10 AI Skills Every Banking and Finance Professional Should Learn Before 2027

Two Analysts, Same Job Title

Picture two liquidity analysts at the same bank. Same job title, similar tenure, same morning meeting. One spends Tuesday afternoon manually pulling funding data from three systems into Excel, reformatting columns, writing a narrative, and emailing the result to the ALCO pack owner. The other has built a Python script that fetches the same data via API, flags outliers against LCR thresholds, and drafts the commentary section using a language model prompt she wrote and refined herself.

Over the next few years, the gap between those two analysts is likely to widen significantly. It probably will not be about who attended an AI awareness session. It will more likely be about who built a working combination of domain knowledge, data skills, and practical automation capability. That combination is already separating professionals in treasury, risk, and regulatory reporting. This article gives you a concrete framework for building it.

Why 'Learn AI' Is Not a Skill

When a hiring manager or a LinkedIn post tells you to "learn AI", they are giving you advice with roughly the same precision as "learn business". It describes a direction, not a destination.

AI is a collection of techniques, tools, and systems. Some are genuinely relevant to banking and finance professionals. Others are not, at least not yet. What matters is knowing which ones apply to your workflow, at what depth, and how they connect to your domain knowledge.

The framework below covers ten skills across a spectrum. Some you can start today with no coding background. Others require meaningful investment over months. All ten are grounded in what banking and finance teams actually do: liquidity reporting, regulatory submissions, credit analysis, model risk management, FTP, IRRBB monitoring, and the rest.

The 10 Skills: A Practical Framework for Banking Professionals

Skill 1: AI Literacy

This is the foundation, and it is genuinely accessible. AI literacy means understanding what large language models, machine learning classifiers, and generative AI tools can and cannot do. Not at a mathematical level. At a conceptual level that lets you evaluate whether a tool is appropriate for a given use case.

A treasury professional with AI literacy knows, for example, that a language model does not retrieve live data unless it is connected to a tool. They know that a model can confidently produce a wrong number. They understand the difference between a retrieval augmented system (one that pulls in relevant documents or data at query time to supplement the model's response) and a model relying purely on its training data. If you want a deeper look at how knowledge retrieval works in these systems, the GraphRAG explainer on the blog covers how knowledge graph retrieval works and when it is worth the complexity.

AI literacy takes weeks, not months. It is the prerequisite for everything else.

Skill 2: Context Engineering

Most practitioners have heard of prompt engineering. Context engineering is the more accurate and more useful frame. It is the practice of providing an AI model with the right information, in the right structure, to produce useful and reliable output.

In a banking context, this means learning how to structure a prompt that includes relevant regulatory context, the correct calculation methodology, the format you need, and the constraints the model should respect. A prompt that is well constructed for a PRA110 commentary section will produce something usable. A vague one will produce something that sounds plausible but contains errors a reviewer will catch.

This skill also includes knowing when to use a system prompt, how to chain instructions across a workflow, and how to test and iterate. These are learnable habits, not technical mysteries.

Skill 3: Data Literacy

Data literacy is the ability to understand, interrogate, and critically assess data: its source, its structure, its granularity, and its limitations. In banking, this matters enormously because a great deal of the data underpinning regulatory reporting and risk analysis is messy, inconsistently defined, or subject to reclassification.

A data literate professional can look at a liquidity report and ask: what is the run off rate applied to this deposit category, does it reflect the contractual maturity or the behavioural assumption, and is the underlying data from the right system of record? An AI tool cannot ask those questions on your behalf unless you have already taught it to.

Data literacy underpins every other technical skill in this list. Without it, code and AI tools produce confident nonsense.

Skill 4: SQL

SQL is the language of structured data, and structured data is the language of banking. General ledgers, treasury management systems, regulatory reporting databases, and risk data marts are all relational. If you can write SQL, you can access and interrogate those data sources directly, without waiting for a data team to produce a report.

A treasury analyst who can write a query to pull funding concentrations by counterparty and product type, filtered by maturity bucket, is operating at a different level of independence than one who cannot. That independence compounds over time.

SQL is also the most transferable data skill in finance. It is consistent across platforms, not especially difficult to learn to a working level, and immediately applicable. Start here if you have not already.

Skill 5: Python

Python is the second technical pillar. It handles what SQL cannot: calculation logic, data transformation, automation, API calls, visualisation, and integration with AI tools. For finance practitioners, you do not need to become a software engineer. You need to reach a level where you can write scripts that solve your own problems.

A realistic entry point is automating a manual reporting process. Pull data from a source, apply a calculation, format the output, and export it. The Python for strings guide on the blog explains how to work with text data in finance, which is a useful early reference once you are working with commentary or transaction descriptions.

