AI in Financial Communication: why Manual Tagging will soon be history
At the end of the year, one thing has become clear: financial communication is facing a profound transformation. After intensive work with large language models (LLMs) in corporate reporting, it has become evident that the industry has massively underestimated the pace of this development.
The question hardly anyone asks: Will AI take over accounting?
Traditionally, accounting has been based on clear, deterministic rules — such as XBRL taxonomies and standardized data structures. In contrast, AI systems work probabilistically, learn from statistical patterns, and generate content based on likelihoods.
These are two completely different approaches — and that’s exactly why the answer to whether AI can take over deterministic accounting is so surprising.
Status Quo: Pattern recognition works — better than expected
Our tests with corporate and sustainability reports show that LLMs can accurately recognize typical structures and formulations, process data reliably according to learned patterns, and already automate 80–90 % of standard tasks.
Where AI still has limitations
The remaining 10–20 % are critical. These include issues such as:
- multi-document safety (avoiding contamination by external templates),
- causal inference for complex topics,
- contextual interpretation of contradictory data,
- and checking logical consistency across different parts of a report.
For now, manual verification is still required — but that will soon change.
Short-term: Inference-capable AI is closer than many think
With new model architectures, larger context windows (200,000+ tokens), increasing compute power, and improved training techniques like reinforcement learning from human feedback (RLHF), AI is developing rapidly.
Over the next 12–15 months, we expect systems that can much more: recognize cause-and-effect in business developments, automatically identify inconsistencies between balance sheet, income statement, and management report, and assess the relevance of information in the specific context of a company. These capabilities are no longer futuristic — they are already emerging in current development cycles.
Long-term: Logical inference — even in regulated markets
In two to three years, AI will not just recognize patterns but draw logical conclusions. It will analyze content without explicit training, evaluate the relevance of business events on its own, and check consistency across all formats — from financial reports and investor presentations to annual general meeting speeches and press releases.
Automated compliance checks will become standard
Another milestone will be automated verification of legal requirements as an integral part of production environments. But the human responsibility remains: even with inferential AI, final approval by legal representatives is indispensable.
The uncomfortable truth about accounting
The core point is quickly stated: AI works probabilistically — accounting traditionally deterministically. But reality shows that probability-based approaches are already everyday practice. Risk assessments, sampling tests, or “reasonable assurance” instead of absolute certainty are accepted practices. The key question is: when will regulators accept 99 % AI accuracy?
Our recommendations for companies
1. Start now with AI-supported processes
The learning curve is steep — those who start today will have a massive competitive advantage in two years.
2. Rethink IT investments
Long-term investments in manual tagging can turn out to be cost traps.
3. Build AI competence within your team
The question is not if but who can use AI competently. Expertise matters.
The Silvester Group has been guiding companies through transformation and innovation in financial communication for over 30 years. As a specialist in digital reporting solutions and AI-supported processes, we help our clients meet regulatory requirements and strategically leverage technological opportunities — precisely, efficiently, and future-oriented.
