May 27, 2014 · 20m · mad

Jason Tan, Sift Science // Data Driven #26 // April 2014 (Hosted by FirstMark Capital)

Jason Tan · 17m spoken Matt Turck · 19s 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

At Data Driven NYC, Sift Science CEO Jason Tan delivers a comprehensive presentation on how machine learning principles transform online fraud detection. He contrasts legacy rule-based systems with dynamic probabilistic models, demonstrating how companies can build scalable, real-time machine learning architectures.

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 2.1% of the talking time here. How this is scored →

Matt as informed peer 0.1 Guest teaching 2.6 Guest disagreement 0.0 Matt pushing back 0.0
05100:0010:0020:001:21–3:37 · Matt as informed peer 0/10 Human Brain Learning and Pattern Recognition This is a solo presentation segment by guest Jason Tan explaining machine learning concepts using human brain pattern recognition. The host is absent during the monologue.3:37–6:03 · Matt as informed peer 0/10 Practical Application: Email Spam Detection Explained Jason illustrates machine learning feedback loops through email spam detection examples. As a presentation monologue, host participation and pushback are non-existent.6:03–8:35 · Matt as informed peer 0/10 Everyday Applications of Machine Learning Technology Jason highlights real-world machine learning applications before contrasting them with flawed rule-based fraud detection systems. The segment is a solo presentation monologue.8:35–10:40 · Matt as informed peer 0/10 Probabilistic Machine Learning Strategies for Fraud Detection Jason details probabilistic scoring and dynamic thresholding in fraud detection. The host does not intervene during this technical talk segment.10:40–12:40 · Matt as informed peer 0/10 Real-Time Online Learning and Combining Human Intuition Jason explains how human intuition pairs with algorithmic data, using email digit patterns as an example. The host remains silent throughout the presentation monologue.12:40–17:30 · Matt as informed peer 0/10 Technical Architecture for Building an In-House Fraud ML System Jason dives into technical architecture, signal selection, and decision trees. Because this is a continuous presentation monologue, host scores remain zero.17:30–20:15 · Matt as informed peer 1/10 Interactive Audience Q&A Session and Conclusion Host Matt Turck briefly enters to joke and facilitate Q&A from audience members Eric and Catherine. Jason politely educates the audience on algorithm trade-offs and pooled data benefits.1:21–3:37 · Guest teaching 2/10 Human Brain Learning and Pattern Recognition This is a solo presentation segment by guest Jason Tan explaining machine learning concepts using human brain pattern recognition. The host is absent during the monologue.3:37–6:03 · Guest teaching 2/10 Practical Application: Email Spam Detection Explained Jason illustrates machine learning feedback loops through email spam detection examples. As a presentation monologue, host participation and pushback are non-existent.6:03–8:35 · Guest teaching 3/10 Everyday Applications of Machine Learning Technology Jason highlights real-world machine learning applications before contrasting them with flawed rule-based fraud detection systems. The segment is a solo presentation monologue.8:35–10:40 · Guest teaching 3/10 Probabilistic Machine Learning Strategies for Fraud Detection Jason details probabilistic scoring and dynamic thresholding in fraud detection. The host does not intervene during this technical talk segment.10:40–12:40 · Guest teaching 3/10 Real-Time Online Learning and Combining Human Intuition Jason explains how human intuition pairs with algorithmic data, using email digit patterns as an example. The host remains silent throughout the presentation monologue.12:40–17:30 · Guest teaching 3/10 Technical Architecture for Building an In-House Fraud ML System Jason dives into technical architecture, signal selection, and decision trees. Because this is a continuous presentation monologue, host scores remain zero.17:30–20:15 · Guest teaching 2/10 Interactive Audience Q&A Session and Conclusion Host Matt Turck briefly enters to joke and facilitate Q&A from audience members Eric and Catherine. Jason politely educates the audience on algorithm trade-offs and pooled data benefits.1:21–3:37 · Guest disagreement 0/10 Human Brain Learning and Pattern Recognition This is a solo presentation segment by guest Jason Tan explaining machine learning concepts using human brain pattern recognition. The host is absent during the monologue.3:37–6:03 · Guest disagreement 0/10 Practical Application: Email Spam Detection Explained Jason illustrates machine learning feedback loops through email spam detection examples. As a presentation monologue, host participation and pushback are non-existent.6:03–8:35 · Guest disagreement 0/10 Everyday Applications of Machine Learning Technology Jason highlights real-world machine learning applications before contrasting them with flawed rule-based fraud detection systems. The segment is a solo presentation monologue.8:35–10:40 · Guest disagreement 0/10 Probabilistic Machine Learning Strategies for Fraud Detection Jason details probabilistic scoring and dynamic thresholding in fraud detection. The host does not intervene during this technical talk segment.10:40–12:40 · Guest disagreement 0/10 Real-Time Online Learning and Combining Human Intuition Jason explains how human intuition pairs with algorithmic data, using email digit patterns as an example. The host remains silent throughout the presentation monologue.12:40–17:30 · Guest disagreement 0/10 Technical Architecture for Building an In-House Fraud ML System Jason dives into technical architecture, signal selection, and decision trees. Because this is a continuous presentation monologue, host scores remain zero.17:30–20:15 · Guest disagreement 0/10 Interactive Audience Q&A Session and Conclusion Host Matt Turck briefly enters to joke and facilitate Q&A from audience members Eric and Catherine. Jason politely educates the audience on algorithm trade-offs and pooled data benefits.1:21–3:37 · Matt pushing back 0/10 Human Brain Learning and Pattern Recognition This is a solo presentation segment by guest Jason Tan explaining machine learning concepts using human brain pattern recognition. The host is absent during the monologue.3:37–6:03 · Matt pushing back 0/10 Practical Application: Email Spam Detection Explained Jason illustrates machine learning feedback loops through email spam detection examples. As a presentation monologue, host participation and pushback are non-existent.6:03–8:35 · Matt pushing back 0/10 Everyday Applications of Machine Learning Technology Jason highlights real-world machine learning applications before contrasting them with flawed rule-based fraud detection systems. The segment is a solo presentation monologue.8:35–10:40 · Matt pushing back 0/10 Probabilistic Machine Learning Strategies for Fraud Detection Jason details probabilistic scoring and dynamic thresholding in fraud detection. The host does not intervene during this technical talk segment.10:40–12:40 · Matt pushing back 0/10 Real-Time Online Learning and Combining Human Intuition Jason explains how human intuition pairs with algorithmic data, using email digit patterns as an example. The host remains silent throughout the presentation monologue.12:40–17:30 · Matt pushing back 0/10 Technical Architecture for Building an In-House Fraud ML System Jason dives into technical architecture, signal selection, and decision trees. Because this is a continuous presentation monologue, host scores remain zero.17:30–20:15 · Matt pushing back 0/10 Interactive Audience Q&A Session and Conclusion Host Matt Turck briefly enters to joke and facilitate Q&A from audience members Eric and Catherine. Jason politely educates the audience on algorithm trade-offs and pooled data benefits.

