StarZero All shows Try the podcast clipper →

Jun 29, 2017 · 20m · top-founders

705: With $8.8M Raised, Is This The Ultimate Machine Learning Tool?

Scott Clark · 10m spoken Nathan Latka · 8m spoken
0:00 / 0:00

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

In this episode of The Top, host Nathan Latka interviews Scott Clark, co-founder and CEO of SigOpt, discussing how the Andreessen Horowitz-backed startup provides black-box Bayesian optimization software for enterprise AI models while scaling to $50,000 in monthly recurring revenue.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Nathan holds 43.6% of the talking time here. How this is scored →

Nathan as informed peer 5.0 Guest teaching 4.0 Guest disagreement 1.9 Nathan pushing back 4.3
05100:0010:0020:000:42–3:12 · Nathan as informed peer 4/10 Introducing Scott Clark and SigOpt Nathan introduces Scott and drills down on SigOpt's SaaS pricing tiers and top-down sales model. Scott clarifies how enterprise packages scale to ten thousand dollars a month and why executive buyers look outside for optimization expertise.3:13–5:51 · Nathan as informed peer 4/10 Enterprise Use Cases and Fraud Detection Nathan pushes Scott to move past vague generalities about Prudential to provide a concrete technical use case. Scott complies while protecting client confidentiality, explaining how SigOpt tunes credit card fraud detection algorithms without replacing existing models.5:51–9:48 · Nathan as informed peer 6/10 Black Box Bayesian Optimization Mechanics Nathan challenges machine learning buzzwords and questions how a generalized platform works without bespoke consulting. Scott educates him on black-box Bayesian optimization, prompting Nathan to synthesize the mechanics using a waterslide analogy.9:48–11:51 · Nathan as informed peer 5/10 Fundraising History and Proprietary IP Security Nathan bluntly challenges whether SigOpt risks cross-contaminating trade secrets between direct competitors like insurance or trading firms. Scott firmly reframes how the architecture abstracts models so proprietary IP never touches SigOpt's servers.11:51–14:41 · Nathan as informed peer 6/10 Customer Count, Churn, and Team Distribution Nathan calculates SigOpt's monthly recurring revenue at fifty thousand dollars from twelve customers and probes on contract discounting and churn. Scott confirms the numbers and details their team allocation across engineering and sales.14:42–17:35 · Nathan as informed peer 7/10 Venture Capital Dynamics and Headcount Burn Nathan aggressively asks if Andreessen Horowitz ignored unit economics like CAC and LTV just to chase AI hype. Scott rejects the premise by defending their due diligence, while Nathan calculates a monthly payroll burn of around one hundred thirty thousand dollars.17:37–19:01 · Nathan as informed peer 3/10 The Famous Five Rapid-Fire Questions Nathan runs through the Famous Five rapid-fire format smoothly, with Scott sharing insights on building sustainable operational habits and leadership readings.0:42–3:12 · Guest teaching 3/10 Introducing Scott Clark and SigOpt Nathan introduces Scott and drills down on SigOpt's SaaS pricing tiers and top-down sales model. Scott clarifies how enterprise packages scale to ten thousand dollars a month and why executive buyers look outside for optimization expertise.3:13–5:51 · Guest teaching 6/10 Enterprise Use Cases and Fraud Detection Nathan pushes Scott to move past vague generalities about Prudential to provide a concrete technical use case. Scott complies while protecting client confidentiality, explaining how SigOpt tunes credit card fraud detection algorithms without replacing existing models.5:51–9:48 · Guest teaching 7/10 Black Box Bayesian Optimization Mechanics Nathan challenges machine learning buzzwords and questions how a generalized platform works without bespoke consulting. Scott educates him on black-box Bayesian optimization, prompting Nathan to synthesize the mechanics using a waterslide analogy.9:48–11:51 · Guest teaching 6/10 Fundraising History and Proprietary IP Security Nathan bluntly challenges whether SigOpt risks cross-contaminating trade secrets between direct competitors like insurance or trading firms. Scott firmly reframes how the architecture abstracts models so proprietary IP never touches SigOpt's servers.11:51–14:41 · Guest teaching 2/10 Customer Count, Churn, and Team Distribution Nathan calculates SigOpt's monthly recurring revenue at fifty thousand dollars from twelve customers and probes on contract discounting and churn. Scott confirms the numbers and details their team allocation across engineering and sales.14:42–17:35 · Guest teaching 3/10 Venture Capital Dynamics and Headcount Burn Nathan aggressively asks if Andreessen Horowitz ignored unit economics like CAC and LTV just to chase