The example below shows a simplified LCR calculation. The deposit categories and run off rates are illustrative only and do not reflect the full regulatory treatment under CRR2 or PRA rules. In particular, the "operational deposits" category has a specific regulatory definition, and the 25 percent run off rate under CRR2 applies only to the portion of balances above the operational deposit threshold. Treat this purely as a coding pattern, not as regulatory guidance.

import pandas as pd

# Illustrative LCR calculation using simplified buckets
# Run-off rates are illustrative only, not regulatory guidance.
# Categories and rates are simplified and do not reflect full CRR2 treatment.

data = {
    "category": [
        "Retail deposits (stable)",
        "Retail deposits (less stable)",
        "Wholesale (illustrative operational bucket)"
    ],
    "balance": [500, 300, 200],
    "run_off_rate": [0.05, 0.10, 0.25]
}

df = pd.DataFrame(data)
df["outflow"] = df["balance"] * df["run_off_rate"]

hqla = 120  # illustrative HQLA balance

total_outflow = df["outflow"].sum()
lcr = hqla / total_outflow

print(df[["category", "balance", "outflow"]])
print(f"\nTotal net outflow (illustrative): {total_outflow:.1f}")
print(f"Illustrative LCR: {lcr:.2%}")

This is a toy example. But the structure: fetch, calculate, present, is the pattern that underpins real automation work. Build fluency with that pattern and you can extend it to genuine use cases.

Skill 6: APIs and System Connectivity

An API is how software systems talk to each other. When you connect a Python script to a data source, or when you call a language model programmatically, you are using an API. Understanding how they work, at a conceptual level and at a basic practical level, is increasingly important.

In treasury and risk, this matters because modern infrastructure increasingly exposes data via APIs rather than file exports. The analyst who can make an API call to retrieve intraday position data, rather than waiting for a batch file, can build faster and more responsive tools.

You do not need to build APIs. You need to know how to use them: authentication, request structure, response handling, and error management. This is a natural next step after Python basics.

Skill 7: Workflow Automation

Workflow automation sits between scripting and full AI integration. Tools like Power Automate, n8n, or even well structured Python scripts with scheduling can handle the movement, transformation, and distribution of data without human intervention at each step.

For a regulatory reporting team, a simple automation that retrieves a data extract, checks it against expected ranges, flags exceptions, and sends a summary to the relevant owner is genuinely valuable. It does not require AI. It requires understanding the workflow well enough to codify it.

This skill is about process thinking as much as technical capability. You need to map what happens, in what order, with what logic at each decision point. That is a skill most finance professionals already possess. The technical layer is the easier part.

Skill 8: AI Agents

AI agents are systems that use a language model to reason through a task, decide which tools to use, execute those tools, and iterate based on the result. They are more capable than a single prompt, and more complex to build and govern.

In a finance context, an agent might be given the task of monitoring a set of positions against policy limits, retrieving the relevant data, identifying any breaches, and drafting a breach notification. Each of those steps involves a decision and a tool call.

This is genuinely emerging territory in banking. The governance challenges are real, and most firms are still working out where agents are appropriate. Understanding what agents are, how they work, and where the risks sit is valuable today, even if you are not building them. The AI and finance job assessment on the blog gives an honest look at the workforce implications.

Skill 9: AI Governance and Model Risk

Banks already have model risk management frameworks. AI introduces new categories of model, and new categories of risk. A language model used to draft regulatory commentary behaves differently from a credit scoring model, and whether it falls within your firm's definition of a "model" under frameworks such as PRA SS1/23 will depend on how the firm classifies it and what role the output plays in a regulated process. That classification question matters before you build governance around it.

What is clear is that AI tools used in material processes require oversight: documentation of purpose and limitations, human review before outputs enter regulated workflows, and some form of ongoing monitoring. AI governance skills include understanding those principles, knowing what questions to ask when a vendor presents an AI tool, being able to contribute to an AI use case review, and knowing where human sign off is non-negotiable.

This skill set is particularly valuable for senior professionals, risk teams, and anyone operating in a second or third line function. It is also increasingly appearing in job specifications.

Skill 10: Critical Thinking and Domain Expertise

This is listed last, but it is the most important. The ability to generate an answer is becoming commoditised. The ability to recognise a wrong answer in a banking context is not.

A language model can produce a plausible sounding NSFR commentary. It cannot tell you whether the stable funding assumptions are consistent with your firm's behavioural estimates, whether the narrative contradicts a disclosure made in last quarter's ILAAP, or whether the number reflects a reclassification that accounting made last month. You can.

Domain expertise becomes more valuable as AI proliferates, not less, because someone must determine whether the output is correct. That person needs to know what correct looks like.

The Banking Skill Stack for AI

Think of these ten skills as a stack with three layers.

The base layer, AI literacy, context engineering, data literacy, and domain expertise, is accessible to every practitioner regardless of technical background. These are the skills you can start building this month.

The middle layer, SQL, Python, workflow automation, and APIs, requires deliberate practice over several months. You will write bad code. You will break things. That is the process.

