive automation using AI to systematically compare proxy statements to internal policies.
AI leaders in sustainable investing are well beyond data extraction however, now tackling the messier, more subjective and semantically complex topics essential to value creation and capital protection. Mature organisations are comfortable delegating some level of initial decision making to the AI to free up resources for the ‘human-in-the-loop’ to make the full and final decision, knowing that under the hood the AI has strong industry knowledge and understanding of their methodology.
Such firms have adjusted their methodologies to be ‘AI-ready’ with scoring frameworks translated into AI prompts detailing assumptions, rules, scenarios and other information necessary to make investment decisions. They are leveraging carefully curated knowledge bases of the latest regulations, frameworks, standards, fundamental research and corporate disclosures for AI workloads. Retrieval augmented generation (RAG) architectures are also widely used to ensure answers are trustworthy, grounded in real content and hallucination free.
Whether to build or buy an AI platform for sustainable investing is understandably a key decision. Firms that get it wrong can waste money building commodity solutions or worse give away differentiation.
Best in class
It is increasingly clear that holistic data sovereignty and in-house development are not critical dependencies for successful AI adoption. There’s little intellectual property (IP) value in investors maintaining vector stores, embeddings, OCR pipelines and duplicative knowledge bases of disclosures and other content. This is not where competitive advantage resides. Similarly, general modelling, context engineering and fine tuning for the complex semantics, acronyms and language of frameworks, regulations and initiatives is a critical building block for higher level reasoning, but not in itself enough. AI platforms have to go beyond educational benefits to deliver decision-ready analysis to deliver a true return on investment.
The pace of change of foundation models and advancements is also frenetic and poses challenges to the data and technology teams of large, highly regulated financial institutions and corporates where concerns around model training, data privacy and cybersecurity often over-rule business value and competitiveness.
In the era of knowledge work, prompt engineering is fast becoming a must-have skill for the AI-literate workforce. Mature sustainable investing methodologies are increasingly deploying a variety of prompt engineering techniques – role prompting, structured prompting, guided reasoning and many other approaches. Leaders in this space demonstrate the maturity to iterate and constantly adjust prompts to test assumptions and improve results and to have confidence that in working with third-party providers the IP embedded in the methodology is protected technologically and contractually such that if circumstances change it can be ported to other platforms without risk of vendor lock-in or IP leakage.
The key opportunity for investors in the AI age is in fully owning an AI-ready methodology and being able to effectively encode through intuitive tooling of proprietary knowledge / your investment thesis – not the input data and the infrastructure where it resides. It’s about how well you can translate interpretations of regulatory developments, socioeconomic trends and other developments into machine readable inputs, prompts or otherwise. It’s in absolutely maintaining strong data sovereignty for the proprietary datasets, research and investment thesis that makes your investment approach unique, but not wasting time and resources developing foundational infrastructure.
Sustainable investors should focus on making their investment processes and strategies ‘AI-ready’, says James Phare, CEO of Neural Alpha.
After two years of intense experimentation, it is becoming clear that many Generative AI pilots in sustainable investing are failing. This is not because the technology is immature, but because the architecture and ownership model are wrong. Initial hype and inflated expectations are giving way to a more sober reality: in-house pilots often stall, fail to scale, or struggle to demonstrate durable business value. As a result, confidence, sponsorship and budgets are increasingly under pressure.
This pattern is not unique to sustainable investing, but the sector faces particular challenges. Geopolitical uncertainty, regulatory complexity and relentless margin pressure have created strong incentives for asset managers and their institutional investor clients to automate and reduce cost. At the same time, ESG data, sustainability frameworks and regulatory interpretations are messy, subjective and constantly evolving. This combination has made sustainable investors both highly motivated adopters of AI and frequent victims of over-ambitious or poorly designed pilots.
The problem is rarely the model. Foundation models are improving rapidly and are increasingly capable across language, reasoning and synthesis tasks. The real issue lies in how organisations choose to build, own and govern their AI capabilities. Too many teams are focused on building out commodity infrastructure while under-investing in the one area that actually creates competitive advantage: their investment methodology.
A clear pattern is now emerging among the organisations seeing real value from AI in sustainable investing. They are redesigning their methodologies to be AI-ready, owning the intellectual property that matters, and deliberately licencing the infrastructure that does not.
The jury is still out
AI pilot investment can yield revolutionary results with many successful examples of companies traversing the pilot-to-production chasm across industries. Companies like BMW, Walmart, Wayfair and Colgate-Palmolive have all been able to demonstrate substantial speed benefits, cost savings, enhanced employee and customer engagement amongst other benefits. In finance, innovative fintech leaders are driving the strongest results with companies like Klarna, which has doubled its revenues and cut its workforce by half over the last three years. Most investors now expect portfolio companies to have a credible AI strategy.
