Dec 5, 2013 · 20m · mad

Panel Discussion // NYC Data Business Meetup // Feb 2013

Shawn Budde · 6m spoken John Fawcett · 3m spoken Zach Perret · 2m spoken Matt Turck · 1m spoken Matt Kroll · 32s spoken
0:00 / 0:00
▶ Watch on YouTube →

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

Hosted by Matt Turck at the NYC Data Business Meetup, this panel discussion features founders from Plaid, Quantopian, and ZestFinance sharing their entrepreneurial journeys and discussing data-driven innovation in financial technology. The speakers address technical architecture, regulatory compliance, machine learning validation, and the challenges of disrupting legacy financial institutions.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 7.2% of the talking time here. How this is scored →

Matt as informed peer 0.3 Guest teaching 5.1 Guest disagreement 2.3 Matt pushing back 0.3
05100:0010:0020:001:54–4:28 · Matt as informed peer 1/10 Founding Stories of Quantopian and ZestFinance The panel opens with Zach, John, and Shawn sharing their respective company founding stories. Host Matt Turck acts primarily as a warm moderator, making a lighthearted banter comment at the end about VCs adding value.4:28–6:52 · Matt as informed peer 0/10 Q&A: ZestFinance on Model Compliance and Risk An audience member asks Shawn Budde about model compliance and handling 100,000 variables under lending regulations like redlining. Budde provides an educational technical answer detailing holdout sets, jitter techniques, and adverse action compliance without any host intervention.6:52–9:16 · Matt as informed peer 0/10 Q&A: Quantopian on Backtesting Data and Algorithmic Trading John Fawcett answers back-to-back technical audience questions regarding Quantopian's historical data depth, intraday trading capabilities, and integration with complex event processing engines like Streambase. The dynamic is collaborative and informational.9:16–12:06 · Matt as informed peer 0/10 Q&A: Plaid on Use Cases, Security, and Privacy Zach Perret from Plaid explains customer use cases and responds to an audience question about privacy and security by describing Plaid's read-only, encrypted database architecture. Host Matt Turck only intervenes to ask the questioner to state her name and company.12:06–14:08 · Matt as informed peer 0/10 Q&A: Skepticism on Predictive Modeling and Validation An audience data scientist challenges ZestFinance's model validation logic as circular tautology. Shawn Budde responds firmly, breaking down their strict 70/15/15 cross-validation protocol and sharing real-world performance metrics showing fraud dropping 98% overnight.14:08–18:53 · Matt as informed peer 1/10 Q&A: Scaling Big Data Underwriting to Institutional Finance Audience members ask why legacy financial institutions haven't adopted big data underwriting and suggest ZestFinance is more of a legal firm than a tech firm. Shawn Budde explicitly rejects the premise, pointing out that major banks only use basic decision trees while ZestFinance ensembles advanced ML algorithms like SVMs and random forests.18:53–20:37 · Matt as informed peer 0/10 Q&A: Computational Latency and Model Deployment Speed In response to a final question about computational latency, Shawn Budde explains how ZestFinance deploys R models using Rserve on AWS clusters to score loans in three seconds, contrasting this with traditional bank 18-month deployment cycles.1:54–4:28 · Guest teaching 1/10 Founding Stories of Quantopian and ZestFinance The panel opens with Zach, John, and Shawn sharing their respective company founding stories. Host Matt Turck acts primarily as a warm moderator, making a lighthearted banter comment at the end about VCs adding value.4:28–6:52 · Guest teaching 6/10 Q&A: ZestFinance on Model Compliance and Risk An audience member asks Shawn Budde about model compliance and handling 100,000 variables under lending regulations like redlining. Budde provides an educational technical answer detailing holdout sets, jitter techniques, and adverse action compliance without any host intervention.6:52–9:16 · Guest teaching 5/10 Q&A: Quantopian on Backtesting Data and Algorithmic Trading John Fawcett answers back-to-back technical audience questions regarding Quantopian's historical data depth, intraday trading capabilities, and integration with complex event processing engines like Streambase. The dynamic is collaborative and informational.9:16–12:06 · Guest teaching 5/10 Q&A: Plaid on Use Cases, Security, and Privacy Zach Perret from Plaid explains customer use cases and responds to an audience question about privacy and security by describing Plaid's read-only, encrypted database architecture. Host Matt Turck only intervenes to ask the questioner to state her name and company.12:06–14:08 · Guest teaching 7/10 Q&A: Skepticism on Predictive Modeling and Validation An audience data scientist challenges ZestFinance's model validation logic as circular tautology. Shawn Budde responds firmly, breaking down their strict 70/15/15 cross-validation protocol and sharing real-world performance metrics showing fraud dropping 98% overnight.14:08–18:53 · Guest teaching 7/10 Q&A: Scaling Big Data Underwriting to Institutional Finance Audience members ask why legacy financial institutions haven't adopted big data underwriting and suggest ZestFinance is more of a legal firm than a tech firm. Shawn Budde explicitly rejects the premise, pointing out that