An analyst today can pull a decade of financials, build a discounted cash flow model and generate a draft investment memo in an afternoon. So can every competitor. The data is the same, the templates are the same, and increasingly the tools that summarise and model are the same.
That raises a question every serious finance professional should ask about their own career. When everyone has the same models and the same data, where does edge come from?
The answer argued here is uncomfortable for anyone who equates expertise with output. The market does not reward vague understanding, and it increasingly does not reward mechanical understanding either. What it rewards is judgment: knowing which model applies, when the data is misleading, and what the number leaves out.
When Analysis Is Commoditised, the Edge Moves
Mechanical analysis used to be scarce because it was slow. Building a model took days, and the analyst who built it faster and cleaner had an advantage. That advantage has been shrinking for years, and automation accelerates the trend.
The practical implication is simple. If a task can be specified precisely enough to be automated, its output becomes available to everyone at low cost. A valuation that any competent tool can produce does not differentiate the person who produced it.
What remains scarce is the layer above the output. Someone still has to decide whether the model is the right one, whether its inputs deserve belief, and whether its conclusion survives contact with what the model cannot see. That layer does not get cheaper when computation does. It becomes the part that matters.
Automation reduces the value of mechanical analysis. It increases the value of the judgment that decides when to trust it.
Judgment, Defined Without the Buzzword
"Judgment" is easy to praise and hard to pin down, which is why it often reads as filler. In finance, it can be defined concretely as three recurring decisions.
The first is model selection. A dividend discount model is a reasonable tool for a mature, stable payer and a poor one for a company that does not pay dividends or is reinvesting heavily. A single-stage growth model is elegant, and it becomes unstable when the assumed growth rate approaches the required return. Knowing which framework fits the asset is a decision no formula makes for you.
The second is assumption scepticism. Most model outputs are dominated by a few inputs. In many discounted cash flow models, the terminal value carries a large share of the result, which means a small change in the long-run growth assumption moves the valuation more than years of detailed forecasting. Judgment is knowing where the output is fragile and interrogating those inputs first.
The third is incentive awareness. Numbers are produced by people with reasons to present them in a certain light. Management guidance, adjusted earnings measures and sell-side forecasts all carry the incentives of whoever prepared them. A professional reads the number and the incentive together.
Each of these is a decision about the model, not a calculation inside it. That is the distinction that matters.
Why Judgment Resists Automation
Three features keep judgment tied to people: context, accountability and incentives.
Context is the information the model does not contain. A model knows the reported figures. It does not know that a key customer contract is being renegotiated, that a regulator's attention has shifted, or that a client's real constraint is a liquidity need they mentioned once in passing. Integrating that context is the work.
Accountability is the second feature, and it is structural. Someone signs the recommendation, answers to the client and carries the professional responsibility. CFA Institute's own framing reflects this. In a November 2025 press release on AI in the investment sector, Mona Naqvi, its Managing Director for Research, Advocacy and Standards, said AI "can supplement professional judgement", and the institute positioned transparency, accountability and ethical governance as the conditions for its use.
Its Research and Policy Center has published commentary in the same direction, describing AI as reshaping workflows while raising demand for human judgment and treating the combination of artificial and human intelligence as the foundation of ethical practice.
Incentives are the third. Clients, especially those with complex needs, want someone who understands their situation and can be held responsible for advice. The US Bureau of Labor Statistics makes a related point in its 2024 to 2034 projections: it expects robo-advisers to have only a mild effect on advisory employment, partly because clients with sophisticated planning needs are unlikely to trust automated advice. That is a projection, not a certainty, but it shows trust and context acting as real limits on automation.
The Spreadsheet Parallel: Tools Eliminate Tasks, Elevate Roles
The fear that software will replace analysts is not new. Spreadsheets automated a large part of the manual calculation and bookkeeping that once defined accounting work. The profession did not disappear. What changed was the mix of work: less time on arithmetic, more on analysis, review and advice.
The current data is consistent with that pattern. The Bureau of Labor Statistics projects employment of accountants and auditors to grow 5% from 2024 to 2034, faster than the 3% average for all occupations, with about 124,200 openings a year. This is a forward-looking projection, not a historical measurement of the spreadsheet era, and it should be read that way. But it is hard to reconcile with the idea that automating core tasks empties a profession.
The same pattern is visible in investment management. CFA Institute research on big data in the investment process finds that professionals are adopting AI and big data tools to complement traditional software such as Excel, not to replace it, and notes that its own 2019 baseline study found minimal AI use in the industry at that time. The toolkit is expanding. The person deciding what to do with the output remains.
The lesson is not that every role is safe. Tasks that are pure execution do shrink. The lesson is that the value migrates to the people who can do what the tool cannot: define the question, check the answer, and take responsibility for the decision.
How Judgment Is Actually Trained
If judgment is the scarce asset, the next question is how to build it. The common answer is seniority: judgment arrives with years. That is only partly true.
Experience without feedback repeats errors. An analyst who makes the same flawed assumption for a decade, without ever comparing forecasts with outcomes, has a decade of repetition, not a decade of judgment. What builds judgment is deliberate repetition with feedback: making a call, recording the reasoning, seeing the result and adjusting.
In practice, that means a few habits. Write down the assumption you are least sure of before you finalise a model. Compare your forecasts with what actually happened and look for patterns in the misses. When a tool produces an output, predict roughly what it should be before you look, and investigate when the two disagree.
The CFA Program has made applied practice a formal requirement. Candidates must complete at least one Practical Skills Module at each level, each taking roughly 10 to 20 hours, and failing to complete it by the time results are released voids the exam result. The same module cannot be repeated across levels, so candidates are pushed to broaden across programming, data science and AI, and analyst skills.
Tool fluency is part of this, not an alternative to it. Institutional Investor, reporting on a CFA Institute career survey, found that 3% of investment professionals said they were already proficient in AI and machine learning, while almost two-thirds said they were interested. Interest without proficiency is a gap. But proficiency without the underlying judgment is a different gap, and arguably the more dangerous one: it produces confident outputs that nobody is equipped to question.
Judgment is not what you know. It is what you do when the output looks right and is not.
What This Means for Candidates: Understanding Beats Coverage
For a CFA candidate, the implication is direct. Coverage, meaning having seen every reading and memorised every formula, is necessary and no longer sufficient. The formula is the part a tool can supply. Understanding why it works, and when it breaks, is the part it cannot.
Consider what that looks like in study terms. A candidate who has memorised the growth model can compute a value. A candidate who understands it knows the value explodes as growth approaches the required return, and treats a result like that as a warning, not an answer.
The first candidate passes a recall question. The second is building the judgment the profession pays for.
The same applies to Ethical and Professional Standards, which some candidates treat as memorisation. Its questions are judgment exercises: applying principles to ambiguous facts. That is the closest thing in the curriculum to the work described above.
This is also why practice questions should be used as feedback, not as a volume target. Each wrong answer is information about where understanding is thin. Our article on using practice questions properly sets out how to work that way, and how AI changes the role of the analyst looks at the same shift from the employer's side.
The durable career asset is not a faster route to the output. It is the ability to question it. That ability starts with mastering the fundamentals deeply enough to know when a number is wrong, and it compounds with every deliberate repetition. Build it now, while the stakes are an exam question rather than a client's capital.