The content on this page has been converted from PDF to HTML format using an artificial intelligence (AI) tool as part of our ongoing efforts to improve accessibility and usability of our publications. Note:
- No human verification has been conducted of the converted content.
- While we strive for accuracy errors or omissions may exist.
- This content is provided for informational purposes only and should not be relied upon as a definitive or authoritative source.
- For the official and verified version of the publication, refer to the original PDF document.
If you identify any inaccuracies or have concerns about the content, please contact us at [email protected].
The use of AI in corporate reporting Case studies
Introduction
The FRC commissioned Lancaster University (with a team including academics from Loughborough University) to undertake research on our behalf on the use of AI in corporate reporting.
A summary of findings was published in July 2026. The research included 39 interviews with senior members of reporting and investor relations teams (plus other external corporate reporting stakeholders including advisors, third party software vendors and auditors), from which the following three case studies are derived.
We note that these case studies highlight interesting examples of existing practice by companies and do not represent an endorsement or recommendation by the FRC.
The FRC and Lancaster University would like to thank Falcon Windsor for help securing interviewees, along with the following organisations who also supported in recruiting research participants: Chartered Governance Institute UK and Ireland, Chartered Institute of Internal Auditors, ICAEW, ICAS, Investor Relations Society, Quoted Companies Alliance, and The 100 Group.
The research team was made up of Prof. Steven Young, Dr Mahmoud Gad and Dr Dasha Smirnow (Lancaster University), Prof. Ken Lee (Loughborough University), and Dr Yasmine Chahed (Independent Senior Adviser and Research Consultant).
2. AI Case study 1: Corporate report production efficiencies
The organisation:
A large multinational with a central finance function responsible for consolidation, forecasting, and external reporting. The reporting process is underpinned by a combination of enterprise tools, including consolidation systems for capturing financial data, a disclosure management platform for report production, and a number of workflow tools to manage review and sign-off processes. Together, these technologies support the end-to-end preparation of financial disclosures and reports.
Approach
The company is working both with AI capabilities within the current tools as well as embedding AI capabilities across various processes, particularly through the use of an enterprise licence of a Large Language Model (LLM) based GenAI assistant.
A key use case for LLM-based tools in the reporting process is AI to support navigating large and complex documents. The organisation’s finance team uses AI as a starting point for gathering relevant data and identifying pertinent sections within lengthy internal reports for inclusion in annual report documents and papers. Similarly, the AI assistant is used by the finance team when addressing technical accounting issues as it can quickly identify and quote relevant standards, supporting the development of the team’s position on a matter.
Another use case for AI in the financial statement preparation process has been the analysis of peer disclosures. By surveying how comparable organisations report on specific topics, the team can benchmark its own disclosures and better understand prevailing practices, ultimately providing useful reference points for discussion (rather than directly changing inputs into published reports).
The organisation is also seeing some AI capabilities being rolled out to existing third party tools such as an AI-driven scenario planning and analysis tool for forecasting and budgeting. The tool is integrated with the reporting platform so that it pulls inputs directly to support the budgeting process. This integration ensures alignment between planning, budgeting and reporting activities.
Outcomes
The organisation considers that AI tools are helping create efficiency around the edges (like saving time searching for data points) but are not yet fundamental. The organisation also noted that AI has allowed them to consider small incremental improvements and enhancements to processes that they can build or trial within the team.
"In the finance team we say: You're aware what are the crunch points? What takes the time? Where are we doing manual work? And if you've got an idea to fix it, great, let's see what we can get on and do ourselves.... This kind of micro automation that people can do on their day-to-day work builds momentum, and all of a sudden, you're starting to move into a better place."
Further Considerations
The organisation notes that while the use of AI is enabling faster access to information and updates, with greater automated efficiency, the role of human oversight and checks is still critical to test and not over-rely on the results and is no replacement for a well-structured and integrated reporting system. The organisation also recognises that their data quality needs to be strengthened to facilitate greater and more productive use of AI.
3. AI Case study 2: Use of AI by IR function
The organisation:
A large multinational with a central investor relations (IR) function responsible for earnings communications, including press releases, analyst presentations, and Q&A preparation.
