nearly every industry — including finance. In fact, the financial sector has long been an adopter of what we now call “traditional machine learning,” using it for predictive modeling, credit scoring, and risk analytics.
But with the current hype around Large Language Models (LLMs) and Agentic AI, I start to question: how can this industry truly take advantage of this new technology? Unlike many other sectors, finance operates under strict regulations, data privacy rules, and governance structures — conditions that, for me, seem a bit contradictory to the autonomy concept of Agentic AI.
To satisfy that curiosity, I attended the Agentic AI for Finance Conference [1] held in Jakarta, Indonesia, on October 16, 2025. The event — organized by Algoritma Data Science School [2] — brought together leading practitioners from banks, insurance companies, fintechs, government institutions, and AI startups in Indonesia to explore how Agentic AI might reshape the financial sector.
Although the discussions focused primarily on Indonesia, many of the insights reflected the challenges and opportunities faced by the broader landscape — from Southeast Asia or even the global finance industry. In this article, I’ll share my key takeaways, insights, and reflections from the event.
Disclaimer: I have no affiliation with any of the companies or organizations mentioned in this article. They are referenced purely for clarity and illustration. For more information about them, please refer to the references section at the end of this article.
1. ROI of AI
Every time a company plans to start a new project or adopt a new technology, the question of Return on Investment (ROI) inevitably arises. This is natural because to start something new — especially with AI — the investment required to build or integrate such systems is not small. Measuring how much financial return the company gets in exchange for that investment is essential.
However, in the case of AI adoption, accurately measuring ROI can be challenging. This happens because most of the time, when a company integrates an AI solution into existing operations, it’s often difficult to isolate the value generated purely by AI. The impact is usually distributed across multiple teams and technologies, making attribution far from straightforward.
One approach I learned from speakers to better understand this is to look at how organizations adopt AI at different maturity levels:
Boosting productivity
Achieving technical excellence
Enhancing or creating revenue streams
By looking at these levels, we know that ROI doesn’t always capture the full picture of AI’s impact. Many organizations are now beginning to complement ROI with Return on Value(ROV) [3], a more holistic approach that measures not only financial return but also aims to answer questions like:
Did AI improve decision quality?
Did it enhance customer satisfaction?
Did it improve internal productivity?
Another equally important perspective is the Cost of Inaction (COI). This represents the potential losses a company incurs by not adopting, or by delaying the adoption of, AI. According to Forbes [4], there are four key areas where a company will face challenges if it chooses to “wait and see”: widening knowledge gaps, difficulty attracting top AI talent, missed learning opportunities, and growing operational inefficiencies compared to competitors that have already become AI-enabled.
To conclude, while ROI will always remain a fundamental component in the decision-making process of AI adoption, companies need to complement it with other perspectives such as Return on Value (ROV) and Cost of Inaction (COI) to capture the full picture of AI’s impact and strategic importance.
2. Challenge as Regulated Sectors
The second key takeaway that I want to discuss here is security.
As I mentioned earlier, one major difference between financial and other sectors lies in the strict regulations and high level of data protection. For instance, the Indonesian Financial Services Authority (OJK) requires banks to keep their data centers and disaster recovery centers within Indonesia’s borders [5].
As the result of this regulation, financial institutions can’t freely adopt cloud-based systems. They must ensure that all data remains secure and follows local regulations. That is why many organizations prefer to run their AI systems on on-premise or hybridinfrastructures rather than relying fully on the cloud.
In addition, with the number of data breaches and phishing attacks, the financial sector must further strengthen their cybersecurity frameworks. One of the speakers emphasized that when integrating AI, everything must be secure and compliant before deployment — or it’s better not to deploy at all. The cost of failure in this area, whether financial or reputational, can be far greater than the cost of delay.
3. Agentic AI in Action
We have discussed two most important aspects before implementing AI in the organization. Now, let’s see some of the use-cases of Agentic AI that have been mentioned by speakers of the event.
Humanless Financial Reporting
Traditionally, financial analysts rely on large and various inputs — such as market prices, company filings, and news sentiment — to build their analysis. This process demands both speed and accuracy, as the financial landscape changes rapidly.
