Latest / The CMO Podcast with Fexingo: Marketing Leadership, Budgets, and Executive Strategy / How CMOs Use Sentiment Analysis on Earnings Calls
Transcript
- Lucas: So there's this fascinating shift happening in how CMOs are getting their consumer sentiment data. They're no longer just relying on surveys or social listening tools. Some of them are now turning to earnings call transcripts. Luna: Earnings calls? The ones where CEOs talk about revenue and margin? How does that help a CMO understand what customers are feeling? Lucas: Right, but think about it. An earnings call is one of the most carefully scripted hours in corporate America. Every word is weighed. And when analysts ask pointed questions about consumer demand or brand perception, the answers are loaded with real insight. CMOs are using natural language processing to scrape those transcripts and extract sentiment signals. Luna: So instead of asking customers how they feel, you're asking the executives who are paid to know how customers feel. That's actually pretty clever. Lucas: Exactly. And one of the best examples I've seen is from PepsiCo. Their CMO, Todd Kaplan, mentioned in an interview last quarter that they've been running sentiment analysis on their own earnings calls and also on competitors' calls. They're looking for shifts in language — words like 'cautious,' 'optimistic,' 'premium,' 'value' — and tracking how those correlate with actual consumer survey data. Luna: And does it correlate well? I imagine there's a gap between what execs say and what consumers actually do. Lucas: Turns out, the correlation is surprisingly strong. A working paper from researchers at Columbia and NYU looked at tens of thousands of earnings call transcripts from consumer goods companies and compared the sentiment scores from those calls with same-period consumer confidence indexes. The R-squared was above 0.7 in many categories — that's a very tight relationship. The language executives use when discussing consumer trends is essentially a leading indicator. Luna: So if a CEO starts using words like 'trading down' or 'promotional' more frequently, that's a signal that a brand shift is coming. Lucas: Exactly right. And the CMO can act on that before the quarterly survey data even lands. For PepsiCo specifically, Kaplan said they used sentiment signals from Q2 earnings calls — both theirs and Coke's — to adjust their messaging strategy for the 2025 Super Bowl. They noticed a dip in the word 'indulgence' and a rise in 'value' across competitor calls, so they leaned into a 'more for your money' angle for Doritos rather than a premium positioning. Luna: That is a concrete example of marketing meeting finance. The earnings call is traditionally a finance event, but the transcript becomes a marketing asset. Lucas: Right. And it's not just PepsiCo. At a recent CMO roundtable I attended, the head of marketing at a major auto brand said they use sentiment analysis on Tesla's earnings calls to gauge consumer excitement about EV features. They look at how often Elon Musk says 'autonomy' versus 'range' and correlate that with their own sales data. Luna: That feels like a smarter way to benchmark than just watching Twitter. But what tools are actually doing this? Is it custom software or off-the-shelf? Lucas: There's a range. Some companies use platforms like AlphaSense or Sentieo, which are built for financial document analysis. But more marketing-specific tools are emerging — one called SignalFire, originally for talent intelligence, now has an earnings call sentiment module. And of course there's the DIY route: pulling transcripts from free sources like Seeking Alpha, running them through a sentiment API like Google Cloud Natural Language or AWS Comprehend, and building a dashboard. Luna: How accurate are these APIs with earnings call language? I imagine there's a lot of jargon and equivocation — 'we remain cautiously optimistic about headwinds' — that might confuse a generic sentiment model. Lucas: Great question. The generic models are actually surprisingly good now. But many CMOs fine-tune them on a corpus of past calls from their industry. So you train the model to recognise that when a beverage company says 'the category is flat,' that's negative, but when a luxury brand says 'the category is flat,' that might be neutral because they're gaining share. The domain tuning matters. Luna: So we're moving from 'what do our surveys say' to 'what does the market say about what consumers want.' It feels like a higher-resolution picture. Lucas: And it's faster. Surveys take weeks to field and analyse. Earnings call transcripts are published within hours, often same-day. So you get a weekly or even daily sentiment signal for publicly traded competitors. For a CMO trying to justify a budget shift to the CFO, having a data point from the CFO's own domain — earnings language — is a powerful persuasion tool. Luna: That makes me think: if these marketing conversations have sparked something