There is a specific genre of ad tech founder who shows up on a podcast, answers every question politely, calls himself "an engineer who loves solving problems," and leaves you with the distinct sensation that you have just been handed a very clean napkin and no explanation for the mess.

Hitesh Chawla is that founder. He is also, and I want to be fair here because fairness is the whole job, one of the more disarmingly honest people I have interviewed in this industry. He did not dodge. He did not lawyer up mid-sentence.

When I asked him for the worst engineering decision in Silverpush's history, he did not do the thing where a CEO says we've made mistakes, and we've learned from all of them, which is corporate for no comment with a smile. He said: "the worst would still be the audio beacon concept which got us infamous for privacy."

Then, three questions earlier, I had asked him the best engineering decision. His answer: "probably the decision we took when we received the FTC notice and then completely pivoted to like completely 180 degrees of it."

The best decision and the worst decision in this company's history are the same event, viewed from opposite ends. That is not a corporate narrative. That is a Greek play with a Series C.

In roughly 2014, Silverpush had the problem all of ad tech had. Advertisers wanted to know whether the person who saw the TV spot was the person who later poked at a banner on a phone. Cookies could not cross that gap. Device IDs could not cross that gap. Nothing could, because televisions and telephones lived in separate universes and the only thing connecting them was a human being on a couch.

Chawla describes the origin without varnish: "this is like I think 2015, probably 2014. We were doing something, building something in India which was around how we could match what are people's viewing habits on TV. So we were using ultrasonic audio for that."

Elegant in the way a heist is elegant. Embed an inaudible ultrasonic tone in the television commercial. Ship an SDK to mobile app developers. Have that SDK sit in the background, microphone open, listening. When the phone hears the ad, the phone tells Silverpush. Congratulations, you have linked a living room to a pocket without a single cookie.

They called it Unique Audio Beacons. The industry called it clever. Privacy advocates used words like surveillance software and creepy, and both of those were descriptive rather than rhetorical.

The user-facing disclosure was a microphone permission. The same permission you grant a flashlight app that has no business asking. Nobody told you the app wanted your microphone so it could take attendance on your television habits.

In 2015 the Center for Democracy and Technology took it to the FTC. In March 2016 the FTC sent warning letters to twelve Android developers suspected of shipping Silverpush code, noting the SDK could generate detailed logs of television content playing near a phone. EPIC piled on. The W3C ended up citing Silverpush in Web Audio privacy discussions, which is the standards-body equivalent of being used as a cautionary tale in a driver's ed video.

Here is where Chawla says the thing I have not stopped chewing on.

"You're right that FTC notice came which somehow assumed that we were doing something in US which we were not and we did not take it seriously, we did not respond so that's why they put it on the website and that became a bigger deal than it was."

Read it again. Slowly.

The Federal Trade Commission of the United States of America sent a notice. The internal response was, and I am paraphrasing only slightly, eh. They did not respond. So the FTC published it. And publication is what turned a regulatory paper cut into the defining fact of the company's first decade.

I want to be precise about what this is and is not. It is not an admission of wrongdoing. His position, consistently held, is that Silverpush built for India, that the FTC's assumption of US deployment was wrong, and that the company had no capacity to argue: "we were too small to fight it out and rather we thought what's something we take out as a learning from the whole incident."

But strip the specifics and what remains is a founder telling you that the most consequential event in his company's history happened because nobody in the building thought a letter from a US regulator was worth answering. That is not a privacy scandal. That is a governance scandal wearing a privacy scandal's coat.

And then the sentence that is the entire pivot, compressed: "what we realized that what we were building is always going to be questioned so why not build something which will not fall into such trap."

Not we were wrong. Not we violated our users' trust. This will always be questioned, so let us build something unquestionable. That is a compliance insight. It is not an ethical one. And I say that as someone who thinks the destination he reached is defensible and maybe even good.

Midway through I raised something sitting in my notes. Nielsen. Wearables. Devices that listen for audio watermarks in broadcast. I had spoken to people over there recently. Functionally the same category of technology.