Approach
The organisation has not used AI to directly draft any disclosures, but it has embedded it in analytical and review processes that shape external reporting. A core use case is that the team use an enterprise version of GenAI to support tracking narrative consistency across disclosures and over time. For example, after drafting performance summaries, the team uses AI to compare current language against numerous prior quarters, identifying shifts in style, emphasis, or message, ultimately assessing how the tone has evolved. The team note that the goal is not to enforce uniformity, but to bring to light any unintended changes.
AI is also used for sentiment analysis where disclosures are reviewed and scored for positivity or negativity; this is to approximate how investors or external tools (including other AI) might interpret them. However, the team notes that limited understanding of actual consumption of disclosures by investors and AI-based tools means that the value is placed on detecting changes in messaging rather than optimising absolute sentiment scores.
A further use case is reviewing and comparing competitor information. The IR team collects significant numbers of earnings transcripts, presentations, and Q&A from peers, using AI to extract themes and topics across the market, helping to assess how content may be interpreted and prepare for questions that may arise.
These capabilities are enabled by several organisational factors:
- extensive data capture practices, including routine transcription of internal discussions, provide rich inputs for AI analysis;
- embedded access to enterprise tools (e.g. Copilot integrated with internal systems) allows querying across large document repositories;
- structured prompts and controlled data inputs help manage output quality; and
- all outputs are subject to human review, although the extent of human oversight depends on materiality of information.
Outcomes
The organisation considers that the use of AI enables synthesis of volumes of unstructured data that would previously have required extensive manual effort. The output informs how the organisation positions its own disclosures, providing contextual market awareness rather than directly leading to the generation of content.
"It gives us a better sense of how what we're going to say will land given the broader market... we can get better insights faster in ways that we just did not have the manpower or time to do previously and can now understand what competitors are saying in a way that we would have never been able to before."
Further considerations
Looking ahead, the organisation aims to automate these workflows further, particularly competitor analysis and narrative benchmarking, reducing manual data handling and enabling real-time insights.
4. AI Case study 3: Enhanced anomaly detection and managing data with multi-agent architecture
The organisation:
A large multinational with complex global reporting requirements wanting to improve its internal reporting processes and assurance activities.
Approach
The organisation launched an enterprise-wide GenAI programme which included AI-enabled anomaly detection combined with diagnostic suggestions of what may be causing the error, discrepancy or omission. The system monitors reporting data and flags missing data, outliers, or late submissions, particularly during reporting cycles. The approach goes beyond simple detection and validation checks by building the diagnostic capability directly into reporting workflows. The resulting alerts are pushed into operational channels (e.g. Teams or email), enabling earlier interventions.
"It will alert authorised users to things like missing data or anomalies or outliers or time critical data that's late [...] And then the AI solution would also provide a suggestion as to what's wrong with the data: it might be incorrect currency or it might be an odd decimal point, given an analysis of the last 12 months' numbers.”
The organisation also introduced an enterprise-wide multi-agent architecture. At the core is a "master agent” (implemented via Copilot), which acts as a unified interface for users. This master agent sits over multiple domain or team specific "child agents” (e.g. finance, tax, treasury, HR), each which is connected to distinct data sources and governed by specific permissions structures allowing the right data to be available only to the right people.
These agents are enhanced with additional data connections and specific models. For example, structured financial data is accessed via enterprise data platforms, while unstructured documents (e.g. policies, reports) are queried using different LLMs optimised for those formats. This approach allows the system to combine multiple data types while maintaining control over access and data integrity.
Several enabling factors underpin this development:
- the organisation has invested in a structured data architecture ("medallion" model), ensuring high-quality, standardised data feeds for AI;
- it has also established a centre of excellence and has close collaboration with technology providers to co-develop and deploy agent-based solutions; and
- strong executive sponsorship, particularly from the CFO, has accelerated adoption and prioritisation.
