AI, auto-tagging & ESEF: Mirage or reality?

With ESEF’s 2020 transition fast approaching, many affected companies will soon have to start getting ready for the implementation of ESMA’s new regulatory mandate and the first reporting deadline in 2021.

Under ESEF, all EEA-listed companies reporting under IFRS will no longer be able to publish their annual financial reports in PDF format and will instead have to deliver them in digital format, using an extensible business reporting language called ‘inline XBRL’ or iXBRL.

To do that, they’ll generally have two main options: to either acquire and implement an iXBRL tagging technology in-house or to outsource the process to a third party iXBRL tagging provider.

While selecting a solution depends on a business’s individual needs in terms of budget, resources and existing processes, there’s one particular feature that most companies would usually look for: auto-tagging.

But, what exactly is auto-tagging?

With the rise of Artificial Intelligence (AI) technology, users are now able to automatically harness large amounts of data and quickly and accurately compile reports. So, by applying machine-readable tags to data, iXBRL allows for the use of AI to compare that data to existing (previously tagged) data and therefore automate the tagging process.

Does it mean that companies will only need an AI-empowered iXBRL technology to comply with ESEF?

Unfortunately, it doesn’t. AI can only provide a partial solution to ESEF as there are some specific nuances that will still require human input. For example:

• ESEF is introducing the concept of ‘extending and anchoring’, which simply means that where an item can’t be tagged back to IFRS, an extending and anchoring tag is created and associated with the nearest existing item. There’re also two types of anchoring, ‘wider’ and ‘narrower’, depending on an item’s level of specificity. Working out those relationships can, therefore, be quite challenging for AI;

• In addition to data in the primary financial statements, ESEF is requiring numbers in related footnotes to be tagged, too. That’s likely to be another issue for AI as a) it’ll need to find the numbers in the text and then b) it’ll need to find out what they relate to. In the situation where those numbers would be extended/anchored, that’ll be practically impossible.

Even in a world where AI works perfectly well, the tagging in an ESEF document needs to be 100% correct. Therefore, companies would still need a human (an experienced one, who knows tagging and iXBRL well) to be reviewing the output before they put it on their website for the world to see.

That’s why at Arkk, we’ve taken an agnostic approach to ESEF and its technical implementation. And even though AI is impacting all parts of how we work, we won’t be relying 100% on its validity for ESEF.

To find out more about our ESEF solutions get in touch.