With Agentic AI, this workflow can be reimagined. By connecting to reliable and real-time data sources, AI agents with different specializations (e.g., market research, company reporting and news, historical data analysis, and report designer) collaborate autonomously to generate concise and data-driven report.
Decision-makers or analysts can simply ask question in natural language, and then multi-agent system orchestrates these specialized agents to deliver the report in PDF or slides format within seconds.
In addition, for recurring reporting tasks (such as daily or weekly updates), the user can schedule the system to generate the report with the most recent market data.
In my view, the reliability of the data source is the most crucial factor in this use case. To make the report trustworthy, we cannot let the specialized agent make up the data — what we called as hallucinating. So instead of using their own training data, we need to feed them with a curated, verified dataset.
Especially for market analysis purposes, platforms like Sectors.app provide a list of API endpoints that can be accessed by AI agents to retrieve actual market data. Using a trusted source such as this — or any other verified sources — helps minimize hallucinations, ensuring accuracy while enhancing analyst productivity.
A couple of months ago, I started my journey on Agentic AI using Sectors.app API and OpenAI Agents SDK. I create a simple Streamlit app that allows users to interact with AI Agents and ask questions related to companies listed on Indonesia Stock Exchange (IDX).
I’ve published this project on my GitHub, and the link to the repository can be found at the end of this article.
AI-powered Audit Process
One discussion that surprised me came from The Audit Board of Indonesia (BPK) — the government institution responsible for auditing the management and accountability of state finances. What caught my attention was how far they’ve already gone in adopting AI solutions within their operations, something I didn’t quite expect from a government body.
In collaboration with Supertype, BPK integrated AI solutions to their BIDICS platform, transforming a vast amount of audit documents into a queryable knowledge base to support the audit process — from searching, analyzing, and visualizing key data contained in reports [7].
They leverage the LLM for data extraction and categorization of documents, as well as for generating preliminary analytical insights. The AI-driven insights assist BPK auditors in planning, risk assessment, and decision-making before conducting detailed field audits [8].
One of the key challenges mentioned by the speaker was ensuring that the data access is limited only to authorized auditors, to prevent any potential misuse. This is particularly important since BPK has authority to collect and manage a large amount of sensitive financial data and documents.
In addition, BPK maintains a human-in-the-loop approach, meaning that all final decisions must involve human oversight — a crucial safeguard given the impact of every action taken by the institution.
Learning from BPK, as one of the highly regulated institution in Indonesia, and seeing how they’ve adopted AI while collaborating with third parties such as Supertype and several universities, made me realize that regulation doesn’t have to be a roadblock to innovation. Instead, it can serve as a framework that guides responsible and impactful adoption of new technologies.
How AI Can Help Notaries
Another interesting use case of AI comes from NOTAPOS [9], a platform developed to manage end-to-end legal documents and assists notaries and legal professionals.
Legal processes — especially in Indonesia — are often manual and time-consuming, from manual data entry, verification document uploads, and so on.
Driven by this fact, NOTAPOS leverages AI to streamline and automate these workflows. According to them, the platform can reduce the processes that typically take 18 hours to several days down to just 30 minutes.
Sounds magical, right?
But that’s not the main point I want to highlight here. In the next section, I’d like to recap what the founder shared about the struggles they faced during the development — lessons that I believe are valuable for us to explore from.
4. When AI Moves Too Fast
This is probably my favorite insight that I can gathered from the event when the experts shared another side of AI excitement.
So let’s continue the discussion about NOTAPOS from before. Back in their early period of the development, building a custom model that could understand domain-specific knowledge — in this case Indonesia’s legal context — required manual fine-tuning. They had to feed the model with hundreds of legal documents so it could learn the necessary context and terminology.
Now, with the rapid advancement of LLMs with larger and more diverse training data, much of this context already exists as part of their knowledge. Tasks that previously required extensive manual fine-tuning can now be done immediately, without additional training and setup, making development becomes faster and significantly more cost-effective.