He did not blink: "It's already happening with not just Nielsen but there are multiple companies which do something similar."

He is right, and that is the deeply uncomfortable part. The ultrasonic-beacon-shaped hole in the ad tech universe never closed. It got a compliance department, an opt-in panel, a consent flow, and a much better lawyer. The technology was never the sin. The absence of disclosure was the sin. Silverpush got caught in the window where the capability existed and the norms did not, which raises the question I will spend Parts Two and Three dismantling: did Silverpush change, or did the industry simply build a legal wrapper around the thing Silverpush got punished for doing early?

Post-FTC, Silverpush killed Unique Audio Beacons. With a lag, it should be said; critics noted the product was still being advertised on the company's own website weeks after the public pledge, which is either an ops failure or a tell, and I have views.

What replaced it was a full inversion. In his words: "the learning from the same thing was that let's not do something where there's any consumer data involved, no questions involved. And then, how do you target consumers if you can't use that personal data? And contextual was the only answer at that time."

Then he adds the detail that dates the whole thing: "in fact, if you think of 2015, no one was talking about or even no one was using the word contextual at that time." He is essentially correct. The category did not have a name yet, and neither did his product: "even we didn't call it contextual initially. Our idea was that we'll understand the content, understand the content and based on that we'll show ads. And then once we launched, then we realized it's called contextual. So that automatically became our story."

"So that automatically became our story." I would like that framed. An entire company narrative, arrived at retroactively, after the fact, because the market supplied a word for what they had already built.

He chose video deliberately, because video was the hard version: "why we started with video was because that was a hard challenge to solve at that time. Now it's easy to understand the content but at that time there was no such AI model so it was a hard problem to solve." And the timing worked, because "video was anyways as a content platform exploding everywhere whether it was YouTube or other social platforms. That content explosion and our story around contextual helped build it out."

He counts the clock from the disaster, not the incorporation: "I would count 2015 as kind of a restart of the company." Ten years, roughly twenty to twenty-two million dollars across seven rounds, strongest in India, Indonesia, the Philippines, Thailand, Malaysia, Vietnam, the Middle East, Africa. Success in the developed markets, which surprised him: "just strangely, developed markets worked." And one persistent failure he cannot explain, which I find weirdly reassuring, because a founder who can explain every market outcome is a founder making it up: "we've struggled so far in South America. We've not found success in South America."

Today the company scans something on the order of half a billion seconds of video a day.

Part One is the origin myth, and origin myths are the easy part. Everyone loves a redemption arc. Redemption arcs sell newsletters.

Parts Two and Three are where I stop being charmed and start being a reporter.

PART TWO: "We Have Never Had A Conversation With YouTube," And Eleven Other Things He Said That Should Terrify His Competitors And His Customers In Roughly Equal Measure. The product teardown. What MIRRORS actually does versus what the marketing says it does. Why the Trend Intelligence Platform's much-advertised "real-time" turns out, when I pushed him on hours versus minutes, to be "we refresh it every 24 hours" and why that is either fine or fatal depending on what you were sold. The connected TV integration horror show he describes with startling candor, including the part where they read subtitles to figure out what show you are watching. And Dennis Pickle's on-air question about whether this entire business is just a gap left by platforms that refuse to grade their own homework, which Chawla answered with "it's built on someone else's gap for sure" and zero hesitation.

PART THREE: The Alienation Tax, The Vibe-Coding Sales Team, And What It Means That The Most Honest Man In Ad Tech Is The One Who Got The FTC Letter. The uncomfortable synthesis. Whether "privacy-safe contextual" is a real category or a laundering mechanism. Why the FTC stopping at warning letters created the template every sensor-based tracking company has run since. The continuity question: same founder, same engineering culture, same market pressure, new vocabulary. Why his entire sales team is learning to code and what that means for your headcount. And the rapid-fire answer nobody in this industry wants printed, which is that the most overrated thing in ad tech is "that everyone is AI native which is not true."

Subscribe. It is cheaper than the DSP fees you are not auditing.