Earnings Call Summaries vs Guidance Tracking in India

A summary can tell you what management said this quarter. To check what happened to a target from last year, you need the earlier statement as well.

Imagine twenty holdings, each with a thirty-page transcript: that is 600 pages before any cross-quarter comparison. This is an illustrative reading load, not a measured average. Summaries can reduce it. But compression and reconciliation remain different jobs, even when the same tool offers both.

An earnings call summary organises the current document: results, new guidance, risks and questions. Guidance tracking connects a dated target to later updates and reported outcomes. A current transcript can include historical context, but it cannot establish what was omitted from an earlier document you have not supplied.

Inve's production snapshot of 4 September 2026 contained 24,328 commitments across 1,648 Indian companies, drawn from 8,220 parsed calls across 1,654 companies. Of the 16,842 commitments with a verdict other than new, 2,271 were classified ghosted: 13.5% of those 16,842 eligible commitments, or 9.3% of all 24,328 tracked commitments. The remaining 7,486 were new. In the same snapshot, 1,000 of 1,648 companies had at least one such classification — 60.7% of those 1,648 companies.

Eligible does not mean terminally resolved. And ghosted means the parser did not identify the follow-up, not that the company necessarily failed to disclose it anywhere. The distinction matters enough to test on a real false positive below.

The company examples are about reading disclosures, not recommendations to buy, sell or hold their shares.

What a single-call summary does well

For applicable post-earnings and quarterly calls, SEBI LODR Regulation 46(2)(oa) requires transcripts on the company's website, with simultaneous submission to recognised exchanges, within five working days of the call's conclusion. This is a disclosure rule for calls that take place; it should not be read as requiring every company to host a quarterly call.

A useful earnings call summary helps you locate four kinds of information:

• The current operating results and the definitions management used.
• New targets, their periods and the speaker's wording.
• Analyst questions and how directly management addressed them.
• Risks, adjustments and qualifications that affect the headline figures.

That is enough to identify what deserves closer reading. It is not enough to conclude that every older target was met merely because the current quarter sounds reassuring.

These are evidence requirements, not claims that every AI summary behaves identically. A researcher can do the reconciliation manually, or give a tool the relevant earlier documents. Merely changing the tool's name does not supply missing evidence.

Welspun: a headline result does not close every guidance row

Welspun Corp provides a useful example of why consolidated results and narrower commitments deserve separate rows. Inve's guidance history records FY26 EBITDA guidance of ₹2,200 crore on the Q2 FY26 call, against ₹1,186 crore reported for H1. The Q3 FY26 update records ₹1,831 crore for nine months. The Q4 FY26 transcript reports full-year EBITDA around ₹2,370 crore and ROCE around 22.3%.

That EBITDA result exceeded the stated target. There is no need to diminish it to make the case for further reading.

The older ledger contains different questions, however. Its Q1 FY25 DI-pipe capacity target was an increase from 500,000 to 600,000 tonnes by the end of Q3 FY25. The parsed history does not identify confirmation against that deadline and labels the row ghosted. A separate US annual-volume target of 250,000–300,000 tonnes is also classified ghosted, with later combined India-and-US disclosures recorded as insufficient to verify that standalone figure (Inve data, commitments 19452 and 19453).

Those labels identify checks to perform. They do not prove the capacity was never installed or the volume was missed. Look for plant disclosures, segment tables and presentations before reaching either conclusion. The Welspun concall archive is a starting point for that source search.

The later Q1 FY27 call reports net cash around ₹2,336 crore and discusses a roughly $300-per-tonne normalised US EBITDA guide. It also describes muted domestic demand and overcapacity in DI pipes. Cash, profitability per tonne and a plant's capacity deadline are different facts. Success on the first two cannot resolve the third by implication.

This is where the ledger earns its place beside the summary: it helps keep the question specific when the latest headline is favourable.

Infosys: correct the quarters before interpreting the pattern

Infosys illustrates the usefulness of a guidance walk — and the risk of copying database quarter labels without checking the filing.

Its FY26 constant-currency revenue-growth guidance moved as follows:

The first four ranges match the dated sequence in Inve's corpus reference. The initial FY26 guidance was issued in April 2025, and the January 2026 call raised it to 3%–3.5%. Some local extracted history labels are one quarter early; the actual call dates and company documents control.

That correction changes the story from a vague sequence of upgrades into a checkable annual cycle. It does not prove that the next opening range will be conservative or that upward revisions predict an attractive share-price return.

The tracker can also be wrong about silence

The local Infosys guidance row for FY26 free-cash-flow conversion is marked ghosted, with a note saying the above-100%-of-net-profit target was not explicitly addressed in Q4 remarks.

