Revealing pension members' sustainability preferences

Given the choice, would you invest your pension sustainably? Anecdotally, people under 40 years old and/or with children overwhelmingly agree. But what does this preference actually mean? how could a pension fund discern these preferences? what should asset managers do to meet this demand?

To answer these questions, Mettle Capital, Wyser and Brunel University London, supported by Innovate UK and ESRC: Economic and Social Research Council, have launched the 'Revealing pension members' sustainability preferences using conversational AI' project in May 2024.

Aligning preferences with practices

The primary research goal is to develop a comprehensive, data-driven system to improve awareness of Environmental, Social, and Governance (ESG) preferences among pension fund members. The project aims to empower pension managers with advanced tools and insights to align investment strategies with the unique ESG priorities of their clients, thereby fostering responsible and sustainable investment practices.

Investor preferences, including ESG priorities, should drive product selection or pension manager's recommendations. Quantifying preferences becomes even harder when attempted at scale - such as a pension fund might need. This research will seek to elicit members' preferences and understand how that interacts with the financial returns they want to achieve.

Off the back of this industrial research we aim to build prototype solutions that will allow members to easily express their ESG preferences for investment opportunities. Our solution will need to work at scale, be engaging and ensure it is not arduous for the beneficiary.

We will do this by prompting the beneficiary through a combination of open and closed questions and then mapping the answers to known metrics. Although we will start with the simplest forms (surveys) given existing research we expect open-ended methods (e.g. conversational AI) to be more effective at scale as they will be like a natural conversation. This research project aims to influence ESG awareness systems that empowers pension managers to make informed investment decisions aligned with the unique ESG preferences of their members while balancing these preferences against their financial goals. The project intends to contribute to the broader goal of fostering sustainable and responsible investment practices in the financial industry.

What is a preference?

Pension managers need enhanced understanding of beneficiary ESG preferences to align investment decisions with members' interests. While some pension providers offer individual-level investment options, their lack of understanding of pension members' interests limits their effectiveness. In the 2023 Autumn Statement, the Chancellor added complexity to the issue by announcing employees' right to request employer contributions to selected funds.

Beneficiary ESG data challenges include:

1. Unknown preferences: User engagement poses a critical challenge due to issues with traditional surveys. Problems include low response rates (1.5% in the NEST Insight experiment), selective responses (structured surveys limit free expression), and data bias (self-selection = response bias). While protocol-compliant surveys can yield robust results, their limited effectiveness is evident in consistently low response rates.
2. Linkage of preferences: Members' preferences must be transformed into actionable metrics by mapping to existing detailed ESG criteria for investment managers/trustees to use when selecting investment opportunities. Such integration has not been undertaken to date, it is crucial for ensuring effective implementation of insights.
3. Values versus value: Gathering preferences must be done conscientiously, ensuring ESG preferences are balanced against desired financial returns. Without performance trade-offs, most would probably choose an ESG investment strategy over non-ESG. However, there is a suggested possibility, (Pástor, 2022), that ESG investments may yield lower long-term returns. This factor should be considered when shaping preferences.

Some efforts exist to collect such data (in the Netherlands), but widespread and sustained activity is lacking (Bauer and Smeets 2003). A UK study on University Superannuation Scheme participants (2% response rate) found that over two-thirds favoured ESG-positive investment, a result mirrored in a financial incentive-based investment "game." However, the initiative employed ESG broadly, overlooking specific/nuanced preferences e.g. plastic-free, combating modern slavery, etc.

How to reveal pension members' preferences

A blend of conversational AI and traditional NLP techniques will be used to overcome project challenges and will increase participation because it will feel like a natural conversation to pension members.

Our solution centres on developing AI models for automating data creation (ESG preferences). The work will build on Mettle's ESG taxonomy and their established linkage to conversational data. There will be two models:

• conversational model: used to dynamically generate questions that will allow a person's preferences to be understood, the model will optimally traverse the ESG taxonomy to keep conversations short and on topic
• mapping model: will map these unstructured responses to structured ESG metrics than can be easily aggregated and compared to investment products.

The conversational model can be used by many different interfaces. Our research will focus on two: a conversational agent/chatbot (lowest cost); and a prompter for an unskilled interviewer (highest engagement).

This research will extend the research and pilots conducted by Professor Rob Bauer (Bauer et al, 2023), his work reflects the most advanced work in this sector. Current commercial offerings are highly focused on self-selecting discrete responses and do not obviously use AI. These solutions often exhibit low response rates, addressing only part of the unknown preferences problem without assisting with the linkage preferences or values versus value issues.

This research will provide working demonstrations of a system that interprets and transforms ESG preferences into structured metrics. The principal outputs are: NLP model; mapping model; and assurance process.

The goal is to tilt UK pension pots towards more sustainable investments and so meet the UK's 2050 #NetZero target. We look forward to working with DC master trusts to build an industry standard for revealing members' preferences.

David Meliveo, Chris Eastwood, Duncan Whitfield OBE, Peter Brackett, Huw Davies, Graeme Griffiths, Jenny Swift, Louise Williamson, Andrew Warwick-Thompson LLB EPMI PTPMI, Daniel Stamford, Robert Branagh, Richard Moody, Richard Sheppard FPFS TEP, Howard Rickard, Tom Higham, Gavin Perera-Betts, Victoria Sant, Gary Smith, Sharon Bellingham AMAPPT, Roz Watson, Ruston Smith MBA, FPMI, FCMI, FRSA, Sandra Carlisle, Mark Pearce , Anthony Dixon , Rufus Grantham

The research consortium consists of:

• Mettle specialises in deriving ESG metrics from textual data, leveraging open-source text repositories (social/trade/traditional media) to form fine-grained ESG metrics for thousands of companies. Mettle creates a pipeline to map text data to ESG metrics
• Wyser brings conversational AI expertise having previously built systems that dynamically question individuals to facilitate context-dependent queries related to employment law (models developed based on audio recordings). The processes/technology employed can be adapted to extract ESG preferences.
• Brunel University London, in particular Federico Castagna will lead author publication in peer-reviewed academic journals around the explanability of the AI involved in the project.