By Tamara O’Brien, TMIL’s roving reporter
So, we come to the final webinar in our series on gen AI in reporting. Having examined gen AI’s crystal maze from many angles, it was fitting to end the series with the experiences of people who are Actually Doing It, and doing it well, namely Alia Fazal, Head of Corporate Governance at BP, and Louise Riordan, Head of Corporate Communications at Informa, ably moderated by FW’s tech partner, Diana Rose.
Before we heard from the doers, FW founder Claire Bodanis did some disclosing of her own, revealing that every year she re-reads The Lord of the Rings. And as she embarked on this year’s Middle-Earth fest, a nagging question she’d had about why gen AI ‘doesn't seem to be all that good at doing the commentary part’ of a report was suddenly answered. Because in his preface, Tolkien spoke about the importance in writing of the author’s ‘lived experience’.
Claire continued (and I paraphrase only slightly): ‘It struck me that this lived experience is at the heart of the role of the human being in corporate reporting, and why generative AI can only ever be a precocious intern. Commentary sections must express the genuine views of management; and to have meaning, they must be informed by lived experience. Generative AI tools have no lived experience, no sense of truth, no reality against which to test their information. They generate their reporting story from selected documents published in the past, and cannot talk to the people involved to determine what that information may mean for the future, in any discursive way. But this future aspect is, of course, an essential aspect of reporting.’
If that isn’t a great apologia for human control of the reporting narrative (and a reason why we humans must keep reading books), show me a better one. But it’s not just Claire’s opinion. Her many conversations with reporting professionals from the likes of AstraZeneca and (AI gurus) Prosus Group revealed that they too found gen AI ‘pretty hopeless at writing anything strategic’.
She then went through some updated research stats, that indicate the use of gen AI in corporate reporting has not changed dramatically since 2024. Good news that companies are treating the technology with appropriate caution? Well, yes – though anecdotally, she’s picking up that smaller, less well-resourced organisations are using external chatbots, irrespective of the risks.
Time to hear from those who really are living the experience of reporting with gen AI.
The proof is in the process
Aware that she was speaking as a representative of one of those well-resourced corporates, Alia opened with a clear statement of how gen AI is viewed at BP. It’s a powerful support tool that helps with the drafting process, synthesis of data and analysis – but it’s not a decision-maker, nor does it replace accountability. Responsibility for content and disclosures in the annual report, no matter how they’ve been created, remains firmly with management and the board.
Where BP is seeing real value in gen AI is at a practical level – in improving, streamlining and accelerating the first draft of narrative disclosures.
It’s particularly good, for example, at gathering complex information from multiple sources and turning it into clear, accessible language; giving a good basis of raw material for humans to assess, enrich and generally shape into meaningful communication. All while being assured that tone and terminology is consistent not just across BP’s (very long) annual report, but other reports they publish each year.
‘Gen AI also identifies trends, flags variances, and even suggests alternative phrases where disclosures are sensitive or nuanced,’ continued Alia. ‘We've seen some tangible benefits from a governance reporting perspective too. For example, we've used AI to help identify source data and verify statements, which used to take hours of manual trawling through minutes and documents.’
And when it comes to the gruelling final compliance checks to ensure that all corporate governance code or listing requirements have been covered, AI can pinpoint where those references sit in the report. A huge saving of time and brain-ache for those doing the checking.
Alia cited many more such benefits. Gen AI’s ability to summarise the considerable amount of information about BP’s global workforce engagement activities not only saves time and effort but helps the drafting team identify key themes and learnings. ‘Which we review, of course, to understand that the information is actually representative and not a hallucination,’ she added. ‘But it has really accelerated the process and given us a good starting point.’ Gen AI has also transformed how quickly they can bring new team members up to speed.
All that wonderful stuff said – Alia emphasised that there are clear boundaries that need to be respected. ‘Gen AI should never be relied upon for anything requiring legal or regulatory interpretation without the expert lens of an appropriately qualified member of the team. It's certainly not appropriate for finalising market-sensitive or forward-looking statements. And nothing’s changed regarding accountability: directors’ duties remain the same. Gen AI is just there to support the process.’
For Alia, the secret to this success is BP’s clear guidance on what is and isn't permitted; for example in the use of secure and enterprise-approved tools, and in protecting confidential data. ‘There's also an increasing focus on documenting where gen AI has been used, not just for external disclosure but for internal use as well, so we have clear links back to the source data. Overall, I think having strong governance frameworks allows you to unlock the benefits of gen AI without introducing new risks.’