major banks only use basic decision trees while ZestFinance ensembles advanced ML algorithms like SVMs and random forests.18:53–20:37 · Guest teaching 5/10 Q&A: Computational Latency and Model Deployment Speed In response to a final question about computational latency, Shawn Budde explains how ZestFinance deploys R models using Rserve on AWS clusters to score loans in three seconds, contrasting this with traditional bank 18-month deployment cycles.1:54–4:28 · Guest disagreement 1/10 Founding Stories of Quantopian and ZestFinance The panel opens with Zach, John, and Shawn sharing their respective company founding stories. Host Matt Turck acts primarily as a warm moderator, making a lighthearted banter comment at the end about VCs adding value.4:28–6:52 · Guest disagreement 2/10 Q&A: ZestFinance on Model Compliance and Risk An audience member asks Shawn Budde about model compliance and handling 100,000 variables under lending regulations like redlining. Budde provides an educational technical answer detailing holdout sets, jitter techniques, and adverse action compliance without any host intervention.6:52–9:16 · Guest disagreement 1/10 Q&A: Quantopian on Backtesting Data and Algorithmic Trading John Fawcett answers back-to-back technical audience questions regarding Quantopian's historical data depth, intraday trading capabilities, and integration with complex event processing engines like Streambase. The dynamic is collaborative and informational.9:16–12:06 · Guest disagreement 1/10 Q&A: Plaid on Use Cases, Security, and Privacy Zach Perret from Plaid explains customer use cases and responds to an audience question about privacy and security by describing Plaid's read-only, encrypted database architecture. Host Matt Turck only intervenes to ask the questioner to state her name and company.12:06–14:08 · Guest disagreement 4/10 Q&A: Skepticism on Predictive Modeling and Validation An audience data scientist challenges ZestFinance's model validation logic as circular tautology. Shawn Budde responds firmly, breaking down their strict 70/15/15 cross-validation protocol and sharing real-world performance metrics showing fraud dropping 98% overnight.14:08–18:53 · Guest disagreement 6/10 Q&A: Scaling Big Data Underwriting to Institutional Finance Audience members ask why legacy financial institutions haven't adopted big data underwriting and suggest ZestFinance is more of a legal firm than a tech firm. Shawn Budde explicitly rejects the premise, pointing out that major banks only use basic decision trees while ZestFinance ensembles advanced ML algorithms like SVMs and random forests.18:53–20:37 · Guest disagreement 1/10 Q&A: Computational Latency and Model Deployment Speed In response to a final question about computational latency, Shawn Budde explains how ZestFinance deploys R models using Rserve on AWS clusters to score loans in three seconds, contrasting this with traditional bank 18-month deployment cycles.1:54–4:28 · Matt pushing back 1/10 Founding Stories of Quantopian and ZestFinance The panel opens with Zach, John, and Shawn sharing their respective company founding stories. Host Matt Turck acts primarily as a warm moderator, making a lighthearted banter comment at the end about VCs adding value.4:28–6:52 · Matt pushing back 0/10 Q&A: ZestFinance on Model Compliance and Risk An audience member asks Shawn Budde about model compliance and handling 100,000 variables under lending regulations like redlining. Budde provides an educational technical answer detailing holdout sets, jitter techniques, and adverse action compliance without any host intervention.6:52–9:16 · Matt pushing back 0/10 Q&A: Quantopian on Backtesting Data and Algorithmic Trading John Fawcett answers back-to-back technical audience questions regarding Quantopian's historical data depth, intraday trading capabilities, and integration with complex event processing engines like Streambase. The dynamic is collaborative and informational.9:16–12:06 · Matt pushing back 0/10 Q&A: Plaid on Use Cases, Security, and Privacy Zach Perret from Plaid explains customer use cases and responds to an audience question about privacy and security by describing Plaid's read-only, encrypted database architecture. Host Matt Turck only intervenes to ask the questioner to state her name and company.12:06–14:08 · Matt pushing back 0/10 Q&A: Skepticism on Predictive Modeling and Validation An audience data scientist challenges ZestFinance's model validation logic as circular tautology. Shawn Budde responds firmly, breaking down their strict 70/15/15 cross-validation protocol and sharing real-world performance metrics showing fraud dropping 98% overnight.14:08–18:53 · Matt pushing back 1/10 Q&A: Scaling Big Data Underwriting to Institutional Finance Audience members ask why legacy financial institutions haven't adopted big data underwriting and suggest ZestFinance is more of a legal firm than a tech firm. Shawn Budde explicitly rejects the premise, pointing out that major banks only use basic decision trees while ZestFinance ensembles advanced ML algorithms like SVMs and random forests.18:53–20:37 · Matt pushing back 0/10 Q&A: Computational Latency and Model Deployment Speed In response to a final question about computational latency, Shawn Budde explains how ZestFinance deploys R models using Rserve on AWS clusters to score loans in three seconds, contrasting this with traditional bank 18-month deployment cycles.