The actual Q4 FY26 earnings-call transcript contradicts that classification. On printed page 4, CFO Jayesh Sanghrajka reported adjusted free cash flow of $3.5 billion for FY26 and adjusted FCF conversion of 106% of net profit. Those figures exclude the specified Labour Code and income-tax-refund effects. The same call reported FY26 CC revenue growth of 3.1% and adjusted operating margin of 21.0%.

The FY26 annual-report highlights separately show FCF conversion of 112.3% of net profit on the report's stated basis. The two conversion figures must retain their adjustment labels; neither supports a claim that no cash-conversion update was disclosed.

That is a real lesson in guidance tracking: compare the target and outcome on the same basis, and let the primary document override the extracted label. A tracker is useful because it points you to the disputed line, not because its verdict is beyond challenge. The Infosys guidance ledger requires this same source check.

A genuine unanswered question is narrower than an accusation

There was also a separate disclosure boundary. At the Q4 FY26 media conference, Chandra R. Srikanth asked about AI-services revenue. Salil Parekh replied: “Is it 10% or 50%, we are not sharing the number.” The media-conference transcript records that exchange on printed page 10.

This was not the analyst earnings call. Nor should it be stitched to a reply to a different journalist and presented as one continuous answer. The evidence establishes that a specific requested number was withheld in that exchange. It does not establish weak AI performance or dishonest reporting; competitive reasons for withholding a breakdown remain possible.

Repeated questions need the full answer too

Finolex Industries reported roughly ₹2,430 crore of net cash at December 2025. Across the five calls from Q3 FY25 through Q3 FY26, analysts repeatedly asked about the deployment of surplus cash without receiving a firm timetable for the remaining balance.

But the Q3 FY26 transcript also records references to planned capex and an earlier dividend. Omitting those facts would turn an unresolved allocation question into an unfair claim that nothing had been invested or returned.

Inve's September snapshot recorded 609 persistent-question dodges across 452 companies, defined as topics deflected on two or more consecutive calls. This is a model-derived finding at medium confidence, not an externally adjudicated governance finding. The guide to Q&A evasions is most useful when read as a way to choose follow-up questions.

How Inve combines summaries and guidance history

Within a four-layer research stack, the summary and ledger sit beside the financial statements and footnotes. They also serve different steps of the post-screener review.

Use Concall AI to locate current results, extracted guidance and Q&A exchanges. Use the Promise Tracker for the original target, subsequent updates and persistent-question history. Detailed Indian company ledgers require Pro access. Inve is describing its own product here; the primary-source corrections above are part of the standard against which that product should be judged.

A practical sequence is:

1. Read the current summary and identify the metric most important to the investment case.
2. Retrieve its earlier target and deadline even if the current quarter looks strong.
3. Read the source exchange behind a missed, revised or silent row.
4. Check the results release, presentation or annual report for the outcome, preserving scope and adjustments.
5. Record what remains unresolved and what disclosure could settle it next quarter.

The full guidance-verification checklist expands those steps. Treat any twenty-minute allocation as an illustrative triage budget, not a claim that a complex business can be fully reviewed in that time.

Where both tools can mislead you

The examples span roughly two years, but company coverage is uneven. The September snapshot averaged 4.18 distinct commitment birth quarters per company. Four or five quarters are not two years, and neither window establishes performance across a full credit or industrial cycle.

The corpus combines gemini-2.5-pro extraction across 1,476 company runs and gemini-3.1-pro-preview across 179, without a common holdout comparison. Errors can affect dates, units, business scope and whether an update was recognised. The Infosys false positive demonstrates why this is a substantive limitation.

Finally, a missed target is not automatically bad faith, and a favourable target history is not a valuation. Customer delays, accounting changes and external shocks can all matter. Keep the original question visible while allowing the explanation to change your interpretation.

Frequently asked questions

What is the difference between an earnings call summary and guidance tracking?

A summary organises the current document. Guidance tracking reconciles earlier targets with later updates and outcomes. A tool needs the relevant history and source checks to do the second job reliably.

Can a single-transcript summarizer identify omitted guidance?

Only if the supplied transcript itself identifies the older target or the tool also has the earlier evidence. Absence from one document does not establish absence from all company disclosures.

Does ghosted mean the company failed to deliver?

No. It is an extraction classification about a missing follow-up in the parsed record. The Infosys FY26 cash-conversion example here shows that even the missing-follow-up classification can be wrong.

How often did guidance go silent in Inve's snapshot?

As of 4 September 2026, 2,271 of 16,842 eligible commitments were classified ghosted (13.5% of those 16,842). Against all 24,328 tracked commitments, the same numerator is 9.3%. Neither is a failure rate for every Indian company's guidance.

Should I still read raw transcripts?

Yes, especially when a disputed target matters to the investment case. Summaries and ledgers help locate the relevant passage; the source and its qualifications determine what you can conclude.

The useful question is not whether today's summary sounds positive. It is whether the evidence now answers the specific question last year's target created.