Tesco’s latest annual report discusses AI’s role in governance, with a case study in the Audit Committee Report focusing on how the company’s internal and external audit teams are using AI and analytics tools to improve assurance and audit processes.
Where can AI help streamline the reporting process?
When deployed in the right way, AI can definitely make the process of producing reports easier and less burdensome for everyone involved. Some of its benefits include:
- Improved accessibility
Google’s NotebookLM creates audio summaries which some users find easier or more convenient than the process of skim-reading a document.
- Quality checks
Large language models (LLMs) like Copilot can be trained to take on time-consuming but basic tasks, including applying text conventions to drafted copy or correcting spelling and grammar.
- More consistent writing style
LLMs are also great at applying a company’s tone-of-voice guidelines consistently to sections that have been drafted by different authors.
- Basic drafting
AI tools can be used for basic drafting based on existing company documents – like creating a summary of Board activity for the year using the Board schedule and minutes.
- Image creation
Tools like Firefly can be used to create imagery that can’t be found in a company’s image library, either because it’s too conceptual (like the front cover of the most recent WPP annual report) or for operational reasons (a group of senior leaders can’t be photographed together because of geographical location).
What are the watch-outs?
While we need to recognise what AI can do well and where it saves time, it’s also important to know when pens should stay firmly in human hands. Two scenarios come to mind:
1. When assessing what to include and what to leave out
Deciding what’s material to the account of the year’s performance and strategy (and what’s just clutter) isn’t the job of AI. That’s the job of the management team, with guidance and constructive challenge from the Board where needed.
2. When judgement and nuance are required
Annual reports must be a ‘fair, balanced and understandable’ account of the company’s performance during the year, and the narrative and KPIs need to reflect this. Similarly, the ‘going concern’ and viability statements are an exercise in assessing the company’s future prospects. The UK Corporate Governance Code and the Listing Rules make it very clear that these disclosures are the responsibility of the directors.
LLMs are known to display a positive bias when drafting content. Research by the University of California San Diego found that they alter the sentiment of source content in about 22% of cases, often reframing neutral or mixed content in a more positive light. There’s a risk then that AI-drafted content will fail the ‘fair, balanced and understandable’ test, even with the help of human editing.
What guardrails need to be in place?
To keep information confidential until report publication, teams should be mindful about their use of AI tools. Giving an LLM access to unpublished content and asking it to polish a draft can put a company at risk of inadvertently putting market-sensitive content into the public domain. To keep content secure, teams need to use the enterprise versions of Copilot or ChatGPT that come with privacy safeguards.
