👋 Hi, welcome to Signal Lab where the team at Innate Language Processing digs deep into language & behavior, the limits of technologies, and the connection between what people say and do. For venture investors and startups.
You invested in a scammy startup. That’s an unavoidable risk of venture investing, right? “Comes with the territory”? Not as much as you’d think.
It seems Delve is the most recent example of a company whose claims exceeded reality. (If you know the story, skip down to the next section.) Founded in 2023 by Karun Kaushik and Selin Kocalar, both Forbes 30 Under 30 members and MIT dropouts, Delve is an hyperscaled AI-native platform for SOC2 and HIPAA compliance that offers certification in weeks not months. They joined Y Combinator in 2024 with a 3.3 million seed round, then raised a 32 million Series A led by Insight Partners in July 2025.
Early this year, a whistleblower alleged that Delve faked this compliance offering, with massive implications for hundreds of customers who may face HIPAA and GDPR fines and liability, if true. In March, an anonymous former client wrote a full exposé. In response, YC dropped Delve publicly, and lead investor Insight Partners tried, insofar as they were able, to wash their hands of them as well.
This is not a post-mortem of the company (which is not dead yet, and the founders are only partially owning up to allegations). Instead, this is an investigation of just how this company got funded. The Delve case reveals a huge source of bias that we have never heard anyone talk about in VC….
Do you hear what I hear?
Acoustic analysis can tell a lot about a person. More than you may think. When we first heard Delve CEO Karun Kaushik’s voice, it struck us how similar his vocal patterns were to Praveen Akkiraju, Managing Director and the Delve deal lead at Insight Partners (lead investor of Delve’s Series A). (Above are separate speech samples, not from the same conversation.) You will probably be able to pick up on some of the similarity as well.
When we ran acoustic parameter tests on Kaushik and Akkiraju, the evidence of their similar vocal patterns was more than anecdotal. Both speakers show significant deviations from norms in several acoustic parameters, i.e., biomarkers in language, indicating pronounced dysphonia, rough-breathy characteristics well outside the average. For example, jitter is a linguistic measure of cycle-to-cycle vocal pitch variation (not “pitch” like decks). Kaushik and Akkiraju are both over 3x a typical speaker. For shimmer, which measures variation in speech amplitude, both Kaushik and Akkiraju are over double the norm. And their HNR (harmonic to noise ratio) which measures spectral quality (i.e., breathiness, turbulence) is not only very low, it is virtually identical. 5.4 dB for Akkiraju, 5.3 dB for Kaushik. Typical is 15-20 dB. But wait, there’s even more dysphonia! Elizabeth Yin, who led the Delve investment at Hustle Fund, shares a distinctive speech sound disorder with the Delve CEO: a lateral lisp. This is when air escapes to the sides of the mouth rather than flowing down the middle when the /s/ or /z/ sound is made.
The question is, why does it matter if these speakers have similar acoustic profiles?
The Bias Investors Aren’t Registering
Bias can be much more subtle than the forms we know by name, like Confirmation Bias, Recency Bias or discrimination on the basis of race, gender, etc.
Decision-making, like it or not, is influenced by personal affinity (i.e., a spontaneous or natural liking for someone or something). What shapes that affinity is often superficial (e.g., perhaps an alma mater, mutual friend, or even acoustic profile), but in decision-making, this is bias. It even has a name: Affinity Bias.
Affinity feels like connection, not decision error. It populates first impressions and gives a pass to plenty of people who might not stand on their own merits. Studies have demonstrated investors’ preference for (and up to 35% greater likelihood of investing in) founders whose political stance aligns with their own, whose facial features are similar to theirs, or simply attractive men (“attractive” females did not do any better than “unattractive” females in the study…). This is how the world works, but it doesn’t make great investments.
Affinity is more risky than you might realize. Peer-reviewed research has demonstrated that Affinity Bias (even in non-linguistic forms) reduces the probability of investment success by nearly 20%, with the worst impact falling on early stage deals. However, the risk can be inverted. If the presence and effects of Affinity Bias are parsed and understood by an investor, investment success can improve by +8% per year. That’s almost a 30% difference in investment success, dependent on whether or not an investor is accounting for their Affinity Bias.
When Affinity Bias breeds faulty intuition, it is a saboteur. But when seen and understood, it becomes an edge.
Affinity Bias in language
Language is one of the most powerful vectors of Affinity Bias. Where language mirrors, perception of credibility is more likely to follow. In one study, funding increased by over 7% with one linguistic tweak in the founder’s pitch. Examples abound outside investing. For instance, criminal psychopaths are over twice as likely to earn release on parole as their non-psychopathic counterparts, manipulating impression by Linguistic Style Matching, an easily accessible form of linguistic Affinity Bias.
Just as we are attentive to and resonate with people who are familiar to or look like us, so we are with people who sound or talk like us – whether prosodically (rhythms, intonations), verbally (idioms, vocabulary), or dynamically (pacing, energy).
The Analysis
When we ran a behavioral linguistic analysis of these two speakers — drawing insight from a level below lexical choice or ethnolect — we identified pronounced similarities in their idiolects, which are structural linguistic patterns that subconsciously build affinity during the evaluation process.
Idiolect is a linguistic fingerprint, the unique habits and patterns that distinguish how someone communicates from everyone else. This radar chart maps five key dimensions of idiolect, where Kaushik and Akkiraju share significant similarities.
Convergence is not a measure of investment quality; it is a measure of aligned idiolect and investor bias potential.
Some excerpts from Innate’s Affinity Bias Assessment:
“The founder's manner of qualifying claims, hedging assertions, and stating convictions mirrors the cadence the investor himself uses, creating an unconscious sense of alignment that can be easily mistaken for evidence of the founder's competence or strategic thinking.”
“The mutual comfort with high uncertainty language is particularly significant. Founders who speak candidly about unknowns can be perceived as either lacking conviction or as intellectually honest. The investor's own elevated use of uncertainty markers means he may be predisposed to interpret that same pattern in a founder as a sign of rigor rather than weakness, creating an affinity-driven interpretive advantage for this founder.”
“Both speakers avoid moral framing, self-deprecation, risk-heavy rhetoric, or urgency-driven appeals; this shared temperance would have made the founder's communication register as measured and pragmatic to the investor, mirroring his own baseline for how ambition ‘should’ sound.”
The figure above (one of ten in the full assessment) demonstrates that both founder and investor score notably high (~2+ standard deviations above the norm) on four measures of cognitive processing, indicating that both speakers openly wrestle with complexity, acknowledge what is not yet known, and revisit assumptions. Their broadly similar baseline cognitive vocabulary evinces a shared cognitive rhythm that renders the founder’s reasoning style as immediately intellectually credible to the investor.
Your Affinity Bias
To the trained ear, linguistic affinity is telling. At Innate, we can often determine within a minute which partner at a fund manages a particular investment. It’s easy because it’s usually the partner who speaks the most like the founder. And yet Affinity Bias is a deeply underreported source of decision error on the part of investors.
If you invest, interrogate your own investments. How often do the founders (or GPs) in your portfolio share a way of speaking similar to yours?
Visualizing where your idiolect converges with another’s is the first step.
Our Affinity Bias assessment measures convergence across eight domains including motivation, cognition, ambition, tense, tone, social style, discourse elements, grammar. It reveals where bias is feeding your intuition.1
See you next month.
1 Our model accounts for gender, ethnicity, age, English as a second language in the metadata, to prevent these variables interfering and confounding results.