Latest / Tech IPO Conversations with Fexingo: Public Software Companies and Capital Markets / How Post-IPO Companies Use AI to Write Their Earnings Scripts
Transcript
- Lucas: If you've listened to enough earnings calls, you start to notice a certain rhythm. The same phrases, the same cadence, the same careful avoidance of anything that might be interpreted as a forward-looking statement. Luna: Right, and lately it feels like that rhythm is getting even more uniform. Like someone's using a template. Lucas: Exactly. And increasingly, that template is written by an AI. We're seeing a growing number of post ipo companies using large language models to draft their earnings call scripts. Not just bullet points, but full paragraphs, Q&A prep, even the tone markers. Luna: So instead of the CFO or the head of IR spending two weeks crafting the message, they feed last quarter's transcript and the new numbers into an AI and get a draft in minutes. Lucas: That's the pitch. And some companies are pretty open about it. I was looking at a recent filing from a mid-cap SaaS company — I won't name them because the practice is still spreading — where they disclosed that the CEO's prepared remarks were 'substantially generated by an AI language model, reviewed and edited by management.' Luna: Substantially generated. That's a new one. I wonder how the SEC feels about that. Lucas: Well, the SEC's rules on fair disclosure haven't changed. You're still liable for what you say. But the AI is changing the economics of narrative control. A small IR team can now produce the same polished language as a Fortune 500 comms department. Luna: But does polished mean better? I've heard some calls lately where the language is so smooth it feels like it's gliding over the real issues. Lucas: That's the risk. An AI trained on past earnings calls learns to mimic the most cautious, least controversial phrasing. So you get a script that avoids any sharp edges — but also avoids genuine insight. There's a term for it: 'earnings call homogenization.' Luna: Homogenization. So every call starts to sound the same, and investors can't tell which companies are actually confident and which are just following the script. Lucas: Exactly. And there's a second-order effect: if everyone uses similar AI tools, the language converges. Suddenly, every CEO is 'pleased with our execution in a dynamic environment' and 'remain focused on driving long-term shareholder value.' It becomes noise. Luna: I remember a call last quarter where the CEO accidentally said 'as the AI suggested' during the Q&A. You could hear the IR person cringe. Lucas: That's the nightmare scenario — the AI's influence bleeding into live conversation. Most companies train their models on their own historical transcripts to keep the voice consistent, but the model doesn't have judgment. It doesn't know when to break the script. Luna: So are we moving toward a world where earnings calls are essentially performances? Scripted, rehearsed, with AI as the playwright? Lucas: To some extent, they've always been performances. But the AI amplifies it. The counterargument is that it also reduces errors. No more accidental forward guidance, no more off-script comments that move the stock. For a post ipo company under scrutiny, that consistency is valuable. Luna: But at what cost? If every call is a polished artifact, the signal gets lost. Investors start to rely more on quantitative beats and misses, and the qualitative context — the stuff that actually explains the numbers — becomes boilerplate. Lucas: That's the tension. And some companies are experimenting with a hybrid model: AI generates the first draft, but the executive team rewrites key sections by hand, especially the parts about strategy and outlook. They want the authenticity to come through. Luna: And what about the Q&A? That's where the real test happens. You can script the opening remarks, but you can't script the questions. Lucas: Actually, some companies are using AI to prep for Q&A too. They feed the draft script into a model that generates likely analyst questions, then they rehearse the answers. It's like a debate prep tool. Luna: So the whole call becomes a closed loop. The AI writes the script, predicts the questions, and the executives just read the approved responses. Lucas: It sounds dystopian, but it's not that different from what large companies have done for years with human communications teams. The difference is speed and scale. A human team might take a week to produce that level of polish. An AI does it in an afternoon. Luna: I wonder if the market will start discounting calls that sound too smooth. Like, if an earnings call sounds like it was written by a machine, maybe investors penalize it. Lucas: There's actually some early research on that. A working paper from a few months ago analyzed the linguistic patterns of earnings calls and found that calls with lower 'perplexity' — basically, more predictable language — were associated with weaker stock performance in the following quarter. The market seems to read between the lines. Luna: Interesting. So even if the AI is helping companies control the narrative, the market might be learning to see through it. Lucas: Exactly. And that creates an incentive for companies to use AI more strategically — not just to sanitize, but to actually clarify. Some forward-thinking IR teams are using AI to ensure consistency while deliberately leaving in a few 'human' moments. A candid aside, an unscripted reaction. Luna: So the art becomes knowing when to break the script. That's still a human skill. Lucas: And that's not something you can outsource to a model. At least not yet. Luna: Speaking of not outsourcing — we should mention that this show itself is a human production. We don't use AI to write our scripts. Lucas: Right. And that's a choice we make because we think the conversation is better when it's two people actually thinking through things in real time. But it does take time and resources. Luna: And a handful of listeners help keep this going. If you find these deep dives useful, you can support the show at buy me a coffee dot com slash fexingo. It's what keeps us ad-free and independent. Lucas: Yeah, no pressure, but if you're in a position to chip in, it genuinely helps us keep doing this kind of work. And we really appreciate the folks who already do. Luna: Alright, back to AI and earnings. So where do you see this going over the next year? Lucas: I think we'll see the SEC eventually issue some guidance on disclosure when AI is used materially in investor communications. Not a ban, but something like 'if a significant portion of your script is ai generated, you should note it.' Luna: That would be a big shift. Right now, most companies don't mention it at all. Lucas: Right. And the ones that do are usually the ones proud of their tech adoption. But as the practice becomes standard, the absence of disclosure might itself become a signal. Luna: So the market adapts. As always. Lucas: As always. And for investors, the takeaway is: listen for the language. If the call sounds too perfect, it might be worth digging deeper. The real story is often in the imperfections.