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

0:00 · Matt 26.4% · guest 73.6%0:00 · Matt 26.4% · guest 73.6%3:00 · Matt 9.6% · guest 90.4%3:00 · Matt 9.6% · guest 90.4%6:00 · Matt 0% · guest 100%6:00 · Matt 0% · guest 100%9:00 · Matt 1.7% · guest 98.3%9:00 · Matt 1.7% · guest 98.3%12:00 · Matt 0.1% · guest 99.9%12:00 · Matt 0.1% · guest 99.9%15:00 · Matt 0.6% · guest 99.4%15:00 · Matt 0.6% · guest 99.4%18:00 · Matt 12.6% · guest 87.4%18:00 · Matt 12.6% · guest 87.4%
Sharpest disagreement ▶ 16:54 Shawn rejects questioner's premise on bank ML knowledge

Shawn Budde explicitly pushes back against an audience member's assertion, stating 'I'm gonna disagree with the assertion that everybody knows how to do this' before detailing how major banks lag in machine learning techniques.

Hardest push from Matt ▶ 4:18 Matt Turck banter on VC value add

In a panel dominated by audience Q&A, the host's primary direct engagement is playfully challenging the guest's origin story narrative by asking if it is a rare case of VCs actually adding value.

Biggest teaching moment ▶ 13:01 Shawn refutes model validation tautology claim

When an audience data scientist accuses ZestFinance of circular reasoning, Shawn educates her on their strict 70/15/15 split validation process and demonstrates real-world empirical proof with a 98% overnight drop in fraud.

Matt holds his own ▶ 4:18 Matt Turck's witty VC value framing

Matt Turck steps in at the end of the founding stories to reframe the guest's anecdote with a sharp, humorous takeaway about VC contributions.

the scores for every segment, with the reasoning behind each
ChapterTopicMatt as informed peerGuest teachingGuest disagreementMatt pushing backWhy
Founding Stories of Quantopian and ZestFinance 1111 The panel opens with Zach, John, and Shawn sharing their respective company founding stories. Host Matt Turck acts primarily as a warm moderator, making a lighthearted banter comment at the end about VCs adding value.
Q&A: ZestFinance on Model Compliance and Risk 0620 An audience member asks Shawn Budde about model compliance and handling 100,000 variables under lending regulations like redlining. Budde provides an educational technical answer detailing holdout sets, jitter techniques, and adverse action compliance without any host intervention.
Q&A: Quantopian on Backtesting Data and Algorithmic Trading 0510 John Fawcett answers back-to-back technical audience questions regarding Quantopian's historical data depth, intraday trading capabilities, and integration with complex event processing engines like Streambase. The dynamic is collaborative and informational.
Q&A: Plaid on Use Cases, Security, and Privacy 0510 Zach Perret from Plaid explains customer use cases and responds to an audience question about privacy and security by describing Plaid's read-only, encrypted database architecture. Host Matt Turck only intervenes to ask the questioner to state her name and company.
Q&A: Skepticism on Predictive Modeling and Validation 0740 An audience data scientist challenges ZestFinance's model validation logic as circular tautology. Shawn Budde responds firmly, breaking down their strict 70/15/15 cross-validation protocol and sharing real-world performance metrics showing fraud dropping 98% overnight.
Q&A: Scaling Big Data Underwriting to Institutional Finance 1761 Audience members ask why legacy financial institutions haven't adopted big data underwriting and suggest ZestFinance is more of a legal firm than a tech firm. Shawn Budde explicitly rejects the premise, pointing out that major banks only use basic decision trees while ZestFinance ensembles advanced ML algorithms like SVMs and random forests.
Q&A: Computational Latency and Model Deployment Speed 0510 In response to a final question about computational latency, Shawn Budde explains how ZestFinance deploys R models using Rserve on AWS clusters to score loans in three seconds, contrasting this with traditional bank 18-month deployment cycles.