Outcomes
The benefits observed by the company are primarily in efficiency and control. AI reduces manual review effort (e.g. scanning reports for completeness) while improving detection of data issues earlier in the reporting cycle. This supports a “90/10” operating model, where AI performs most of the processing and allows the team to focus on final validation and assurance.
Further considerations
Currently, the level of trust in outputs is seen as a key constraint to further adoption, with the organisation now focusing on building trust through testing and user involvement. There is also a focus on extending anomaly detection and summarisation capabilities to regulatory updates and narrative reporting.
5. Risk Considerations
These case studies highlight some possible common risks and concerns arising from the use of AI in corporate reporting practices. Whilst each organisation needs to judge for themselves the best approach to controls and mitigations, some common areas that may be considered include:
| Risks | Mitigations and controls |
|---|---|
| Over-reliance on AI outputs and risk of errors There is a risk that incorrect, incomplete or misleading outputs are accepted too readily, particularly where outputs appear plausible or are produced at speed, especially where sourcing and quoting accounting standards and providing diagnostics. GenAI could still be prone to hallucinations or faulty reasoning. | AI should support, not replace, human judgement. Ultimately directors are responsible for the output of the annual reporting process. Often, companies have:
|
| Weak or inconsistent human oversight The "90/10" operating model and increasing automation could move human review later in the process or up the seniority curve, creating a risk that staff lose practical experience of the underlying reporting process (especially where turnover is high). | Valuable "human in the loop" oversight is essential to ensure the right review at the right time. Often, companies have established:
|
| Data quality, integrity and access-control risk Poor data, incomplete repositories or overly broad permissions could lead to unreliable outputs or inappropriate access to sensitive reporting information. | Multi-agent architecture, enterprise search and large document repositories depend on quality input data, appropriate permissions and well-governed data connections. To support this, companies often implement:
|
| Misuse of AI in narrative reporting and loss of management voice Use of AI in drafting and supporting tasks such as tone and sentiment analysis may erode authentic management commentary or create unintended shifts in tone and messaging if not used and reviewed appropriately. | We see companies cautiously adopting AI as a supportive co-worker rather than in a co-creator role. Often, companies:
|
6. Conclusion
The case studies are from large multinational organisations which already had large and complex IT estates and have refined their technology use. AI processes and control require significant staff resources and a level of training and skill. Organisations with less mature use of AI and technology may lack the skills and resources to evaluate outputs, design appropriate controls, or identify safe and productive use use cases. While the level of resources can be a barrier, executive sponsorship and support are always critical to enable responsible adoption, balancing innovation with risk management. Furthermore, sharing of expertise and experience among colleagues through AI champions or a Centre of Excellence can allow controlled experimentation, with benefits shared more widely across teams.
7. Definitions
Enterprise search: technology, usually AI-enabled, that indexes, retrieves, and organises information across an organisation’s internal data repositories. It enables employees to find documents, messages, records, and knowledge across multiple systems, as bound by permissions. It typically delivers answers instead of documents, with full citations and strict permission controls.
Large language model (LLM): an artificial intelligence model (typically a neural network) trained on a vast amount of text for natural language processing tasks, especially language generation. LLMs can typically generate, summarise, translate, and analyse text in many contexts. They are the basis for many modern GenAI chatbots or assistants, such as ChatGPT, Claude and Copilot.
Medallion model: a layered data architecture that organises data into multiple layers, each representing a higher level of quality and structure: Bronze (raw data), Silver (validated and cleansed data), and Gold (curated and analytics-ready data). Its primary goal is to incrementally improve data quality and reliability as data flows from ingestion to analytics, making it suitable for business intelligence, machine learning, and reporting applications.
Multi-agent architecture: a structured system where multiple specialised AI agents collaborate to solve complex tasks under a coordination layer that manages communication and governance, and will include communication protocols and governed components, including agent registries, memory, observability, and secure tool access.
Get in touch
London office: 13th Floor, 1 Harbour Exchange Square, London, E14 9GE
Birmingham office: 5th Floor, 3 Arena Central, Bridge Street, Birmingham, B1 2AX
+44 (0)20 7492 2300
www.frc.org.uk Follow us on LinkedIn