An annual report is more than a display of bare numbers and facts. It is also about language. How companies describe information, what terms they use, how often and in what context, it all contributes to the sentiment that the text evokes in the reader. In other words, language in an annual report drives readers' perceptions regarding the performance and reputation of the relevant company.
How do Dutch listed companies use language in their annual reports? After an extensive ‘sentiment analysis’ of recent annual reports, analysts at KPMG found that in their annual reports, the companies surveyed write in more positive terms about topics that are strategically important to them compared with topics they consider less strategically important, such as regular compliance topics. With regard to the climate policy pursued, the text sometimes comes across as too positive, which can lead a company to be accused of greenwashing. At the same time, dilemmas and challenges are hardly described, if at all. This observation raises the question of whether companies should adjust their use of language in reporting to allow the diverse readers of the annual report a more balanced or, as the case may be, more realistic picture. And if so, how should they adjust their language to create a realistic image of the company. This would ensure the outside world gains better and correct information.
Algorithms are continuously improving
Analysts of KPMG's department Digital Assurance & Innovation collaborate with experts from other fields within KPMG, such as ESG (Environmental, Social and Governance) & Sustainability, as well as consultants from Advanced Analytics & Big Data Engineering. They applied statistical methods with machine learning techniques, enriched with domain knowledge of ESG and audited by KPMG professionals, enabling this textual sentiment analysis. Through continuous innovation, KPMG manages to refine this AI-based technology on an ongoing basis.
Meanwhile, our auditors regularly apply sentiment analysis in audit practice. This allows them to better indicate whether the combination of financial facts and the sentiment evoked by the text are sufficiently balanced. The auditor's opinion is the basis of determining the extent to which the quality of reporting should be improved. The technology is based on an algorithm that we have trained extensively to consistently and efficiently provide a clear picture of sentiment by topic in the annual report. Part of this training involved presenting our auditors with thousands of sentences from annual reports and having to qualify each sentence as positive, neutral or negative. This allowed the algorithm to become increasingly discerning on how to judge the sentiment of an annual report text. As a result, this is now a very rapid process: the software is capable of reading and analyzing 200 pages in 15 minutes.
Discrepancies visible
We see a variety of discrepancies in the texts of annual reports based on this sentiment analysis. A few examples: on the one hand, a company states that it is a Great Place to Work, where talents feel at home and thrive; while the numbers show that the company is having great difficulty retaining talents, necessitating them to retain employees by offering a substantial salary increase. Or a company boasts of reducing CO2 emissions, while later in the annual report, it reveals that production has increased substantially while no significant CO2 mitigation measures have been taken. In addition, sentiment analysis also filters out when certain words are used with excessive frequency, or when one section paints a much more positive picture than another (which can happen because multiple departments provide text for the annual report).
Are the claims true?
An advantage of using this sentiment analysis is that companies can properly compare their own information with that of industry peers, as far as publicly available information is concerned (such as an annual report). This is possible for any financial data, but also, for example, on how often industry peers use certain ESG terms and in what context compared to the company’s own terms and information. Are a company's ESG claims and suggestions true ("we are the most sustainable player in our industry")?
KPMG offers this sentiment analysis as a standard part of services to a growing group of Dutch companies. With our English-language version, we also serve companies with an international scope. In addition to the annual report, the tool can also be used for press releases, sustainability reports and other corporate texts.
Assessing ESG Data Quality: is your information correct and complete?
KPMG can also use this technology to check whether certain words are inappropriately omitted or understated in the annual report. That is useful, for example, when the Corporate Sustainability Reporting Directive (CSRD) goes into effect in early 2024. Pursuant to this directive, the European Commission requires large European companies to report on the environmental and social impact of business activities, and to have this information reviewed by an auditor. The CSRD aims to increase the quality of information and transparency about companies' environmental and social impacts. A sentiment analysis also enables the auditor to have the text of a report compared with the mandatory CSRD topics and KPI’s (water consumption, CO2 emissions, equal pay, child labour, et cetera). Are these CSRD topics integrated? And are they presented in the appropriate or desirable context?
