Investment Research Software Checklist for Indian Firms

Investment research software should be evaluated by the evidence it preserves, not by the number of charts in the demo. For an Indian PMS, AIF, family office, investment adviser or research team, the important question is whether the platform can carry a company from source document to analysis to decision to continuous monitoring without turning the process into a black box.

This checklist is deliberately vendor-neutral. Use it with one company your team already knows well. A familiar company makes it much easier to spot a wrong period, missing qualification, invented figure or impressive answer that never addressed the actual question.

1. Audience and job to be done

First establish which job the tool is meant to perform.

• Retail stock discovery?
• Professional company research?
• Quantitative screening and backtesting?
• Financial modelling?
• Portfolio accounting and reporting?
• Mutual-fund due diligence?
• Continuous monitoring?
• Client communication?

No platform should receive credit for solving a job it never claims to solve. A family-office accounting system and a listed-equity research system can both be excellent while doing almost nothing alike.

Pass condition: the vendor defines its user, scope and limitations clearly.

2. Source coverage and traceability

Ask which primary sources sit behind the product:

• exchange filings;
• annual reports and quarterly results;
• investor presentations;
• concall transcripts;
• shareholding disclosures;
• mutual-fund factsheets and holdings;
• official macro or alternative-data releases.

Then test whether a figure or claim opens the supporting document, page, period and context. A link to a 300-page annual report is better than no citation, but it is not the same as claim-level evidence.

Pass condition: material facts can be traced to a specific source and date.

3. Entity, period, basis and units

Many errors look reasonable because the number itself exists somewhere. It simply belongs to the wrong frame.

Check whether the platform distinguishes:

• consolidated from standalone;
• quarter from year-to-date and full year;
• current period from restated comparative;
• rupees from lakhs, crores or another currency;
• attributable profit from a broader profit line;
• a ratio from a percentage-point change.

Pass condition: the answer exposes enough metadata to identify exactly what the number represents.

4. Point-in-time history

Ask the vendor to recreate a screen or company view on a past date.

The test is not whether the platform possesses old observations. It is whether it knows when each observation became available. A result for March may be published in May. A later filing may restate the March number. A historical test should not use either before the appropriate availability date.

Pass condition: past analysis uses information that could actually have been known then.

5. Deterministic calculations

Language models should not be the hidden calculator for metrics used in screens, rankings or valuations.

Ask:

• Is the formula documented?
• Which inputs were used?
• How are missing inputs treated?
• Is the metric applicable to this sector?
• Can the same inputs reproduce the same result?

Pass condition: a derived value behaves like a calculation, not a generated opinion.

6. Honest unavailable states

Ask a question whose answer is absent from the available documents. A safe research system should say it cannot support the answer.

Also look for silent substitutions: a missing consolidated result replaced by standalone data, a missing dividend replaced by zero, or a bank forced through an industrial-company formula.

Pass condition: the product fails closed when evidence or required inputs are missing.

7. Quantitative screening and scorecards

Screens and scorecards should expose:

• universe and date;
• eligibility gates;
• factor definitions;
• peer groups;
• weights and transformations;
• missing-data policy;
• ranks and tie treatment;
• version history;
• human overrides.

The result should be reproducible. A score that changes because an AI agent “reconsidered” the company is not a governed factor model.

Pass condition: the same version and data produce the same result.

8. Qualitative research with evidence

AI can compare transcripts, extract guidance, map risks and draft research notes. The platform should distinguish evidence from interpretation.

A useful output shows:

• the exact management statement;
• when it was made;
• whether it was numerical, directional or conditional;
• what changed later;
• the documents supporting the conclusion;
• any important counter-evidence.

Pass condition: the reader can inspect why the system reached a conclusion and where uncertainty remains.

9. Models and assumption ownership

The platform should separate reported history from analyst assumptions.

Ask whether:

• historical data can be refreshed without overwriting assumptions;
• formulas and scenarios are inspectable;
• assumptions have owners and dates;
• the system identifies what changed after a new result;
• Excel or Python can remain part of the workflow.

Pass condition: the analyst can explain every forecast input without saying “the AI chose it.”

10. Portfolio context

Research rarely ends at the company.

Check whether the system can connect a new idea or event with:

• existing holdings and weights;
• sector and factor concentration;
• liquidity constraints;
• overlapping exposures through funds;
• portfolio-specific monitoring rules.

Pass condition: the platform can show why the same company may matter differently to two portfolios without issuing an unexplained allocation instruction.

11. Continuous monitoring

Price alerts are not thesis monitoring.

Useful monitoring connects new filings, calls, guidance, KPIs, ownership and forensic events with company-specific conditions defined by the team. It should route important exceptions and suppress repetitive noise.

Pass condition: an alert explains which research assumption or rule may require review.

12. Exportability and verification

Ask the vendor to export its work.

A serious export should preserve enough structure to inspect:

• source inputs;
• formulas;
• filters and gates;
• factor scores and ranks;
• included and rejected companies;
• missing data;
• lineage and run metadata.

PDF is useful for communication. Excel is stronger for verification because the analyst can recompute and challenge the work.

Pass condition: the software can show its working outside its own interface.

13. Security, privacy and permissions

Before loading internal notes, portfolios or client information, ask:

• what data is stored;
• who can access it;
• whether accounts and firms are isolated;
• how permissions and exports are controlled;
• whether external models receive private content;
• what the retention and deletion policy is;
• what audit logs are available.

The right answer depends on the firm’s own policies and regulatory advice. The wrong answer is ambiguity.

Pass condition: the vendor can explain the complete data path in plain language.

14. Integration and exit risk

No tool exists alone. Establish how it fits with spreadsheets, document storage, portfolio systems and existing data vendors.

Also ask how the firm retrieves its data and research if it leaves. A product that creates institutional memory should not make that memory impossible to export.

Pass condition: the platform improves the workflow without holding the firm’s research hostage.

The 45-minute vendor test

Use this sequence in every demo:

1. Open a company the team knows.
2. Retrieve one reported figure and inspect the source.
3. Reproduce one calculated ratio.
4. Compare two concalls and trace a changed claim.
5. Run one historical screen.
6. Change a scorecard rule and inspect the effect.
7. Connect a new event to a portfolio guidepost.
8. Ask an unsupported question.
9. Export the result to Excel.
10. Give the export to an analyst who did not attend the demo.

If that analyst cannot reconstruct what happened, the workflow is not yet verifiable.

Where Altys fits

Altys is a source-linked, point-in-time research and monitoring system for Indian investment teams. It connects financials, filings, concalls, management guidance, ownership, factors, screening, scorecards, modelling, portfolio context and alerts.

The design principle is calculated, cited and exportable. Strategy screens and scorecards can be exported as formula-native Excel workbooks; GenGrid and analytical tables can be exported for independent checking. Unsupported evidence stays unavailable rather than being filled with a plausible guess.

That does not make Altys the right product for every investor. It is built for teams that need a repeatable professional research process, not tips, execution or family accounting. Start with the Altys solutions map or the new-firm research stack.