AI hype. Scott rejects the premise by defending their due diligence, while Nathan calculates a monthly payroll burn of around one hundred thirty thousand dollars.17:37–19:01 · Guest teaching 1/10 The Famous Five Rapid-Fire Questions Nathan runs through the Famous Five rapid-fire format smoothly, with Scott sharing insights on building sustainable operational habits and leadership readings.0:42–3:12 · Guest disagreement 1/10 Introducing Scott Clark and SigOpt Nathan introduces Scott and drills down on SigOpt's SaaS pricing tiers and top-down sales model. Scott clarifies how enterprise packages scale to ten thousand dollars a month and why executive buyers look outside for optimization expertise.3:13–5:51 · Guest disagreement 2/10 Enterprise Use Cases and Fraud Detection Nathan pushes Scott to move past vague generalities about Prudential to provide a concrete technical use case. Scott complies while protecting client confidentiality, explaining how SigOpt tunes credit card fraud detection algorithms without replacing existing models.5:51–9:48 · Guest disagreement 2/10 Black Box Bayesian Optimization Mechanics Nathan challenges machine learning buzzwords and questions how a generalized platform works without bespoke consulting. Scott educates him on black-box Bayesian optimization, prompting Nathan to synthesize the mechanics using a waterslide analogy.9:48–11:51 · Guest disagreement 3/10 Fundraising History and Proprietary IP Security Nathan bluntly challenges whether SigOpt risks cross-contaminating trade secrets between direct competitors like insurance or trading firms. Scott firmly reframes how the architecture abstracts models so proprietary IP never touches SigOpt's servers.11:51–14:41 · Guest disagreement 1/10 Customer Count, Churn, and Team Distribution Nathan calculates SigOpt's monthly recurring revenue at fifty thousand dollars from twelve customers and probes on contract discounting and churn. Scott confirms the numbers and details their team allocation across engineering and sales.14:42–17:35 · Guest disagreement 3/10 Venture Capital Dynamics and Headcount Burn Nathan aggressively asks if Andreessen Horowitz ignored unit economics like CAC and LTV just to chase AI hype. Scott rejects the premise by defending their due diligence, while Nathan calculates a monthly payroll burn of around one hundred thirty thousand dollars.17:37–19:01 · Guest disagreement 1/10 The Famous Five Rapid-Fire Questions Nathan runs through the Famous Five rapid-fire format smoothly, with Scott sharing insights on building sustainable operational habits and leadership readings.0:42–3:12 · Nathan pushing back 2/10 Introducing Scott Clark and SigOpt Nathan introduces Scott and drills down on SigOpt's SaaS pricing tiers and top-down sales model. Scott clarifies how enterprise packages scale to ten thousand dollars a month and why executive buyers look outside for optimization expertise.3:13–5:51 · Nathan pushing back 5/10 Enterprise Use Cases and Fraud Detection Nathan pushes Scott to move past vague generalities about Prudential to provide a concrete technical use case. Scott complies while protecting client confidentiality, explaining how SigOpt tunes credit card fraud detection algorithms without replacing existing models.5:51–9:48 · Nathan pushing back 5/10 Black Box Bayesian Optimization Mechanics Nathan challenges machine learning buzzwords and questions how a generalized platform works without bespoke consulting. Scott educates him on black-box Bayesian optimization, prompting Nathan to synthesize the mechanics using a waterslide analogy.9:48–11:51 · Nathan pushing back 6/10 Fundraising History and Proprietary IP Security Nathan bluntly challenges whether SigOpt risks cross-contaminating trade secrets between direct competitors like insurance or trading firms. Scott firmly reframes how the architecture abstracts models so proprietary IP never touches SigOpt's servers.11:51–14:41 · Nathan pushing back 3/10 Customer Count, Churn, and Team Distribution Nathan calculates SigOpt's monthly recurring revenue at fifty thousand dollars from twelve customers and probes on contract discounting and churn. Scott confirms the numbers and details their team allocation across engineering and sales.14:42–17:35 · Nathan pushing back 7/10 Venture Capital Dynamics and Headcount Burn Nathan aggressively asks if Andreessen Horowitz ignored unit economics like CAC and LTV just to chase AI hype. Scott rejects the premise by defending their due diligence, while Nathan calculates a monthly payroll burn of around one hundred thirty thousand dollars.17:37–19:01 · Nathan pushing back 2/10 The Famous Five Rapid-Fire Questions Nathan runs through the Famous Five rapid-fire format smoothly, with Scott sharing insights on building sustainable operational habits and leadership readings.

