Reporting. The one discipline everyone working on creating it agrees (generally):
- It’s tedious,
- It can be nerve-racking.
- It is just words on paper, and does not create (positive or negative) impact per se.
- It’s often legally required, and with that comes the ‘buzz’ of it being legal compliance. And
- It can suck the life out of an individual or a team.
For those reading the result however, it can be rather exciting stuff – if a report is well done, insightful, leverages complex, interlinked data, and manages to structure in an understandable manner what otherwise is more intuitive, unstructured bits and pieces of data and knowledge.
Hence, with this in mind, I’d like to focus today’s blog post on digging into where and how specifically AI can be of use and benefit for the ESG reporting discipline.
The three areas of reporting we’’ll be looking at in particular are:
- Automatisation of Compliance Tracking
- Data Collection and Integration
- Support Audit & Review Preparation
All of these applications areas fall into Step 2 and Step 3 – depending on exact need and required results – of the below AI maturity model, and broadly fall into Retrieval Augmented Generation (RAG) uses cases.
Automatisation of Compliance Tracking
The reporting process often is tedious not the least because it requires – as of today – large amounts of manual cross checking against regulatory requirements.
The good news is: AI can be the helping hand of compliance officers.
The most evident such case is the tracking and interpreting regulatory requirements (e.g. those outlined in the EU’s CSRD) and generate / maintain / update the related reporting templates.
In this way, via technology, it can be ensured that it is highlighted where what data is required, and gaps can be found somewhat easier and in a more streamlined manner.
In the case of the CSRD, like for most new regulation, this ability is particular of relevance in the initial phases of its implementation, where guidelines may still change and be updated, or indeed clashes with other pieces of regulation emerge.
Of course AI can come in handy also to fill those pesky data fields, too – see the next chapter for further detail on that.
Data Collection and Integration
With the help of AI, data can be aggregated across a broad variety of sources, of which some could be structured data (e.g. typically from databases) as well as unstructured data (e.g. from text documents, meeting minutes, social media etc.).
Let’s take a look at a few concrete examples of what ‘structured’ and ‘unstructured’ data means in practice:
| Sustainability Area | Structured Data | Unstructured Data |
|---|---|---|
| Environmental Data | Greenhouse Gas Emissions | |
| CO2, CH4, N2O emissions (metric tons) Data drawn from Emissions monitoring system data by scope (Scope 1, 2, 3) EPA datasets on GHG emissions | Satellite Imagery: Satellite data on pollution and deforestation, requiring analysis | |
| Energy Consumption | ||
| Electricity, gas, water usage data (kWh, GJ) Energy meter data by facility or equipment | Maintenance Logs: Free-text logs from technicians detailing system efficiencies and repairs | |
| Waste Management | ||
| Hazardous vs. non-hazardous waste data by disposal method (landfill, recycling, incineration) | Inspection Reports: Qualitative audit reports with handwritten notes | |
| Social Data | Labour Practices | |
| Employee headcount by region, gender, role Occupational health/safety data (incident rates, lost workdays) Training hours per employee by type | Survey Responses: Open-ended feedback on workplace conditions Interviews: Transcripts/audio recordings discussing labour practices | |
| Labour Compliance in the Supply Chain | ||
| Labour Compliance Metrics: Data on adherence to labour laws, including minimum wage, working hours, and safety regulations. | Inspection Reports and Audit Narratives: Detailed descriptions and qualitative findings from labour audits and inspectionss. | |
| Diversity and Inclusion | ||
| Diversity Statistics: Quantitative data on the demographic composition of the workforce, including gender, ethnicity, and age. | Diversity Program Reports and Employee Testimonials: Narratives and descriptions of diversity and inclusion initiatives, personal stories, and qualitative feedback from employees about the workplace culture and inclusivity efforts. | |
| Governance Data | Board Composition | |
| Board member diversity (gender, age, independence) Attendance records for board/committee meetings Executive compensation data | Meeting Minutes: Narratives from board meetings on governance Personal Statements: Public letters/statements from board members on governance and ethics | |
| Compliance and Ethics | ||
| Reported incidents of non-compliance Internal audit results (number of audits, findings) | Whistleblower Reports: Narrative reports/recordings of non-compliance incidents Legal Documents: Court filings and legal briefs related to compliance cases | |
The beauty of this is of course, that such data sources can be internal, external, paid for, or open source.
