John-Paul Edwards, CIO at Broadstone – a leading independent pensions, insurance, employee benefits and investment consultancy
Amid the hype and anticipation surrounding the use of Artificial Intelligence (AI), Broadstone is taking a realistic, pragmatic approach to get the most effective use out of this exciting technology in support of our ambitious growth objectives.
The first point to make about how we, as a business, are implementing AI is that it requires co-ordinated change across people, processes, and technology rather than an ad hoc approach.
By embedding these processes across Broadstone we are able to democratise access to generative AI, automating tasks to boost productivity, support cost efficiency, and create opportunities across our services. In this way, AI is augmenting our competitiveness and work and potentially giving us an edge in the market.
But how exactly are we using AI? Broadstone has predominantly identified three streams to engrain in our operations.
1. Consuming prebuilt models/tools for task productivity outcomes
Broadstone uses prebuilt AI-as-a-service tools for creating videos, marketing, content generation, summarisation, and semantic search. These tools provide ‘easy wins,’ making information accessible to business users via natural language conversational interfaces. For example, GitHub Copilot has increased our software development velocity by 15%-20%, and financial modelling in our Insurance, Regulatory and Risk (IRR) division have also leveraged Copilots to deliver time savings of 15%-25% in Python coding.
2. Fine-tuning with accurate, relevant reference data for business-specific outcomes
As such, AI is allowing Broadstone to reshape how we operate many of our business processes, particularly as we look to target enhancing our diverse services. We’ve been piloting and developing solutions that fine-tune models and use more accurate, relevant reference data. One such use case is the deployment of a solution that uses AI to transform unstructured data, extract values, run calculations, and create a monthly dashboard. This thereby removes an estimated 85-90% of the manual data entry in the analysis of claims data and premium forecasts for corporate healthcare client reports.
3. Custom development for innovation using accurate representative data for training
Custom development typically requires accurate representative data for training and more investment spent in Machine Learning (ML) skills, infrastructure, and security. However, Broadstone is currently piloting generative AI, cognitive search, and ML within our IRR division to query data safely and accurately, ensuring appropriate training for specific business outcomes.
This use of AI aims to enhance our underwriting and credit risk services by enhancing analytical model building. Systems to learn from data, identify patterns, and augment consultant expertise. These advancements not only support our consultants but also position Broadstone to capitalise on operational efficiencies and improve our service offerings.
Other examples include our customisation of a Large Language Model (LLM) for IT support queries. This has the capability to answer core issues, add context to incident tickets and improve response and resolution times. In practice, an impressive 59% of our first-line support tickets were successfully closed by the AI.
Broadstone has also developed its own app using Azure AI cognitive services to extract key points from large PDF documents and reports to generate summaries. The tool uses a secure LLM to extract pension accounting assumptions used by the actuarial team and includes a separate feature to create minutes and write up actions from Teams transcripts.
All these processes save time, save cost, and improve our productivity as a consultancy.
Bringing employees with us. Alongside our commercial use of AI, we recognise the importance of winning hearts and minds amid some scare stories.
So as part of our use of AI, Broadstone prioritises employee training and knowledge to use generative AI tools safely and confidently. We have established ethical principles that guide our application of AI including strong leadership commitment, open communication, comprehensive ethical guidelines and policies.
We have also established an AI Council to monitor compliance and promote responsible and trustworthy AI practices. By aligning AI with our business objectives, testing carefully and putting in place thorough security protocols, we can make AI realistic, safe, and viable, maximising its potential to achieve our growth ambitions while not wasting valuable resources.