you've actually used — like a new way to pitch your budget or a tool you're evaluating — that kind of insight has real value for you. It definitely does for us. And honestly, if today was worth a coffee to you, that's the link — buy me a coffee dot com slash fexingo. Lucas: Yeah, listener support is what keeps this show ad-free and lets us keep digging into these concrete examples. No pressure, but if you got something actionable from this episode, that's the best way to say thanks. And we appreciate it more than we can say. Luna: Exactly. So back to the data — you mentioned that this sentiment analysis can also help with crisis management. How does that work? Lucas: Great segue. If a competitor has a product recall or a PR disaster, the sentiment shift on their earnings call is immediate. The CEO's language becomes defensive. You can detect that within hours and decide whether to adjust your own messaging — for example, running ads that emphasise your safety record or quality control. One CMO I know at a consumer electronics company set up an alert: if negative sentiment on a rival's call spikes more than 20 percent in a single transcript, their team gets a notification. Luna: That's competitive intelligence at machine speed. It used to be that you'd read about a competitor's trouble in the Wall Street Journal the next day and then plan a response. Now you have a real-time signal. Lucas: And it scales. You can monitor hundreds of companies simultaneously. That's impossible with human reading. The CMO of a large retailer told me they track sentiment across their entire supplier base — if a key vendor's earnings call shows signs of financial stress, they flag it for the procurement team to consider alternative sourcing. It's become a cross-functional tool. Luna: So the CMO is now also a risk manager and a supply chain watcher. That's a big expansion of the role. Lucas: It is. And it reinforces the argument that marketing belongs at the strategy table. If you can bring data from earnings calls — the language of the CFO — and show how consumer sentiment is shifting before quarterly results, you're speaking the same language as the rest of the C-suite. Luna: Are there any pitfalls? I imagine the sentiment model might misinterpret sarcasm or hedging. CEOs sometimes say 'we're pleased with our progress' when things are actually dire. Lucas: That's the biggest limitation. Sarcasm and irony are still hard for NLP. And there's the 'spin' problem — executives are trained to put a positive gloss on bad news. But researchers have found that even the spin itself is a signal. If a CEO uses more hedging words like 'might,' 'could,' 'possibly,' than in previous quarters, that's a negative signal even if the surface sentiment is positive. Luna: So you're measuring not just the words, but the confidence behind them. That's a level of nuance that most traditional sentiment tools miss. Lucas: Exactly. And some advanced platforms now incorporate tone analysis from the audio itself — pitch, pace, hesitation. But most CMOs are starting with the text and getting plenty of value. The threshold to start is low: you can pull a transcript today, run it through a free sentiment API, and get a score in five minutes. Luna: Let's talk about the cost. Is this something only big brands with dedicated data science teams can do, or is it accessible to smaller marketing teams? Lucas: It's actually becoming quite accessible. There are SaaS tools that start at a few hundred dollars a month for a limited number of transcripts. And the DIY route with open-source models like those from Hugging Face costs essentially nothing for compute if you use a small model. A small marketing team could set up a pipeline in a weekend with Python and a free API tier. Luna: What about privacy or legal concerns? Are earnings call transcripts fair game for competitive analysis? Lucas: Completely. Earnings calls are public. The transcripts are widely distributed. There's no non-public information issue. However, if you're using sentiment analysis on your own calls to gauge internal sentiment, that's fine too. But the real power is in benchmarking against competitors. Luna: So what's the one thing a CMO should do tomorrow to start exploring this? Lucas: Pick three competitors. Go to a site like Seeking Alpha or the SEC's EDGAR system and download their last two earnings call transcripts. Paste them into a sentiment analysis tool — there's a free one called MonkeyLearn that has a pre-built sentiment model. Run the same analysis on your own last call. Compare the emotional arc of the conversation. If your competitors are using more optimistic language about consumer demand than you are, that might signal you're being too cautious. Luna: That's a concrete action. And it costs nothing but a couple of hours. Lucas: Exactly. And once you see the pattern, you'll never look at earnings calls the same way again. They're not just financial reports — they're rich marketing intelligence documents.