Goldfib Key takeaways
- The most valuable output of a concall is not this quarter's commentary. It is how today's guidance compares with what management said four and eight quarters ago.
- A guidance-vs-delivery tracker turns qualitative calls into a credibility score per management team.
- Evasive answers, vanishing topics and analyst questions that go unanswered are signals in their own right.
- Use AI for extraction and comparison, and humans for verification and judgement. Never paste AI-extracted numbers into a model unchecked.
Most Indian listed companies of any size now hold a call after each quarterly result and file the transcript or recording with the exchanges. For an investor that is a remarkable resource: years of management statements, on the record, in a consistent format.
Most people read a concall the day it happens and never look at it again. That wastes most of its value.
Treat each call as one point in a series
A single call tells you what management wants you to think this quarter. A series of calls tells you whether management can be believed. Analysts get the most out of transcripts when they compare what was promised with what was delivered.
The guidance-vs-delivery tracker
Here is what a tracker reveals for two hypothetical management teams over eight quarters of guidance:
Quietly dropped guidance is the most important category and the hardest to catch by reading calls one at a time. A capex target, a new product, or a margin goal is mentioned for three quarters, then never again. Nobody asks, because nobody is tracking.
Language signals worth measuring
Beyond hard numbers, transcripts carry softer signals. These are useful when measured consistently over time, not when one sentence is over-interpreted.
| Signal | How to measure | Why it matters |
|---|---|---|
| Hedging language | Frequency of “challenging”, “headwinds”, “we remain cautiously…” per 1,000 words | Rising hedges often come before guidance cuts |
| Topic share | % of management remarks spent on each segment or theme | Management talks most about what’s working and avoids what isn’t |
| Answer directness | Share of analyst questions answered with a number or a direct yes/no | Evasion on specific questions is informative |
| Repeat questions | Same question asked by several analysts across quarters | The market’s unresolved worry |
| KPI substitution | Metrics added or removed from the opening remarks | A metric that disappears is often a metric that turned bad |
Where AI helps, and where it fails
AI language models are well suited to transcript work. They read fast, they are consistent, and they can compare the same language across dozens of documents. Used carelessly, they are also dangerous.
- Target read as actual. “We aim to reach 20% margins” extracted as “margins of 20%”.
- Segment confusion. Numbers for one segment attributed to another, or consolidated figures mixed with standalone.
- Period confusion. FY versus calendar year, or H1 guidance read as full year.
- Lost hedges. “If demand holds, we could see…” summarised as a firm commitment.
- Speaker confusion. An analyst’s suggestion recorded as management guidance.
The fix is a workflow rule rather than a better prompt: every AI-extracted number carries the exact quote and location, and a human signs it off before it enters a model. We cover this division of labour in AI for investment research: what it can and can’t do.
Putting it into practice
- Start with your five largest holdings and build eight quarters of trackers.
- Score each management team’s hit rate and note any quietly dropped guidance.
- Add one line to each investment memo: Management credibility: hit rate X%, key slippages Y.
- Refresh within a week of each results season.
Done this way, transcripts stop being a quarterly reading chore and become one of the most useful inputs in your research process.
Frequently asked questions
What is a concall in Indian stock markets?
A concall is a conference call, usually held after quarterly results, where company management discusses performance and answers questions from analysts and investors. Indian listed companies upload recordings or transcripts to the stock exchanges, which makes them a key primary source for research.
How do you analyse an earnings call transcript?
Extract quantitative guidance such as revenue growth, margins, capex and debt targets, then compare it with past guidance and actual results. Note topics that appear or disappear, track changes in tone, and flag questions management avoided. The comparison across quarters matters more than any single call.
Can AI analyse earnings call transcripts accurately?
AI models are good at extracting statements, summarising themes and comparing language across quarters. They can misread numbers, confuse segments, or present a target as an achievement. Treat AI output as a first draft and verify every number against the transcript and the filed results.
What are red flags on an earnings call?
Guidance that keeps being pushed out, metrics that management stops reporting, repeated non-answers to the same analyst question, a sudden change in which KPIs management emphasises, and unexplained changes in the CFO or auditor.