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

0:00 · Matt 0% · guest 100%0:00 · Matt 0% · guest 100%3:00 · Matt 0% · guest 100%3:00 · Matt 0% · guest 100%6:00 · Matt 0% · guest 100%6:00 · Matt 0% · guest 100%9:00 · Matt 0% · guest 100%9:00 · Matt 0% · guest 100%12:00 · Matt 0% · guest 100%12:00 · Matt 0% · guest 100%15:00 · Matt 10.4% · guest 89.6%15:00 · Matt 10.4% · guest 89.6%18:00 · Matt 5.7% · guest 94.3%18:00 · Matt 5.7% · guest 94.3%
Sharpest disagreement ▶ 19:33 Rejection of big data downside premise

Jason politely rejects Catherine's premise that big data can be counterproductive in predictive models, asserting that pooled data is always beneficial in fraud detection.

Hardest push from Matt ▶ 17:34 Host wraps up presentation and opens Q&A

Host Matt Turck intervenes after the rapid presentation, directing Jason to sit down and drink water before taking audience questions.

Biggest teaching moment ▶ 11:00 Explaining the logic behind programmatic email creation

Jason educates the audience on why machine learning alone isn't enough, explaining the fraudster behavior behind numerical email addresses.

Matt holds his own ▶ 17:34 Host re-engages audience with event context

Matt Turck steps back onto the stage, making a lighthearted remark about burning calories at Data Driven NYC while moderating the transition to Q&A.

the scores for every segment, with the reasoning behind each
ChapterTopicMatt as informed peerGuest teachingGuest disagreementMatt pushing backWhy
Human Brain Learning and Pattern Recognition 0200 This is a solo presentation segment by guest Jason Tan explaining machine learning concepts using human brain pattern recognition. The host is absent during the monologue.
Practical Application: Email Spam Detection Explained 0200 Jason illustrates machine learning feedback loops through email spam detection examples. As a presentation monologue, host participation and pushback are non-existent.
Everyday Applications of Machine Learning Technology 0300 Jason highlights real-world machine learning applications before contrasting them with flawed rule-based fraud detection systems. The segment is a solo presentation monologue.
Probabilistic Machine Learning Strategies for Fraud Detection 0300 Jason details probabilistic scoring and dynamic thresholding in fraud detection. The host does not intervene during this technical talk segment.
Real-Time Online Learning and Combining Human Intuition 0300 Jason explains how human intuition pairs with algorithmic data, using email digit patterns as an example. The host remains silent throughout the presentation monologue.
Technical Architecture for Building an In-House Fraud ML System 0300 Jason dives into technical architecture, signal selection, and decision trees. Because this is a continuous presentation monologue, host scores remain zero.
Interactive Audience Q&A Session and Conclusion 1200 Host Matt Turck briefly enters to joke and facilitate Q&A from audience members Eric and Catherine. Jason politely educates the audience on algorithm trade-offs and pooled data benefits.

Statements from this episode (6)

Assertion Supported
Tan: Sift Science has 25 employees and 175 customers including Twitter and Uber
“We have 25 people out in San Francisco just closed our Series B recently, and we have about a 175 customers around the world, including Twitter, Square, Match.com, OpenTable, Uber, Airbnb, Kickstarter and others.”
Jason Tan May 27, 2014 ▶ 0:19
Insight
Tan: Traditional rule-based fraud detection systems do not scale
“So this is the way that things are done, is really building these rule-based systems that do not scale, okay?”
Jason Tan May 27, 2014 ▶ 7:40
Assertion Not checkable as stated
Jason Tan: Email addresses containing digits have higher fraud rates
“If you have numbers in your email address, more likely to be a fraudster.”
Jason Tan May 27, 2014 ▶ 8:42
Assertion Not checkable as stated
Jason Tan: Gmail domains carry lower fraud risk than obscure domains
“Gmail is much more trustworthy than some of these other domains, right?”
Jason Tan May 27, 2014 ▶ 8:54
Assertion Supported
Tan: Google generates $55B annually due to its search data lead
“They make fifty-five billion dollars a year, because they have so much data by being the number one search engine.”
Jason Tan May 27, 2014 ▶ 13:19
Assertion Not checkable as stated
Jason Tan: Sift Science builds global models across 175 customers
“We build this global model that learns across all of our 175 global customers”
Jason Tan May 27, 2014 ▶ 19:37
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.