Researching earnings calls out loud: what changes when you can just ask

The follow-up is where the value is

A first question rarely produces the answer. It produces a direction: a company you had not considered, a number that does not match your assumption, a phrase worth chasing. The work is in what you ask next, and typing imposes a small tax on each one that adds up until you stop asking.

Spoken follow-ups cost almost nothing, which changes how many you ask. "And the quarter before?" "Who else said that?" "Was that the CFO or the CEO?" Each is two seconds of speech and none of them would have been worth the typing. The research goes deeper because the friction that used to end it is gone.

What voice is worse at, honestly

• Numbers in bulk. A table of eight companies and four metrics is unreadable aloud and obvious on screen.
• Precise recall. A spoken figure passes and is gone; a written one can be reread without asking again.
• Anything you need to copy. A quote you intend to paste into a note wants to be text from the start.
• Noisy places and open offices, for the obvious reason and the less obvious one that the transcription degrades exactly when you most want it accurate.

The pattern that works: talk wide, read narrow

Use speech for the exploring, where the questions are short and each answer only needs to be good enough to decide the next one. Which sector moved, who raised, does anyone contradict that. This is the phase where volume of questions beats precision of any single one.

Then switch to reading for the part you will act on. The transcript of the conversation is right there, the quotes are attached to the turns that produced them, and the figures can be checked at your own pace rather than at the speed of speech. The same conversation carries both, so the switch costs nothing.

Why grounding matters more, not less, in speech

A conversation pressures a model to keep talking, and the cheapest way to fill a pause is with something plausible. That is exactly the failure an earnings tool cannot afford, because a fabricated figure said confidently out loud is far more convincing than the same figure on a page.

The defence is that spoken answers come from the same retrieval as written ones: the archive is searched, the passages are found, and the claim is built from them. When nothing is found, the honest answer is that nothing was found. An assistant that says "the calls do not cover that" is worth more than one that never disappoints you.

How earnings.chat helps

Voice runs over the same archive and the same tools as the written chat: more than 253,000 earnings calls from 12,853 companies back to 2020, searched by sector, country or period. The spoken exchange lands in the same conversation as anything typed, with the passages behind each answer attached to it, so stopping the call and carrying on in writing continues one thread rather than starting a second.

Every lookup is shown while it happens, which matters more in speech than in text: a few seconds of silence with nothing explaining it is the moment a user assumes something broke.