Statements from this episode (11)

Disclosure
Plaid was built after discovering developers lacked financial data tools
“What we're building now was an evolution because we simply just didn't have the tools to build what we wanted to build. And we talked to a bunch of other developers and found out that they didn't have the tools either.”
Zach Perret Dec 5, 2013 ▶ 1:26
Assertion Not checkable as stated
Budde: ZestFinance does not scrape Facebook data for credit decisions
“Everybody kind of, everybody first off assumes that we're scraping Facebook. We're not.”
Shawn Budde Dec 5, 2013 ▶ 6:19
Insight
Budde: Top credit variables hold 3-5% weight in ZestFinance models
“The data that has the most impact Is the same data that has always had the most impact. It's just instead of being 10 or 15 or 20% of the equation, it's, you know, it's three to five percent of the equation.”
Shawn Budde Dec 5, 2013 ▶ 6:28
Disclosure
Perret: Plaid spends 80% of its time on B2B products
“So we spend about 80% of our time building towards the business to business side.”
Zach Perret Dec 5, 2013 ▶ 10:26
Disclosure
Perret: Plaid has six active developers in early 2013
“So we have six people kind of actively in development and a big list of people that we're trying to add to it as we deal with scaling right now, which is all of our time.”
Zach Perret Dec 5, 2013 ▶ 11:51
Assertion Not checkable as stated
ZestFinance saw fraud drop 98% overnight after launching new model
“When we implemented in December, we saw fraud drop overnight by about 98% and we saw first payment failures drop by more than 40% you know, between December first and December second when we implemented the model.”
Shawn Budde Dec 5, 2013 ▶ 13:40
Assertion Not checkable as stated
Budde: Major lenders are not assembling advanced ML models for credit decisions
“Nobody is taking an SVM, ah, naive Bayesian, you know, A hidden mark off a random forest, and assembling those into a credit decision, as far as I know. We haven't come across them. I've talked to some of the biggest lenders.”
Shawn Budde Dec 5, 2013 ▶ 17:20
Disclosure
The CFPB approached ZestFinance offering regulatory protection for algorithmic underwriting
“We've, ah, we've actually been approached by the Consumer Financial Protection Bureau, ah, about, in essence, giving us kind of a waiver to say that, you know, we're gonna protect you from regulators who dispute this or don't like it.”
Shawn Budde Dec 5, 2013 ▶ 18:14
Disclosure
Budde: ZestFinance operates as a direct lender, not a software vendor
“We're a lender. So we're, you know, we're not in the business, ah, primarily in the business of selling models or techniques. We're not a consulting company. We are actually making loans, you know, on our own portfolio, and that's where we believe the money is…”
Shawn Budde Dec 5, 2013 ▶ 18:38
Disclosure
Shawn Budde: ZestFinance scores credit models in three seconds using R and AWS
“We build our models in R, and we implement our models in R. So, ah, we're using R serve you know, on Amazon clusters to score our models. Ah, we score in three seconds.”
Shawn Budde Dec 5, 2013 ▶ 19:22
Assertion Not checkable as stated
Shawn Budde: Typical banks update credit models only once every 18 months
“A typical bank is gonna do a new model once every 18 months or so, ah, and they're gonna spend three to six months coding it up, you know, into a production system, and, ah, and validating that they didn't miss a decimal.”
Shawn Budde Dec 5, 2013 ▶ 19:54
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 400 conversations 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.