Portfolio Analysis
How does AI portfolio analysis work?
Answer
AI portfolio analysis reads an investor's holdings, resolves every fund to the securities it actually holds, computes the portfolio's real exposures, compares them against thresholds appropriate to that investor, and explains the findings in language. Most of the work is deterministic arithmetic — look-through weights, concentration, overlap, allocation drift. The part that benefits from a model is the messy work either side of it: reconciling inconsistently named holdings into identified instruments, and turning a set of measurements into an explanation a person can act on.
What is arithmetic and what is judgement
This distinction is the most useful thing to understand about any system described as AI portfolio analysis, because it tells you which outputs you can check.
Concentration, overlap, sector weights and allocation drift are arithmetic. Given the same holdings and the same disclosures, any correct implementation produces the same numbers, and they can be verified. Nothing about them is probabilistic, and a system that produces different figures on two runs of the same portfolio has a defect, not a model.
Judgement is where the thresholds are set, how a finding is prioritised, and what is worth saying at all. That is where the system is making a choice, and it is where a reader should want the reasoning shown.
The five steps
- Ingest. Holdings arrive from a broker connection, a Consolidated Account Statement or an uploaded statement, in inconsistent formats with inconsistent names.
- Normalise. Each line is resolved to an identified instrument — ISIN, scheme code — rather than to the string printed on the statement. This is the step where language models genuinely help, because "HDFC Bank Ltd", "HDFCBANK" and "HDFC Bank Limited-EQ" are one company and no lookup table has every spelling.
- Look through. Every mutual fund and ETF is replaced by its disclosed portfolio, so the analysis runs on securities rather than on scheme names. Without this step none of the interesting findings are computable at all.
- Measure. Look-through weights are summed across the portfolio and concentration, overlap, sector and factor exposure and allocation drift are computed. Pure arithmetic, fully reproducible.
- Interpret and explain. Measurements are compared against thresholds appropriate to this investor, and what crosses one becomes a finding stated with the holdings and numbers behind it.
Why look-through is the whole game
A portfolio tracker treats a mutual fund as one line with a name and a value. Every question worth asking about a portfolio — how much is in one company, how much of it is in financials, whether two funds are doing the same thing — requires knowing what the funds hold.
An investor holding four equity funds appears to hold two hundred companies. Resolved to holdings, the same portfolio frequently has forty to fifty per cent of its value in ten companies. Both descriptions are of the same portfolio; only one of them is what the investor is actually exposed to.
What to ask of any AI portfolio analyser
- Does it look through funds to underlying holdings, or does it treat a fund as one line? Without look-through, its concentration and overlap figures cannot be correct.
- Does every finding name the holdings and the number behind it, or does it assert conclusions?
- Is the methodology written down — inputs, calculation, assumptions, limitations — or is the method the product?
- Does it distinguish an external fact from its own calculation from its own opinion?
- Does it take custody of assets or execute trades, and is it compensated by anything it recommends?
What it cannot do
It cannot forecast returns, and a system presenting a projection as a prediction is misrepresenting a simulation. It cannot know an investor's circumstances beyond what they provide — a job in the sector their portfolio is concentrated in, a liability nobody entered. And it cannot make the decision: what a portfolio should hold depends on what the person wants, which is not in the data.
How Vestmind does it
VestFolio follows exactly the five steps above, and publishes the method for each in the methodology library. Findings carry the holdings that produced them. Vestmind does not execute trades, does not take custody, and is not compensated by any instrument it analyses.
Related questions
Can AI analyse my mutual funds?
Yes, and the useful analysis is the one that looks through each scheme to the securities it holds rather than reading its name and category. Once funds are resolved to holdings, the portfolio's real concentration, its overlap between schemes, its sector exposure and its allocation drift are all computable — none of which can be seen from a list of fund names and current values.
Is AI portfolio analysis accurate?
The arithmetic is exactly as accurate as the holdings data it runs on, and it is reproducible: the same portfolio and the same disclosures produce the same numbers. The two real sources of error are stale fund disclosures — Indian schemes disclose holdings monthly, so an analysis reflects the last disclosure rather than today — and misidentified instruments at the normalisation step. Interpretation is judgement, and should be presented with its reasoning rather than as a fact.
Does an AI portfolio analyser need my broker password?
It should never need one. Broker integrations in India use authorised read-only connections rather than credential sharing, and a statement or Consolidated Account Statement upload needs no connection at all. Any system asking for a broker password or a trading PIN is asking for more access than analysis requires.
Is an AI portfolio analyser the same as a robo-advisor?
No. A robo-advisor allocates and manages money it takes custody of. An analyser reads a portfolio the investor holds and transacts themselves, and produces findings rather than trades. The difference matters for what each is compensated by and for who makes the decision.
Sources
- SEBI — Categorization and Rationalization of Mutual Fund Schemes — defines the scheme categories and what each may hold
- Securities and Exchange Board of India (SEBI)
How Vestmind does this
VestFolio is Vestmind's AI portfolio analysis product. It reads an Indian investor's holdings — stocks, mutual funds and ETFs — and reports what the portfolio is actually exposed to: concentration, mutual fund overlap, sector and factor tilts, asset allocation drift and the risks those create.