Of course, the human error component is probably one that comes immediately to mind when peddling AI as a helping hand in aggregating and compiling data: Automated data entry and data validation process – including enhancing data quality or flagging the lack thereof—can take those tasks off human beings, and free their time for the more demanding areas (e.g. figuring out the whereabout of such data ….)
Some of that data may be available additionally in real time – and AI can help to digest that as soon as available, including relevant reactions and process optimisation, at a scale while, and while dealing with massive amounts of data.
If done well, AI can hence ‘grow with’ a company’s growth, and the related growth in reporting requirements, available and in need of digestion data volumes, as well as the complexity that arises from country-by-country variability.
This is not to say that it is all smooth sailing, and that all that benefit comes without its complexities. Complexities keep existing – just shift to a different level, and different areas. Maybe the one issue though that gets indeed significantly more difficult to get right with the application of AI is Data Privacy and Security, as notably internal data is most likely of sensitive nature ….
It is important to say though that while AI could take a lot of the tedious work off our hands, it is remains to be seen how the tech is able on the long run to compensate / work around the lack of data quality.
At this very moment ‘crap in is crap out’ , or in other words: if data quality is insufficient at the outset, the tech per se cannot compensate for it. Hence human supervision is indispensable.
And lastly: integrating yet another system into the corporate digital tool park, and notably the one fine tuned for reporting, is a not to underestimate undertaking and requires careful planning. IT and reporting departments need to ensure that AI tools are compatible with their current systems and can handle the specific reporting requirements of their industry.
Support Audit & Review Preparation
The analytical capabilities of AI can be adapted to act as ‘first level audit’ scrutinizing ongoing reporting and fact check the underlying data and the links to original audit trails to assess the consistency and plausibility of data in a first instance.
For auditors, it could give them a helping hand to assess the accuracy and completeness of sustainability reports, and to ensuring that the audit trails truly go back to every data entry and modification. Cross checking and probability checks on the reported insights could also be automated, as can – at the same time – the preparation of fine granular data break downs to assist this purpose.
Beyond audit, but supporting Due Diligence by stakeholder groups (e.g. investors, customers), AI tools could be used to generate custom-made break downs; or to generate for each of these groups their very own version of the financial statements based on the different types of analysis they run and information they are looking for.
Nevertheless: AI can at this moment NOT give fundamental support in interpreting the information contained in reports, or indeed any kind of overview / Dashboard. Which means: the design of precisely things like dashboards or progress / impact reports is per se still consequence of our very human intelligence, albeit the manual work and ‘caching’ of information may no longer be so.
Conclusion
There is no shade of doubt that AI can, and will, benefit reporting from a number of angels: from reducing the time spent on the boring tasks of data aggregation to the more sophisticated first level audit reviews, and including the support of compliance officers in keeping abreast of regulatory data reporting requirements.
This said, like so often, just because available processing power allows us to do with tech what was deemed hard before, there remain critical points to be kept in mind:
- Crap in is crap out: for the time being at least, the quality of the AI application’s result rely still on the quality of the data that goes into the system. By ‘data quality’ I do not mean legibility, or availability etc, but rather: ‘meaningfulness’. However amazing the AI’s juggling of data may be, if the data has no meaning to start with, or indeed is forged data at the outset, the results cannot possibly result in anything desirable for the purpose at hand.
- IT always costs money – and well done IT certainly does.
So does wasting time of highly qualified sustainability professionals.
Clearly outlining the costs and wins is as important as ever. Without forgetting much harder to monetise aspects such as people’s work motivation, self-efficacy, and career development. - Last and not least: good AI experts, that understand the software development side of it, are few and far between. Focusing on leveraging ‘of the shelf’, relatively easily customisable AI software offers is hence priority above any custom build.
Leveraging those successfully will allow to do the learning.