The top layer, AI agents and AI governance, builds on everything below. You cannot govern what you do not understand. You cannot build agents without the technical and conceptual layers in place.

Do You Still Need to Learn to Code?

Yes. But the bar has changed.

AI tools can write code. They write it faster than most humans, and for many standard tasks the output is adequate. What they cannot do is tell you whether the code is solving the right problem, whether the calculation logic reflects the correct regulatory treatment, or whether the output should be trusted.

The ability to read, understand, and modify code is now more important than the ability to write it from scratch. You need enough fluency to direct the tool, evaluate the output, and debug when something is wrong. That is a lower barrier than it was five years ago, but it is still a real one.

What Happens to Excel?

Excel does not disappear. It changes role.

Excel remains the fastest tool for ad hoc analysis, scenario modelling, and communicating numbers to people who do not use code. It is also deeply embedded in governance processes, model documentation, and ALCO packs. That is not changing rapidly.

What is changing is the boundary of what Excel is the right tool for. Large data sets, automated pipelines, and AI integration are better handled in code. The professionals who thrive will be fluent in both, using Excel for what it does well and Python or SQL for what they do better. The Tableau and AI article on the blog asks whether specialised visualisation tools still earn their place in a finance team using AI.

Three Levels: User, Practitioner, Architect

Not everyone needs the same depth. A useful frame is:

User: Can use AI tools effectively, writes good prompts, understands limitations, applies domain judgement to outputs. No coding required.

Practitioner: Can write SQL and basic Python, build simple automations, call APIs, and integrate AI tools into their own workflows. This is the level most banking professionals should target.

Architect: Can design, build, and govern AI systems. Understands agentic workflows, model risk management, and system integration deeply. Fewer people need this, but those who have it are in significant demand.

Most of the advice in this article targets the practitioner level, because that is where the career differentiation sits for the majority of treasury, risk, and finance professionals over the next few years.

A 12 Month Learning Roadmap

This is structured for someone learning alongside a full time role, which means realistic time blocks of roughly three to five hours per week.

Months 1 to 2: Foundation

Focus on AI literacy and data literacy. Read broadly about how language models work. Take a short structured course. Pick a repetitive task in your current role and map out exactly what data it uses, where it comes from, and what the logic is. That mapping exercise is more valuable than it sounds.

Months 3 to 4: SQL

Work through a structured SQL course using banking relevant examples where possible. Write queries against a publicly available financial dataset. Set yourself a small project: replicate a report you currently receive manually using a SQL query instead.

Months 5 to 7: Python

Start with the basics, data types, loops, functions, and pandas for data handling. Then pick one real problem from your role and build a script that solves part of it. It will have rough edges. Refine it as you go.

Months 8 to 9: Context Engineering and Automation

Combine your Python skills with an AI API. Build a script that takes a data output and generates a draft commentary. Separately, automate one recurring workflow using a scheduling tool or a simple pipeline.

Months 10 to 12: Governance and Agents

Study your firm's model risk framework and think about how it applies to the AI tools you are now using. Read about AI agents. Understand the risks. If your firm has an AI governance process, engage with it. You now have enough context to contribute meaningfully.

Where to Start Based on Your Role

The Academy at The Industry Portal has structured courses across Python, finance, and treasury that are worth looking at if you want guided learning with a banking context. Browse the course catalogue if you want a specific course, or the learning paths if you want a guided journey with a shareable certificate at the end.

The Skills AI Cannot Replace

Regulatory judgement, stakeholder management, commercial awareness, and ethical reasoning do not sit in a model. They sit in a professional who has built them over years.

A language model cannot tell you whether a proposed FTP methodology is consistent with the firm's funding strategy. It cannot read the room in an ALCO meeting. It cannot make a judgement call on whether a model assumption is defensible to the PRA given the current supervisory climate. It cannot take responsibility for a submission.

These capabilities become more valuable as AI handles more of the execution, precisely because they are the layer of oversight that regulated institutions require. The professional who combines them with technical fluency is the one who will be genuinely hard to replace.

If you are building towards a career in senior treasury, risk, or finance leadership, these are the areas where continued investment pays the highest return. Technical skills get you to the table. Judgement is what keeps you there.

Where to Go from Here

Pick one skill from the base layer and start this week, not this quarter. If you have no SQL experience, open a free SQL course today and write your first query by Friday. If you have not used an AI tool in a work context yet, draft your next piece of analysis commentary using one and then edit it critically. That editing process teaches you more than reading about AI ever will.

Structured Learning Options

If you want to go deeper with structured support, the Pro membership at The Industry Portal gives you access to member only courses, downloadable resources, and a practitioner community where these questions get worked through in real time. The newsletter is worth subscribing to as a free starting point to keep up with what is changing and what is worth your time.

The combination of domain expertise and practical technical capability is the differentiator. You already have the domain side. The technical layer is learnable. Start small, build consistently, and apply everything to real problems in your actual role. That is the only roadmap that works.