StarZero All shows Try the podcast clipper →

Sep 30, 2016 · 22m · mad

Making On-Demand Delivery Profitable // Jeremy Stanley, Instacart (Data Driven NYC / FirstMark)

Jeremy Stanley · 17m spoken Matt Turck · 1m 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 a Data Driven NYC event, Instacart VP of Data Science Jeremy Stanley explains how advanced machine learning, predictive demand forecasting, and vehicle routing algorithms turned on-demand grocery delivery into a profitable business model while navigating complex multi-sided marketplace dynamics.

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 6.4% of the talking time here. How this is scored →

Matt as informed peer 1.0 Guest teaching 0.8 Guest disagreement 0.3 Matt pushing back 0.3
05100:0010:0020:000:30–3:03 · Matt as informed peer 0/10 Mapping the Customer and Shopper Experiences This is a solo presentation by Stanley outlining Instacart's four-sided marketplace model. The host does not participate in this segment.3:03–5:56 · Matt as informed peer 0/10 Unit Economics and the Path to Profitability Stanley presents data on unit economics, fulfillment speed improvements, and revenue growth. The segment is entirely monologue with zero host interaction.5:56–9:19 · Matt as informed peer 0/10 Demand Forecasting and Capacity Shock Absorbers Stanley details demand forecasting algorithms and capacity management techniques. The host is absent during this portion of the presentation.9:19–14:21 · Matt as informed peer 0/10 Optimizing Fulfillment Times and Vehicle Routing Stanley concludes his talk by walking through vehicle routing problems and recruitment efforts. Because it is a monologue, host scores remain zero.14:21–18:14 · Matt as informed peer 5/10 Presentation Conclusion and Recruitment Call Host Matt Turck engages Stanley in a dialogue, offering context from previous speaker sessions about organizational design and tech stacks. Stanley explains their organizational model and choices regarding infrastructure like Spark, while gently reframing a hypothesis about data scientists becoming startup CEOs.18:14–22:44 · Matt as informed peer 1/10 Audience Q&A and Event Conclusion The segment consists of audience Q&A moderated by Turck. Stanley answers questions about delivery batching, retail partner models, and startup hiring advice.0:30–3:03 · Guest teaching 0/10 Mapping the Customer and Shopper Experiences This is a solo presentation by Stanley outlining Instacart's four-sided marketplace model. The host does not participate in this segment.3:03–5:56 · Guest teaching 0/10 Unit Economics and the Path to Profitability Stanley presents data on unit economics, fulfillment speed improvements, and revenue growth. The segment is entirely monologue with zero host interaction.5:56–9:19 · Guest teaching 0/10 Demand Forecasting and Capacity Shock Absorbers Stanley details demand forecasting algorithms and capacity management techniques. The host is absent during this portion of the presentation.9:19–14:21 · Guest teaching 0/10 Optimizing Fulfillment Times and Vehicle Routing Stanley concludes his talk by walking through vehicle routing problems and recruitment efforts. Because it is a monologue, host scores remain zero.14:21–18:14 · Guest teaching 3/10 Presentation Conclusion and Recruitment Call Host Matt Turck engages Stanley in a dialogue, offering context from previous speaker sessions about organizational design and tech stacks. Stanley