Making it personal, interactive and fun
Aware, for her part, that Informa was less of a household name than BP – Louise began by explaining what the company does. Which is, connect people with knowledge. As well as being one of the world’s leading events organisers, behind shows such as London Tech Week, Informa is also a major academic publisher, with brands including Taylor & Francis and Routledge. In the spirit of connecting people with knowledge, they’ve always seen their annual report as a prime communications opportunity; a chance to inform and engage with audiences from investors to job applicants, partners and potential investees.
So when it came to gen AI, Informa’s focus was on the user experience. How could it help readers get to know them and their business better? Bring gen AI and the AR together, of course! So two years ago, they embedded an app of Informa’s home-grown AI assistant Elysia into their AR microsite. The result is an AR hub chock-full of extra material, videos, interviews, imagery, quizzes – all from trusted sources, and enabling users to dig deep into whatever topics interest them.
‘What happens is, you land on the site and the Elysia app acts as your conversation and research partner along the way,’ Louise explained. ‘So you might go to the section which talks about one of our brands, WHX, a big healthcare brand, and see how that brand has grown, because that's a material part of our growth as a company.
‘Same with financial data. Say you go to our KPI section, which goes back three years, but you want to go back further. So while you’re reading about it, you can ask the Elysia app on the side to give you a five- or 10-year trend line. It’s like a second screen that brings the AR to life.’
Knowing the value of good listicle, Louise shared five things Informa has learned from its experience:
Doing this well takes significant time, thought and testing. We had a separate squad working on the microsite app, alongside the core reporting team. We trained the AI only on trusted company material – some 20 years of our annual report, sustainability report, financial statements, product database and some product websites. So essentially it searches a walled garden. Even so, it required extensive guardrails, testing and careful tuning of tone.
It’s not for the faint-hearted. We all know that nuance and context are everything in reporting, and LLMs do not give identical answers each time. So we include disclaimers, tell users to check source material, and set clear limits to reduce hallucinations and prevent actions we wouldn’t take as reporting professionals. For example, the tool will not predict future share prices or compare us to other companies.
It requires leadership buy-in. We could not have created this resource without our leadership team’s strong support for both reporting and AI. We also had an existing platform in Elysia, and great collaboration from our technology and AI teams, who seized the opportunity to develop it further.
Gen AI can’t replace everything. Investors and analysts want facts and information but also feeling, and as Claire said, lived experience. They won’t rely on a chatbot to make an investment decision. That reflects our wider experience of gen AI in business: it can support processes up to a point, but when decisions matter, there will always be a human involved. People will still want to consult source material, speak to someone directly, experience a product or business firsthand, or read the chair’s original letter.
Gen AI is an ongoing commitment. The technology itself keeps changing, and we’ll always be refining the microsite app and finding ways to increase personalisation. I’m sure we’ll be making even more use of gen AI in next year’s reporting process too.
Top tips from wise adopters
What advice would our panel give to someone wanting to improve their reporting with this technology?
Alia: Be clear about the guardrails and guidance for using gen AI. And when drafting content, consider how you’ll maintain an audit trail. A good framework will achieve both aims, helping people to use gen AI properly and record how and why they used it. I’d also say, start with a lower-risk area – a data-heavy section, or a non-financial committee report – rather than something high profile like the Chair’s statement.
Claire: I’d say start with Your Precocious Intern! Alia’s point is exactly right: before using gen AI, you need to think through how you’ll use it, along with the risks and benefits. If you put sensitive material into it without thinking, you’ll run into trouble. Our research offers a framework to help people get started. You don’t have to agree with all of it, but even the summary gives practical suggestions on when to use AI, when not to, and what process to follow.
Louise: I agree with Alia and Claire: start by asking what you are trying to achieve rather than jumping in. A good place to begin is with pain points in the process, and small experiments where AI could help. Reporting is a high-pressure period, so plan ahead, decide what you will and won’t use AI for, and make sure everyone working on the report follows the same standards.
This year we used our internal AI tool in some disclosure statements, not to create material but to improve existing text. Teams drafted content, then used AI to test alternative structures or wording to improve clarity and brevity. Also, like BP we have many people working on the report, and ensuring correct and consistent terminology in the end product is a mammoth task. Using the tool to spot inconsistencies was hugely helpful.
And finally…
I hope this has given you the gist of the debate. The panel’s championing of ‘lived experience’ in reporting certainly made for a satisfying finale to our first series on this challenging and rapidly evolving technology. We hope you’ll join us for the next series, which begins in autumn. But first, enjoy the summer!