The Role of AI in Finance: How We Think About It at Flex

This will be the first in a series of blog posts that dive deep into exactly how we use AI at Flex down to the smallest details. 

In this first article, I’m going to give you a glimpse into some of the ways we think about AI in finance at Flex at a high level — particularly on the engineering side. We’ll talk about why AI matters for our engineering teams and how we use it responsibly to balance speed, compliance, and trust.

When people ask me about how we use AI in finance at Flex, I often tell them it’s not just about automation — it’s about sharpening the quality of our thinking.

At Flex, we build financial tools that business owners rely on to manage money, streamline operations, and stay compliant in a regulated industry. To make that possible, we don’t just “add AI” to our workflows. Instead, we use it as a thought partner; a way to test ideas, refine them, and ultimately build a financial platform that solves real problems for our users.

It’s important to note that when I refer to AI, I’m speaking to the collective capabilities of Large Language and Large Reasoning Models (LLMs + LRMs). We very much understand they are all different but we’ll use AI colloquially here.

The Wonders of AI in Finance and Accounting

AI isn’t just a tool for engineers or customers — it has broad applications across the banking and finance industry as a whole:

• In accounting and finance: Automating reconciliation, anomaly detection, and reporting.
• In credit and banking: Supporting fraud detection, risk modeling, and customer service.
• In business finance platforms: Enhancing cash flow insights, streamlining payments, and helping companies make smarter decisions.

At Flex, our role as engineers is to integrate AI thoughtfully into our workflows, so it supports these outcomes while staying transparent, secure, and user-focused.

How We Think About AI at Flex

Quality of Thinking Comes First

The most important part of any engineering project isn’t the model, the language, or the tools — it’s the thinking. Before writing a single line of code, we ask ourselves:

• What is the end user trying to do?
• Why are they doing it?
• How do they feel while doing it?

AI helps us answer these questions more thoroughly and quickly. Instead of rushing into execution, we can use AI as a first-principles partner by asking: “Does this really solve the user’s problem? Could the experience be simpler? Are we missing edge cases? What are problems with this approach?” The most important aspect is to solve the right problem. Solving the right problem saves significant time and energy!

AI as a Thought Partner

When we start building something new, like a monitoring system for high-performance APIs, we don’t just dive into the deep end. Instead, we’ll often sketch a one-pager or a technical spec that clearly outlines our goals and plans for the project (frequently with the help of AI). From there, we further utilize AIs capabilities for:

• Sharpening ideas: Asking AI to critique our goals or suggest alternative approaches.
• Structuring the plan: Breaking projects into phases, milestones, and execution steps.
• Parallel exploration: Running multiple AI prompts to compare different strategies at once.

This is where the power of AI in finance industry engineering really shines. AI doesn’t replace our thinking — it supercharges it.

We utilized this strategy heavily when we built out our transaction receipt policy feature. First, we defined our goals. We critiqued it until we found something we could both build out quickly and extend in the future with minimal hurdles. Then we broke up the work to be tackled by a combination of human and robot teammates.

AI as a Tool for Execution

Defining Clear, Granular Plans

Once we’ve shaped a strong idea, AI becomes a powerful execution partner. The key is giving it tight guidelines. Let’s walk through the steps using an example: In this case, we’ll touch on how our team built out Optical Character Recognition (OCR) support for our Bill Pay product. 

1. Define the goal in detail.
   
   • Our main goal was to build out the system as designed that allows users to simply upload a bill they wish to pay. From there, we parse all of the data from the document automatically and allow the user to confirm that data and the terms prior. Once confirmed, the payment can be completed.
2. Break it into small, verifiable tasks.
   
   • Take the designs and determine if there are any relevant workflows not incorporated.
   • Build out each page according to the designs using mock data.
   • Add document upload capability with relevant location, TTL, privacy configurations. 
   • Add parsing logic that extracts the data from the uploaded document.
   • Populate as many of the relevant fields as possible (ensuring only data from the document is used) and return an error if the document is unparseable or you can not continue.
   • Allow the user to confirm the values, select their terms, and continue with the transaction.
3. Use AI to generate a first draft.
   ‍
4. Use the above steps to define the technical spec and be sure to identify two things:
   
   • Where is the data coming from? 
   • Where is the data being stored?
5. Review and refine the draft with human judgment.
   
   • Include other team members as needed.
6. Repeat until the final product is where we want it to be.

In this  example, we made sure we verified outputs, ran tests, and ensured compliance with data integrity standards. AI allowed us to work in parallel on multiple components, all while keeping a tight leash to ensure nothing slipped through unchecked.

The Check-In Process

When it comes to utilizing AI, a structured feedback loop is critical. At Flex, our engineers:

• Ask AI to review its own outputs (“Does this code achieve the stated goal?”)
• Cross-verify with team knowledge
• Refine and re-run iterations until the final solution meets both technical and compliance standards

This disciplined cycle keeps AI as a helpful tool, rather than a potential risk factor. We also find that as AI improves, we are able to further simplify the process by codifying these refinement loops directly into the context using memory or some sort of rules file.

Traditional Engineering vs. AI-Enhanced Engineering at Flex

To further explain how our approach differs from a more manual process, here’s a comparison of how traditional engineering workflows differ from AI-enhanced engineering at Flex:

‍

AI in Finance: The Importance of Trust

Compliance and Consumer Trust

The finance industry is highly regulated. Unlike other sectors, financial platforms like Flex handle sensitive personal and financial data (PII). This makes trust and compliance non-negotiable in our processes and products.

When people ask (often with a hint of skepticism) how we use AI in our work on the engineering team at Flex, my answer is that AI is always most effective when paired with strong human oversight. At Flex, we maintain:

• Strict data access controls to prevent data leakage.
• Data integrity checks at every step.
• Regulatory compliance across all workflows, with special attention to those driven by AI-driven.
• Human verification of all AI outputs.

AI helps us move faster, but only when we protect consumer trust first.

Final Thoughts: AI as a Partner, Not a Replacement

At Flex, we see AI as more than an exciting trend — it’s a way to think better and build smarter. From brainstorming technical specs to executing products and functionalities and maintaining compliance, AI plays a critical role in how we engineer financial solutions.

But the most important thing to remember is this: AI is ahelpful partner, not a replacement. It’s here to sharpen our thinking, accelerate our execution, and help us build financial tools that business owners can trust.