What AI Tools Do Portfolio Managers in India Actually Need?

The best AI tool for an Indian portfolio manager is not the chatbot with the longest answer. It is the system that helps the desk move from evidence to decision to monitoring without losing the source, the calculation, the historical context or the reason a position entered the book.

A general language model can be an excellent reader and drafting partner. A professional research stack has to do more. It must know which financial figure belongs to which entity and period, preserve what was knowable on a past date, calculate ratios consistently, connect company research with portfolio exposure, and keep watching the assumptions after the investment committee has moved on.

That distinction matters because portfolio management is not one research task. It is a repeating operating loop.

The seven jobs in a portfolio manager’s AI stack

Job	What the system should do	What still belongs to people
Discover	Screen a defined universe using repeatable rules	Decide which questions and constraints matter
Understand	Organise filings, concalls, KPIs, segments and guidance	Form a view of the business and industry
Model	Prepare sourced history and test arithmetic	Choose drivers, assumptions and scenarios
Compare	Run scorecards and peer analysis consistently	Decide whether the comparison is economically meaningful
Decide	Assemble evidence, risks and portfolio impact	Approve, reject, size and record the decision
Monitor	Watch company-specific guideposts and disclosures	Judge whether a change is material to the thesis
Learn	Compare forecasts and decisions with outcomes	Change the process when the evidence supports it

Tools that cover only the first two rows can save reading time. Tools that connect all seven can change how much research a lean investment team can sustain.

1. A sourced Indian financial-data layer

The first requirement is not AI. It is clean, dated and appropriately classified data.

Indian listed-company research has recurring traps: standalone and consolidated results, financial-year and calendar-year periods, restated comparatives, bonus and split adjustments, bank-specific financial identities, changing segment definitions and metrics that do not apply equally across sectors.

If the data layer gets these wrong, a more capable language model simply explains the wrong number more persuasively.

The practical test is traceability. A portfolio manager should be able to ask:

• Which filing supplied this figure?
• Is it standalone or consolidated?
• What period does it represent?
• When did it become available?
• Was it reported directly or calculated?
• What happens when the necessary input is unavailable?

The safe answer to the last question is often a blank or an explicit unavailable state. Absence should not silently become zero.

2. Screening that preserves the rule

A screen should express the investment process, not merely return a list.

That means preserving the universe, liquidity gates, metric definitions, exclusions, ranking rules and date. If a portfolio manager says, “profitable mid-caps with improving cash conversion, acceptable leverage and reasonable valuation,” the system should turn that into inspectable conditions rather than an opaque AI score.

The distinction becomes critical in a backtest. A rule evaluated on today’s cleaned history may include data that was filed, restated or corrected after the historical portfolio-formation date. Point-in-time data prevents that look-ahead.

The language model can make the rule easier to express. Deterministic code should decide which companies actually pass it.

3. Company research that goes beyond summarisation

Summarising a concall is useful. Portfolio managers need the comparison behind the summary:

• What changed from the previous call?
• What did management originally guide?
• Which KPI or segment moved against the thesis?
• Did cash conversion support the reported profit?
• Did ownership, pledging, the auditor or a rating action change?
• Does the evidence contradict something in the original investment memo?

This is where an AI research assistant becomes a research system. It does not merely compress a document. It attaches the new document to the company history and the questions the desk already cares about.

4. Financial models that remain inspectable

AI can accelerate the mechanical parts of modelling: finding source lines, spreading history, checking links and preparing scenario scaffolding. The forecast assumptions still need an owner.

A useful workflow separates three layers:

1. Reported facts from primary documents.
2. Calculated history produced through explicit formulas.
3. Forecast assumptions chosen and approved by the analyst or portfolio manager.

When these layers blur, it becomes impossible to explain whether a surprising output came from the company, the software or the analyst.

Excel remains important here. The point of connecting AI with spreadsheets is not to eliminate Excel. It is to stop spending analyst hours on transcription while preserving a familiar environment in which every formula can be inspected.

5. Scorecards with visible definitions

Many firms use scorecards to turn an investment philosophy into a consistent company review. A score might combine quality, growth, valuation, momentum, governance or company-specific factors.

The danger is false precision. A score of 82 is meaningless unless the team can inspect:

• the factor definitions;
• the applicable peer group;
• missing-data treatment;
• weights and gates;
• point-in-time inputs;
• any human override;
• how the score changed.

AI can help draft a scorecard from a written philosophy. The published scorecard should still behave like a model: deterministic, versioned and exportable.

6. Portfolio context, not isolated stock answers

A company can look attractive on its own and still be the wrong addition to a portfolio.

The portfolio manager needs to see position size, sector concentration, factor exposure, liquidity, correlation and overlapping business sensitivities. A new lender may deepen an existing credit-cycle bet. A defensive consumer company may add more of a valuation factor the book already owns. A mutual fund may duplicate direct holdings.

An AI answer about a company becomes more useful when it understands this surrounding context. The system should explain the exposure; it should not make the allocation decision invisibly.

7. Monitoring that remembers the thesis

Most of a portfolio position’s life is spent being held. Yet many research tools are optimised for the moment before purchase.

Useful monitoring starts with a small set of company-specific guideposts: the revenue driver, margin band, asset-quality measure, order flow, working-capital condition, management promise or forensic threshold that could genuinely change the view.

New information should then be routed against those guideposts. The result is not “good news” or “bad news.” It is a reason to review a particular assumption, model line or portfolio exposure.

This is how a team can monitor 100 companies without pretending to read everything.

A due-diligence test for any AI research product

Take one Indian company the team knows well and ask the vendor to do the following:

1. retrieve a specific historical financial figure and open the source;
2. explain the period, basis and units;
3. show the formula behind a derived metric;
4. answer an unsupported question with an honest unavailable state;
5. recreate a historical screen using only information known then;
6. compare current results with the original guidance;
7. export the screen, scorecard or analysis so the team can independently inspect it;
8. connect a company change with a real portfolio rule or thesis condition.

A polished memo is no longer a sufficient demo. The evidence chain is the product.

Where Altys fits

Altys for portfolio managers is an India-first research and monitoring system for PMS firms, AIFs, family offices, investment advisers and research teams.

The platform connects point-in-time financial data, filings, concalls, management guidance, ownership, forensic signals, factor scorecards, strategy testing, financial models, portfolio context and company alerts. Material claims link back to evidence. Calculations remain deterministic. Core analytical workflows—including screens and scorecards—can be exported to Excel so the desk can verify what the software did.

The positioning is deliberately narrower than “AI manages your portfolio.” Altys prepares the data, preserves the process, watches the evidence and helps the team move faster. The portfolio manager remains responsible for the assumptions, exceptions and capital-allocation decision.

For a broader comparison, see the best AI research tools for Indian stocks. If the firm is still assembling its first research operation, continue with how to set up an equity research desk in India.