The apprenticeship is the thing that broke
The headline version of this story is that AI is coming for financial analysts. The accurate version is more specific and more uncomfortable. What automated is data gathering and consolidation, routine model building, comps tables, standard reporting packs, variance commentary and first-draft memos. That is not a random slice of the profession. It is precisely the junior job description. The full role assessment is in will AI replace financial analysts.
What survives is the judgement layer: choosing the assumptions, which is where a model is actually made, recognising when an output is wrong before anyone acts on it, understanding the business behind the numbers, persuading decision-makers, and carrying responsibility for the call.
The problem is how you learn those things. You develop a feel for when a number looks wrong by pulling ten thousand numbers yourself. Automating the mechanical work does not remove the need for the judgement it produced; it removes the paid apprenticeship that produced it. If you are senior, this is leverage. If you are two years in, the ladder has fewer rungs than it had when the people above you climbed it, and that is worth taking seriously rather than waiting out.
You are much closer to this than you think
Analysts routinely describe themselves as non-technical while doing something structurally identical to programming.
A model is a dependency graph of pure functions. Each cell derives from other cells, an input change propagates through the whole structure, and your job is to keep that propagation correct. That is the mental model that trips up most beginners when they meet state and data flow, and you have been living in it for years.
You already do the hard discipline. Handling edge cases, tracing why a figure changed, making an amendment without silently breaking something three tabs away, checking a result against something independent before it goes to a committee. That is engineering practice, learned in a tool that gave you none of the support engineers take for granted.
And you have the instinct that matters most in the AI era: you do not trust an output you have not checked. Finance beats that into people, because a wrong number that reaches a decision is a career event. In a market where the common failure of new developers is accepting plausible generated code at face value, arriving with a professional verification reflex is a genuine advantage. It is the same argument made in AI is the new Excel, and finance people tend to get it faster than anyone.
Three lanes, and the best-paid one keeps you in finance
Lane one: stay in finance, add the technical half. The highest-ceiling option and the one most analysts overlook. Quantitative work, fintech product and engineering, and data roles inside banks and insurers all pay a premium because they need both halves, and the finance half is the one a technology team cannot easily hire. Python and SQL alone change what you can do; inside most finance functions the person who builds the automation becomes considerably more valuable than the person operating it.
Lane two: data engineering and analytics. The natural adjacency, since you already live in structured data. It pays above analysis at every level in Malaysia and has more open roles, and the comparison between the three data jobs is laid out in the data engineer guide and the data scientist guide.
Lane three: the full switch. General software development, most open roles, entering at roughly RM 6,000 to RM 9,000 a month at the AI-capable junior level (the data). Be honest with yourself here: if you are already mid-level at a bank, this specific lane is probably a pay cut, which is exactly why lanes one and two matter more for you than they do for most switchers. They price your existing experience instead of discarding it. Accountants face a near-identical decision, and the accountant switch guide is worth reading alongside this one.
The plan, and what to build
Six to twelve months of consistent part-time study, 10 to 15 protected hours a week, alongside the job. The scheduling is what defeats people, not the difficulty, and finance hours make that harder than average, so plan around the reporting calendar rather than pretending it does not exist.
Build the things your own desk needed and never got. Take the model you rebuild every month and turn it into something that runs on a schedule. Build the data pull that currently costs you a morning. Build the reconciliation checker, the reporting automation, the alerting that tells you a figure moved before somebody else notices. These interview far better than a generic to-do app, and unlike most portfolio projects they may pay for themselves at your current job before you have even left it.
Every lane rests on the same foundation: being able to structure a problem, build, debug, ship and operate real software, with AI fluency and the judgement to catch what the model gets wrong. That last clause is the part finance already gave you. None of that learning is wasted whichever lane you take, which is why it is worth starting before you have decided which one it will be. The cheapest way to find out whether it suits you is the free trial: a week of real projects, a live instructor session, no card.