explains their organizational model and choices regarding infrastructure like Spark, while gently reframing a hypothesis about data scientists becoming startup CEOs.18:14–22:44 · Guest teaching 2/10 Audience Q&A and Event Conclusion The segment consists of audience Q&A moderated by Turck. Stanley answers questions about delivery batching, retail partner models, and startup hiring advice.0:30–3:03 · Guest disagreement 0/10 Mapping the Customer and Shopper Experiences This is a solo presentation by Stanley outlining Instacart's four-sided marketplace model. The host does not participate in this segment.3:03–5:56 · Guest disagreement 0/10 Unit Economics and the Path to Profitability Stanley presents data on unit economics, fulfillment speed improvements, and revenue growth. The segment is entirely monologue with zero host interaction.5:56–9:19 · Guest disagreement 0/10 Demand Forecasting and Capacity Shock Absorbers Stanley details demand forecasting algorithms and capacity management techniques. The host is absent during this portion of the presentation.9:19–14:21 · Guest disagreement 0/10 Optimizing Fulfillment Times and Vehicle Routing Stanley concludes his talk by walking through vehicle routing problems and recruitment efforts. Because it is a monologue, host scores remain zero.14:21–18:14 · Guest disagreement 1/10 Presentation Conclusion and Recruitment Call Host Matt Turck engages Stanley in a dialogue, offering context from previous speaker sessions about organizational design and tech stacks. Stanley explains their organizational model and choices regarding infrastructure like Spark, while gently reframing a hypothesis about data scientists becoming startup CEOs.18:14–22:44 · Guest disagreement 1/10 Audience Q&A and Event Conclusion The segment consists of audience Q&A moderated by Turck. Stanley answers questions about delivery batching, retail partner models, and startup hiring advice.0:30–3:03 · Matt pushing back 0/10 Mapping the Customer and Shopper Experiences This is a solo presentation by Stanley outlining Instacart's four-sided marketplace model. The host does not participate in this segment.3:03–5:56 · Matt pushing back 0/10 Unit Economics and the Path to Profitability Stanley presents data on unit economics, fulfillment speed improvements, and revenue growth. The segment is entirely monologue with zero host interaction.5:56–9:19 · Matt pushing back 0/10 Demand Forecasting and Capacity Shock Absorbers Stanley details demand forecasting algorithms and capacity management techniques. The host is absent during this portion of the presentation.9:19–14:21 · Matt pushing back 0/10 Optimizing Fulfillment Times and Vehicle Routing Stanley concludes his talk by walking through vehicle routing problems and recruitment efforts. Because it is a monologue, host scores remain zero.14:21–18:14 · Matt pushing back 2/10 Presentation Conclusion and Recruitment Call Host Matt Turck engages Stanley in a dialogue, offering context from previous speaker sessions about organizational design and tech stacks. Stanley explains their organizational model and choices regarding infrastructure like Spark, while gently reframing a hypothesis about data scientists becoming startup CEOs.18:14–22:44 · Matt pushing back 0/10 Audience Q&A and Event Conclusion The segment consists of audience Q&A moderated by Turck. Stanley answers questions about delivery batching, retail partner models, and startup hiring advice.