speaking balance: gold is Nathan, purple is the guest (3 minute bins)

0:00 · Nathan 53.2% · guest 46.8%0:00 · Nathan 53.2% · guest 46.8%3:00 · Nathan 15.8% · guest 84.2%3:00 · Nathan 15.8% · guest 84.2%6:00 · Nathan 30.5% · guest 69.5%6:00 · Nathan 30.5% · guest 69.5%9:00 · Nathan 38.6% · guest 61.4%9:00 · Nathan 38.6% · guest 61.4%12:00 · Nathan 35.7% · guest 64.3%12:00 · Nathan 35.7% · guest 64.3%15:00 · Nathan 59.5% · guest 40.5%15:00 · Nathan 59.5% · guest 40.5%18:00 · Nathan 75.2% · guest 24.8%18:00 · Nathan 75.2% · guest 24.8%
Sharpest disagreement ▶ 10:46 Defending black box data isolation

Scott firmly rejects Nathan's skepticism that enterprise customer secrets or IP could accidentally leak to competing firms.

 Hardest push from Nathan ▶ 15:36 Challenging VC due diligence on unit economics

Nathan cuts through VC talking points to ask bluntly if Andreessen Horowitz ignored CAC/LTV ratios entirely to chase AI hype.

 Biggest teaching moment ▶ 6:12 Explaining hyperparameter optimization

Scott breaks down how black box Bayesian optimization replaces inefficient human trial-and-error in multi-dimensional parameter spaces.

 Nathan holds their own ▶ 8:52 Formulating the waterslide domain analogy

Nathan demonstrates quick comprehension of a complex algorithmic workflow by creating an accurate analogy of domain experts building a waterslide with provided component parameters.
the scores for every segment, with the reasoning behind each
ChapterTopicNathan as informed peerGuest teachingGuest disagreementNathan pushing backWhy
Introducing Scott Clark and SigOpt 4312 Nathan introduces Scott and drills down on SigOpt's SaaS pricing tiers and top-down sales model. Scott clarifies how enterprise packages scale to ten thousand dollars a month and why executive buyers look outside for optimization expertise.
Enterprise Use Cases and Fraud Detection 4625 Nathan pushes Scott to move past vague generalities about Prudential to provide a concrete technical use case. Scott complies while protecting client confidentiality, explaining how SigOpt tunes credit card fraud detection algorithms without replacing existing models.
Black Box Bayesian Optimization Mechanics 6725 Nathan challenges machine learning buzzwords and questions how a generalized platform works without bespoke consulting. Scott educates him on black-box Bayesian optimization, prompting Nathan to synthesize the mechanics using a waterslide analogy.
Fundraising History and Proprietary IP Security 5636 Nathan bluntly challenges whether SigOpt risks cross-contaminating trade secrets between direct competitors like insurance or trading firms. Scott firmly reframes how the architecture abstracts models so proprietary IP never touches SigOpt's servers.
Customer Count, Churn, and Team Distribution 6213 Nathan calculates SigOpt's monthly recurring revenue at fifty thousand dollars from twelve customers and probes on contract discounting and churn. Scott confirms the numbers and details their team allocation across engineering and sales.
Venture Capital Dynamics and Headcount Burn 7337 Nathan aggressively asks if Andreessen Horowitz ignored unit economics like CAC and LTV just to chase AI hype. Scott rejects the premise by defending their due diligence, while Nathan calculates a monthly payroll burn of around one hundred thirty thousand dollars.
The Famous Five Rapid-Fire Questions 3112 Nathan runs through the Famous Five rapid-fire format smoothly, with Scott sharing insights on building sustainable operational habits and leadership readings.