This and other similar cases that happened due to the rapid progression of AI raise a new dilemma:
Should we keep building now, or wait for the next big leap that might make today’s efforts obsolete?
On a personal reflection, this question connects back to the earlier discussion of this article — the cost of inaction. In such a fast-evolving field, waiting might seem like a way to avoid “unnecessary” cost, but the real risk lies in failing behind — missing the chance to learn, experiment, and adapt to the technology as it evolves.
We cannot just wait. As the speaker wisely noted, “the key is to be able to predict where the technology will be in the next six months“. It’s not always about chasing every new trend, but about staying adaptable and positioning ourselves to seize each opportunity as it comes.
5. The Ultimate Question: Will AI Replace Humans?
One thing I really appreciate about this conference is how it brought together people from different backgrounds. This diversity was reflected in how the they responded slightly differently to the same big question — will AI replace humans?
Of course, this question can’t be answered by a simple yes or no. Most of the companies believe that AI is not here necessarily to replace humans, but rather to empower them. To ensure this, organizations need to invest heavily in targeted training programs to help employees leverage AI — at least to enhance productivity.
However, training alone doesn’t guarantee that people will adopt AI effectively, especially in companies with diverse generations and cultures. There will always be a group resistant to change, viewing AI as a threat, or simply too complex to use. This is where leadership becomes crucial — to guide and shift mindsets.
Still, if AI can perform certain repetitive or clerical tasks — probably better than humans — what happens to those whose roles are defined mainly by this work alone? One company shared that they have shifted a number of back-office roles toward more customer-facing positions as AI took over administrative tasks. (Unfortunately, I didn’t get a chance to get more clarity on why from back-office to customer-facing roles.)
Ultimately, both companies and employees share the responsibility to ensure continuous reskillling and upskilling to stay relevant in the ongoing AI transformation in the workplace.
Conclusion
Agentic AI opens up many opportunities; boosting productivity, achieving technical excellence, to creating entirely a new stream of business. However, none of this can happen without a solid foundation on regulation, data security, infrastructure, and human readiness.
In this article, we have discussed some of my key learnings from the conference. I’m grateful for the chance to see how different organizations are reacting to this new era of transformation.
Of course, this is not the end of this journey. In fact, parts of what we discussed here soon to be not relevant. However, that’s the reality of AI’s rapid evolution. Yet, we cannot afford to simply wait and see.
Reference
[1] Supertype.ai Conference — Agentic AI for Finance. https://supertype.ai/conference
[2] Algoritma Data Science School https://algorit.ma/
[3] Yedda Stancil — ROI (Return on Investment) vs. ROV (Return on Value): Understanding the Key Differences https://www.linkedin.com/pulse/title-roi-return-investment-vs-rov-value-key-yedda-stancil/
[4] Forbes — The Hidden Cost of Inaction on AI: Why You Can’t Afford to Wait and See https://www.forbes.com/councils/forbestechcouncil/2025/06/26/the-hidden-cost-of-inaction-on-ai-why-you-cant-afford-to-wait-and-see/
[5] OJK Regulation No. 11/POJK.03/2022 — Implementation of Information Technology by Commercial Banks. https://ojk.go.id/en/regulasi/Documents/Pages/Implementation-of-Information-Technology-by-Commercial-Banks/OJK%20Regulation%2011%202022%20concerning%20Implementation%20of%20Information%20Technology%20by%20Commercial%20Banks.pdf
[6] Sectors.app — Sector Financial API https://sectors.app/api
[7] Supertype.ai — LLM Development and Collaboration with BPK https://supertype.ai/llm-development
[8] INTOSAI Journal — BPK BIDICS: From A Question That Has No Answer https://intosaijournal.org/journal-entry/bpk-big-data-analytics-bidics-from-a-question-that-has-no-answer/
You can find the source code and my early learning milestone in this GitHub repository. In this project, I used OpenAI Agents SDK, Sectors.app API, and Streamlit to build a simple interactive financial app that allows users to interact with AI Agents and ask questions related to companies listed on Indonesia Stock Exchange (IDX).
We use cookies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it.AcceptPrivacy Policy
