A model can now pull a company's filings, populate a comparables table and draft an earnings summary in the time it takes an analyst to open a spreadsheet. For anyone training for an analyst career, the obvious question is whether the job will still exist.
That question is framed slightly wrong. The more useful one is what the job will be paid for.
Automation reduces the value of mechanical analysis. It increases the value of professional judgment. The analyst role is shifting from producing analysis to validating it, placing it in context and standing behind it.
This is not a comfortable shift for everyone. It rewards a different set of skills from the ones most junior analysts were hired for, and it changes how those skills are acquired.
The Automation Gradient: What Goes First, What Resists
Analyst work is not one task. It is a chain, and the links do not automate at the same speed. No source we could verify measures this gradient for financial analysts specifically, so what follows is a practitioner's reading of the work, not a statistic.
The first links to go are the ones with clear inputs and checkable outputs:
- Data collection: pulling financials, transcripts and market data into a usable form.
- Screening: filtering a universe of securities against defined criteria.
- First-draft models and summaries: a three-statement model built from reported figures, or a two-page note on a quarter's results.
The links that resist are the ones where the right answer depends on judgment, not retrieval. Setting assumptions is the clearest example: a model can extend a revenue trend, but deciding whether the trend should continue requires understanding the business, its competitors and management's incentives.
Cross-checking resists because it requires knowing what a wrong number looks like. Client accountability resists because it is not a task at all. It is a responsibility.
The practical implication is simple. The part of the job that used to take the most hours is the part automating fastest. The part that took the fewest hours, and carried the most weight, is the part that remains.
What the Evidence Actually Says
The discussion around AI in finance tends toward confident predictions. The verified evidence is narrower, and it is worth reading carefully.
In a December 2025 review, the CFA Institute Research and Policy Center summarised benchmarking research (Xu et al.) finding that autonomous AI agents completed only about 30% of complex, multi-step tasks without human help. The same review cited a Microsoft Research study of more than 200,000 Copilot interactions in which the model's actions diverged from what the user intended in roughly 40% of cases. Both findings concern general knowledge work, not analyst tasks specifically. Do not read them as automation rates for financial analysis.
What they do show is the shape of the problem. Current systems handle discrete steps well and long chains of dependent steps less reliably, and they often solve a slightly different problem from the one they were given. In finance, a slightly different problem is a wrong answer. That is the empirical case for treating human validation as a standing requirement, not a courtesy check.
Employer sentiment points in the same direction. In a CFA Institute survey of 200 investment industry representatives, fielded in February 2024, 85% of employers said the industry needs industry-wide standards and ethical guidelines for AI and generative AI, and 82% said the absence of standards is holding back faster adoption.
The sample is employers, not practitioners, and modest in size, so the percentages should not be stretched. The direction is still clear: firms are not waiting for better models. They are waiting for a way to trust them.
The Accountability Point: Someone Must Own the Number
This is a core concept. In investment work, a number is not just an output. It is a claim that someone is prepared to defend to a portfolio manager, a client, an investment committee or a regulator.
A model cannot sign off on a valuation. It cannot explain to a client why a recommendation changed, or accept responsibility when an assumption proves wrong. When a price target is questioned, the question goes to a person, and that person needs to understand every assumption beneath it well enough to defend or revise it.
Machines can produce the analysis. They cannot be accountable for it. That gap is where the analyst's value now sits.
The same CFA Institute survey found that 70% of employers consider training on AI regulatory compliance and risk-related skills preferred or essential, and 47% judged their own organisation unprepared for potential regulatory change on AI. Firms are signalling that the scarce skill is not operating the tools. It is knowing when their output can be relied on, and taking responsibility when it is.
CFA Institute's own 2025 publication on AI in asset management frames the practitioner's task in similar terms: balancing automation against human oversight, preserving governance and accountability, and treating AI as a supplement to professional judgment rather than a replacement for it.
Team Structure: Fewer Producers, More Validators
No verified data yet measures how analyst teams are changing in headcount or ratio, so this section is our view of the direction, not a measurement.
Traditional research and investment teams are shaped like a pyramid. A broad base of junior analysts produces models, comps and first drafts; a smaller layer of senior analysts reviews and refines; a portfolio manager or partner makes the call. When production becomes cheap, the base of that pyramid is the part under pressure. The review and decision layers are not.
The likely result is a flatter structure in which each analyst covers more ground, with software handling production and the analyst acting as editor, validator and owner of the conclusion. That is a more demanding job, not an easier one.
It also creates a real problem. Senior judgment has traditionally been built by years of junior production: you learn what a sound model looks like by building many flawed ones. If fewer people do that work, the industry has to find another way to develop the validators it now needs. Candidates who can demonstrate understanding without the traditional apprenticeship will have an advantage.
The Skills That Gain Value
If the job moves from producing to validating, three skills become more valuable.
Judgment. Choosing assumptions, recognising when a result is implausible, and deciding what matters in a set of numbers. Do not confuse this with intuition. Judgment in finance is structured: it rests on knowing how a business, a security and a market actually work.
Domain depth. Validation requires knowing more than the system being checked. An analyst who understands revenue recognition, the drivers of a sector's margins, or how duration behaves when rates move can spot a model's error in seconds. One who does not will approve it.
The market does not reward vague understanding, and automation makes that more true, not less.
Communication. When analysis is abundant, the scarce product is a clear conclusion someone can act on. Explaining why a number is right, where it might be wrong and what would change the view becomes a larger share of the job.
Workforce sentiment is mixed. In the CFA Institute survey, employers described their staff as curious (68%), anxious (60%) and resistant (48%).
Curiosity is the useful response. Anxiety and resistance both assume the job is standing still. It is not. It is moving toward the skills that were always the most valuable, and away from the ones that were merely the most time-consuming.
Where Structured Knowledge and the CFA Fit
If domain depth is what makes validation possible, the question becomes how to build it. This is where structured knowledge regains importance.
A validator needs a map of the whole territory: financial reporting, valuation, fixed income, derivatives, portfolio construction and ethics, and how they connect. Learning those areas piecemeal, on the job, as tasks happen to arise, leaves gaps. And gaps are exactly where an unchecked model error passes through.
The CFA Program is one of the most established ways to build that map. CFA Institute counts more than 200,000 charterholders across more than 160 markets, which makes the designation a widely understood signal of a common knowledge base.
It will not teach judgment by itself. No curriculum can. But it supplies the structured foundation that judgment is exercised on, and the ethics and professional standards component speaks directly to the accountability question that AI has made more pressing.
Be precise about what the credential does. The CFA improves your ability to evaluate analysis. It does not make mechanical skill more valuable, and it is not a defence against automation on its own. For a fuller view of where the designation pays, see whether the CFA is still worth it in 2026, and for the broader argument about judgment, why professional judgment still matters in an automated market.
What This Means for Your Career
The analyst role is not disappearing. It is being redefined around the parts of the work that cannot be delegated: owning the assumptions, checking the output and answering for the conclusion.
That redefinition has a clear consequence for anyone still building their skills. Time spent becoming faster at mechanical tasks is time invested in the part of the job that is shrinking. Time spent understanding why the numbers are what they are is invested in the part that is growing.
The market will not reward mechanical skill for much longer. It will reward the professional who can tell whether the machine is right.
Build judgment deliberately. Study the material deeply enough to explain it, check it and defend it. That is the skill set automation makes more valuable every year.