Statements from this episode (11)

Disclosure
SigOpt pricing starts at $2,500/mo, with enterprise plans around $10,000/mo
“Our work group pricing that we publish on our website starts at 2500 dollars a month getting you a little more than a dozen models a month enterprise plans ramp up from there around the 10,000 dollar a month mark is typical.”
Scott Clark Jun 29, 2017 ▶ 1:54
Assertion Not checkable as stated
Insurance giants like Prudential are actively investing in machine learning
“Prudential is really investing in machine learning and data science. I think we see this across the board in a variety of different insurance companies where some of the more traditional models are being augmented by the amount of new data that's Being able to…”
Scott Clark Jun 29, 2017 ▶ 3:23
Disclosure
SigOpt fine-tunes existing client models instead of entirely replacing them
“So instead of just taking a raw data set of decades of fraud data and giving them some model to like rip and replace what they already have, we sit on top of what they have and provide this additive boost by fine tuning it.”
Scott Clark Jun 29, 2017 ▶ 5:37
Insight
Clark: Humans are bad at performing 10-dimensional optimization mentally
“Turns out humans are pretty bad at doing 10 dimensional optimization in their head.”
Scott Clark Jun 29, 2017 ▶ 6:28
Assertion Supported
Black-box Bayesian optimization outperforms standard machine learning tuning techniques
“What we're able to do is without any domain expertise, without making any assumptions about that underlying model, outperform these standard techniques by providing this ensemble of black box Bayesian optimization strategies.”
Scott Clark Jun 29, 2017 ▶ 6:42
Disclosure
SigOpt has raised $8.8M to date, led by Andreessen Horowitz
“We have about 8.8 million dollars to date. Went through Y Combinator in winter 15. Andreessen Horowitz led our seed round immediately following that. They also led our series A last July.”
Scott Clark Jun 29, 2017 ▶ 9:58
Assertion Not checkable as stated
Algorithmic trading firms use SigOpt because client data never touches its systems
“This allows us to work with some of the most secretive algorithmic trading firms in the world, where their domain expertise and their models are literally how they make their billions of dollars, but they can still use SigOpt because all we're tuning are these…”
Scott Clark Jun 29, 2017 ▶ 11:12
Disclosure
SigOpt has about a dozen paying enterprise customers globally
“About a dozen customers around the world.”
Scott Clark Jun 29, 2017 ▶ 11:48
Assertion Supported
Huawei is a paying customer of SigOpt
“Huawei is another customer.”
Scott Clark Jun 29, 2017 ▶ 12:02
Assertion Not checkable as stated
SigOpt generates approximately $50,000 in monthly recurring revenue
“That's a fair assumption, yeah.”
Scott Clark Jun 29, 2017 ▶ 12:18
Assertion Not checkable as stated
SigOpt has experienced absolute zero customer churn to date
“No, that's the nice thing is once customers start putting this into their system to replace it, they have to go back to one of these Previous techniques, like trying to brute force the problem or something like that.”
Scott Clark Jun 29, 2017 ▶ 12:54
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 2,600 episodes transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.

Try the podcast clipper Learn more