Another enticing job posting on LinkedIn: Bank of New York Mellon (BNY) is looking for no less than a **Senior Vice President, POM Technical Product Management**. The role is Product Owner for Fund Accounting in the EMEA region. Sounds impressive.
**Briefly about the company:** BNY Mellon is not just a bank. It is one of the pillars of the global financial system. They themselves state that they influence nearly 20% of global investment assets. It’s a gigantic, hyper-regulated, conservative (despite all claims of innovation) machine. A machine that cannot afford to make mistakes.
**What pain point are they trying to solve?**
Let’s be frank. I’ve managed IT teams for twenty years and have seen dozens of such vacancies. If you strip away the corporate fluff about “transformative solutions” and “empowering employees,” the essence boils down to this:
BNY has an extremely complex system for fund accounting in Europe, the Middle East, and Africa (EMEA). It’s a tangle of client demands, constantly changing regulatory requirements (hello, European directives!), internal operational needs, and draconian risk management rules.
They need a one-person orchestra. A person who will:
1. **Listen to everyone:** clients, operations staff, lawyers, compliance, developers.
2. **Understand everything:** from the intricacies of NAV (Net Asset Value) calculation to the specifics of corporate actions and taxation in Poland and Luxembourg.
3. **Translate:** convert business language (“We want it faster and with fewer errors”) into development language (“Create an epic with a user story for automatic data reconciliation from source X via API Y with acceptance criteria Z”).
4. **Prioritize:** decide what’s more important — a new feature for a key client that will bring millions, or closing a security vulnerability mandated by a regulator, which will bring nothing but peace of mind.
Essentially, BNY is looking for a super-communicator and super-prioritizer with a vast knowledge base in their head. A person who will serve as a living hub, constantly juggling conflicting requirements and trying to keep all the balls in the air. The salary, I’m sure, is commensurate. And so is the headache.
**What if this hub isn’t a person?**
Imagine a dialogue between two IT directors over coffee.
— They’re looking for a “product person” for fund accounting again. The third one in five years. One burned out, the second left for a promotion with a competitor. It’s not a job, it’s constant stress.
— Why do they even need a person? They boast about using “cutting-edge AI.” Why not let it do the job?
— Are you serious? Put AI in charge of the backlog in a system where an error costs millions? Compliance would eat them alive.
— And do they have no errors now? Human error is always a factor. A tired Vice President missed something, misunderstood a priority, forgot a dependency — and boom. AI doesn’t get tired, doesn’t burn out, and doesn’t have bad days. The only question is how to properly “prepare” it.
**AI Solution: Creating a “Product Brain”**
Instead of hiring another expensive specialist who will become a bottleneck in the system, an AI-powered analytical core can be built. This isn’t a single tool, but a combination of technologies.
**Step 1: Collection and Digitization of All Incoming Data.**
The AI system must “read” and understand everything that a human product person does. For this, we connect it to the following sources:
* **Regulatory Documents:** We set up parsers that monitor updates on the websites of European regulators (ESMA, EBA, etc.). An LLM (Large Language Model) analyzes the text of new directives and extracts specific requirements for IT systems. For example: “Change reporting format for form X by July 1, 2025.”
* **Client Requests:** We integrate AI with Jira, Zendesk, or even email. It classifies requests, determines their urgency based on text analysis and client data from CRM (client size, their influence).
* **Operational Data:** We connect the system to logs, monitoring systems (Datadog, Splunk). The AI analyzes error rates (break rates), NAV calculation time, process bottlenecks (cycle time), and generates technical tasks for optimization itself.
* **Dependencies:** The AI scans the codebase and architectural diagrams to build a dependency graph between systems. It will know that a change in the pricing module might affect the financial reporting module.
**Step 2: Prioritization and Backlog Creation.**
This is the most interesting part. A human product person has an evaluation model in their head. For AI, it will be formalized and objective.
* A **scoring model** is created, where each potential task (feature, bug, technical debt) is evaluated based on several parameters: revenue impact, implementation cost, risk reduction (regulatory, operational), client satisfaction, technical debt.
* The AI automatically assigns scores. For example, a regulatory task with a clear deadline receives the maximum risk score. A feature requested by 10 small clients receives a lower score than one requested by a single, but largest, client.
* The output is a **dynamically prioritized backlog**. It is recalculated daily based on new data.
* Then, based on this, the LLM **generates ready-to-use user stories** with “crisp acceptance criteria,” as the job posting requires. It takes the essence of the requirement and frames it in a form ideal for developers.
**Step 3: Communication and Reporting.**
AI cannot go for coffee with stakeholders. But it can do what takes a human hours:
* Generate **personalized dashboards and reports** for each group: for compliance — a report on regulatory requirement fulfillment; for client service — status on their requests; for top management — a summary of product KPIs.
* Automatically send **notifications about releases, risks, and roadmap changes**.
**How to Overcome Distrust?**
Of course, no one will give AI the keys to the financial system from day one. Implementation should be phased:
1. **”Advisor” Mode:** Initially, the AI operates in the background. It analyzes data and proposes its version of the roadmap and priorities. The human Vice President makes the final decision. This allows for model calibration and proving its viability with real data.
2. **”Partner” Mode:** The AI is given control over a small, non-critical part of the product. For example, internal analytical reporting. The team works according to its backlog, proving effectiveness in practice.
3. **”Autopilot with Observer” Mode:** The AI manages 90% of the process, while a human expert (no longer a Vice President, but rather a “Chief System Architect”) verifies key decisions and intervenes only in exceptional situations.
**How to Validate the Results?**
Very simply. With data.
* **Backtesting:** We load two years of data into the system (all requests, all incidents, all regulatory changes). We compare the roadmap that the AI would have proposed with what was implemented by humans. What would have brought more value? Where would there have been fewer emergencies?
* **KPIs:** We compare key product metrics before and after implementing the AI assistant: time to value (cycle time), number of bugs in production (defect escape rate), timeliness of NAV calculation, client satisfaction. Numbers don’t lie.
BNY states in its job posting that they “use cutting-edge AI to create transformative solutions.” So, isn’t it time to start with themselves? Perhaps the biggest transformation is not to hire another Vice President to manually untangle this complexity, but to build a system that makes this complexity manageable. And then, instead of one Vice President, you’ll need a couple of capable engineers and analysts to support it. And that, as my experience tells me, is both cheaper and more reliable.
Источник: https://www.linkedin.com/jobs/view/4390630928/