Yet the results are mixed according to studies. Perhaps the most pessimistic study is a2025 Massachusetts Institute of Technology report, which found 95% of Generative AI pilots fail to deliver measurable business impact. Similar studies conclude high failure rates of between 70% and 95%.
GenAI pilots fail for a variety of reasons. Lack of a clear business problem is a significant one with broad, ill-defined, even ‘cool’ use cases championed by executives lacking grounding in operational revenue, risk or cost outcomes. Related to this is misunderstanding the core strengths of large language models (LLMs), a key technology powering most GenAI tools, and overestimating model capabilities – leading to unrealistic expectations particularly for advanced reasoning, accuracy or agentic workflows. Scalability challenges with prototypes that work for ten users but fail to scale for 500+ is also a common problem.
AI success stories
LLM adoption in sustainable investing is disrupting many areas traditionally serviced by data vendors and consultants. Reporting, due diligence, investment research, corporate benchmarking and screening are amongst those areas with successful adoption delivering days of work in hours. Indeed AI innovation is even commoditising some traditionally data-intensive areas such as controversy detection and adverse media monitoring with significant business model challenges for incumbent vendors.
ESG data acquisition is a strong use case for AI in sustainable investing. Parsing sustainability reports and filings to extract ESG metrics, targets and policies is a common use case, again much to the detriment of incumbent data vendors with large analyst teams. Corporate sustainability scoring and benchmarking takes this a step further by comparing peer group performance and calculating intensities to distinguish leaders from laggards. Similarly, stewardship and proxy voting workflows are also seeing extensive automation using AI to systematically compare proxy statements to internal policies.
AI leaders in sustainable investing are well beyond data extraction however, now tackling the messier, more subjective and semantically complex topics essential to value creation and capital protection. Mature organisations are comfortable delegating some level of initial decision making to the AI to free up resources for the ‘human-in-the-loop’ to make the full and final decision, knowing that under the hood the AI has strong industry knowledge and understanding of their methodology.
Such firms have adjusted their methodologies to be ‘AI-ready’ with scoring frameworks translated into AI prompts detailing assumptions, rules, scenarios and other information necessary to make investment decisions. They are leveraging carefully curated knowledge bases of the latest regulations, frameworks, standards, fundamental research and corporate disclosures for AI workloads. Retrieval augmented generation (RAG) architectures are also widely used to ensure answers are trustworthy, grounded in real content and hallucination free.
Whether to build or buy an AI platform for sustainable investing is understandably a key decision. Firms that get it wrong can waste money building commodity solutions or worse give away differentiation.
Best in class
It is increasingly clear that holistic data sovereignty and in-house development are not critical dependencies for successful AI adoption. There’s little intellectual property (IP) value in investors maintaining vector stores, embeddings, OCR pipelines and duplicative knowledge bases of disclosures and other content. This is not where competitive advantage resides. Similarly, general modelling, context engineering and fine tuning for the complex semantics, acronyms and language of frameworks, regulations and initiatives is a critical building block for higher level reasoning, but not in itself enough. AI platforms have to go beyond educational benefits to deliver decision-ready analysis to deliver a true return on investment.
The pace of change of foundation models and advancements is also frenetic and poses challenges to the data and technology teams of large, highly regulated financial institutions and corporates where concerns around model training, data privacy and cybersecurity often over-rule business value and competitiveness.
In the era of knowledge work, prompt engineering is fast becoming a must-have skill for the AI-literate workforce. Mature sustainable investing methodologies are increasingly deploying a variety of prompt engineering techniques – role prompting, structured prompting, guided reasoning and many other approaches. Leaders in this space demonstrate the maturity to iterate and constantly adjust prompts to test assumptions and improve results and to have confidence that in working with third-party providers the IP embedded in the methodology is protected technologically and contractually such that if circumstances change it can be ported to other platforms without risk of vendor lock-in or IP leakage.
The key opportunity for investors in the AI age is in fully owning an AI-ready methodology and being able to effectively encode through intuitive tooling of proprietary knowledge / your investment thesis – not the input data and the infrastructure where it resides. It’s about how well you can translate interpretations of regulatory developments, socioeconomic trends and other developments into machine readable inputs, prompts or otherwise. It’s in absolutely maintaining strong data sovereignty for the proprietary datasets, research and investment thesis that makes your investment approach unique, but not wasting time and resources developing foundational infrastructure.
Share via:
Recommended for you