Recently, I was speaking to James Palczynski – the CEO of DeCue Technologies – and he drummed up this summary of what you need to know: The new exposure in the earnings call analysis arms race is not the words management says, but entirely how executives say them. Machine learning now scores the delivery of each word, phrase and message for stress, confidence, hesitation, and possible deception, turning a routine call into a high-resolution read your team never authorized — and one investors are already trading on. The data these systems use to reach these insights is generally readable only by machine, a super-human capability.
Here are 10 things that James said:
- The exposure has moved from your script to the delivery. The words you carefully draft are now somewhat less interesting. What’s mined for advantage is understanding how they’re said. Machines are reading the acoustics under the language, particularly at a level that can’t be controlled or coached, content which speakers are unaware they have even made available.
- Every word is scored, roughly 3,500 different ways. These systems tag stress, confidence, hesitation, “sense of power,” emotional state, and supposed deception with mathematical markers of meaning, word by word. One service vendor sells exactly “what cannot be scripted… the unspoken meaning beneath the script.” It’s essentially a polygraph run on the whole call that yields more than a polygraph would, and that no one consented to.
- It’s used to second-guess your message and to trade against it even if the conclusions reached are inaccurate or misattributed. Those readings become positions against your guidance and expectations. Operators adjust and act on the acoustic score whether or not it’s accurate, so a misread of your CFO becomes someone’s trade regardless of the truth with no visibility or recourse for you.
- M&A and deal calls are considered the softest target. One provider, Speech Craft Analytics, markets a particular edge on M&A calls, reasoning that hastily assembled, less-rehearsed remarks leak the most signal. Most deal calls announce intent, not a completed transaction – so deal risk is still live and management confidence could be signal or even self-fulfilling prophecy.
- The scale is the tell. A standard system takes ~88 acoustic measurements every 10 milliseconds — roughly half a million data points across a 20-minute prepared remarks section… a nearly ideal format of lengthy, uninterrupted, measured commentary in a clean recording environment.
- It’s already commercialized — and quietly lucrative. Institutional investors are reportedly paying $10,000–$100,000 a month for these feeds, chasing something approaching a point of alpha. Some vendors market openly; others have scrubbed their sites as scrutiny grew.
- It reaches past emotion into health and cognition. The same data and methods are reliably used to flag clinical depression from call audio, with diagnosis of PTSD, early Parkinson’s, and anxiety also well documented in the research literature. The problems of an ability to gain this knowledge from involuntary, deeply personal signal leaking from a routine corporate event are clear.
- This is a disclosure-integrity problem first. The message your team builds is being overwritten by inferences from involuntary delivery cues — some of which are impossible to attribute and may have nothing to do with the content — but are traded on nonetheless. Your messages and your executives are already exposed.
- The options are few, and the tech industry is chasing capability, not defense. Most companies are choosing to shoulder risk… but by default because they are not aware of the capability at all. Only one tech company, DeCue Technologies, llc (www.decuetech.com) has emerged with a defense that is not synthetic voice. The other choices are not great: accept the risk; switch to synthetic voice (which kills the signal, but also the value of a live call); stop holding calls (hugely disruptive); or, as DeCue does, find a way to neutralize the machine signal while preserving the human voice.
- Bottom line for boards and GCs: make this a deliberate decision by a small group — the executives, in-house and outside securities counsel, IR — not a default reached because no one knew it was happening.