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 4.5% · guest 95.5%12:00 · Matt 4.5% · guest 95.5%15:00 · Matt 34.3% · guest 65.7%15:00 · Matt 34.3% · guest 65.7%18:00 · Matt 6.2% · guest 93.8%18:00 · Matt 6.2% · guest 93.8%21:00 · Matt 5.8% · guest 94.2%21:00 · Matt 5.8% · guest 94.2%
Sharpest disagreement ▶ 16:50 Slight reframe on data scientist CEOs

Stanley playfully pushes back against Turck's hypothesis about data scientists becoming CEOs by suggesting instead that CEOs will need to act like data scientists.

 Hardest push from Matt ▶ 17:42 Prompting guest to justify avoiding Spark

Turck immediately presses Stanley to explain why Instacart avoids using Spark when possible, forcing a breakdown of the tool's additional operational complexity.

 Biggest teaching moment ▶ 22:15 Advising startups on data science hiring timing

Stanley offers definitive guidance to the audience, explaining why early-stage startups should not hire data scientists right away and should instead focus on product traction first.

 Matt holds his own ▶ 15:30 Synthesizing cross-interview insights on org design

Turck demonstrates domain expertise by synthesizing broader patterns from months of interviews regarding the isolation of data science teams in corporate silos.
the scores for every segment, with the reasoning behind each
ChapterTopicMatt as informed peerGuest teachingGuest disagreementMatt pushing backWhy
Mapping the Customer and Shopper Experiences 0000 This is a solo presentation by Stanley outlining Instacart's four-sided marketplace model. The host does not participate in this segment.
Unit Economics and the Path to Profitability 0000 Stanley presents data on unit economics, fulfillment speed improvements, and revenue growth. The segment is entirely monologue with zero host interaction.
Demand Forecasting and Capacity Shock Absorbers 0000 Stanley details demand forecasting algorithms and capacity management techniques. The host is absent during this portion of the presentation.
Optimizing Fulfillment Times and Vehicle Routing 0000 Stanley concludes his talk by walking through vehicle routing problems and recruitment efforts. Because it is a monologue, host scores remain zero.
Presentation Conclusion and Recruitment Call 5312 Host Matt Turck engages Stanley in a dialogue, offering context from previous speaker sessions about organizational design and tech stacks. Stanley explains their organizational model and choices regarding infrastructure like Spark, while gently reframing a hypothesis about data scientists becoming startup CEOs.
Audience Q&A and Event Conclusion 1210 The segment consists of audience Q&A moderated by Turck. Stanley answers questions about delivery batching, retail partner models, and startup hiring advice.

Statements from this episode (12)

Assertion Not checkable as stated
Instacart achieved profitable unit economics in summer 2016
“We've achieved profitable unit economics, it happened in the summer”
Jeremy Stanley Sep 30, 2016 ▶ 4:17
Assertion Not checkable as stated
Instacart grew revenue by 500 percent between early 2015 and late 2016
“Since the beginning of last year, we've grown revenue 500%.”
Jeremy Stanley Sep 30, 2016 ▶ 4:39
Assertion Not checkable as stated
Ninety percent of Instacart customers are repeat buyers
“90% of our customers are repeat customers.”
Jeremy Stanley Sep 30, 2016 ▶ 4:48
Assertion Not checkable as stated
Instacart's repeat customers spend around $500 per month
“They spend about 500 dollars a month with us”
Jeremy Stanley Sep 30, 2016 ▶ 4:50
Prediction Didn’t hold up
Instacart projects reaching profitability and positive cash flow by late 2017
“And in the next 12 months, we're gonna be a profitable company, cash flow positive.”
Jeremy Stanley Sep 30, 2016 ▶ 4:50
Disclosure
Instacart tests delivery discounts to incentivize demand during off-peak capacity
“Ah, and we're increasingly testing, ah, reversing it, and actually giving discounts when we have a lot of capacity to try to incentivize demand to times when we have lots of shoppers available.”
Jeremy Stanley Sep 30, 2016 ▶ 9:10
Assertion Not checkable as stated
Instacart's routing models predict Manhattan travel times twice as well as Google
“If you just take the predictions out of Google Maps API, you'll be able to explain about 25% of the variance. If you look at the models that we built internally that use a lot more of our historical data, we can get up to about 50% of the variance explained.”
Jeremy Stanley Sep 30, 2016 ▶ 10:20
Disclosure
Instacart recomputes routing plans every minute in every market
“We recompute our plans every minute in every one of our markets”
Jeremy Stanley Sep 30, 2016 ▶ 13:21
Insight
Unifying objectives into one function beats coarse optimization constraints
“Oftentimes you start out with a system like this, and because it's such a huge space to explore, you put in a lot of really coarse constraints up front. You know, don't do this if x, right? Most of those constraints are reasonable, but bad 10% of the time, and…”
Jeremy Stanley Sep 30, 2016 ▶ 13:30
Assertion Not checkable as stated
Instacart's 100-person engineering team includes 25 data professionals
“Engineering is around say a hundred people. Out of that there's about 25 that are data related. About five are data engineering. About 10 are analytics, which is really decision science, helping to inform the decisions that the product managers or operations t…”
Jeremy Stanley Sep 30, 2016 ▶ 14:51
Assertion Not checkable as stated
Instacart's data stack relies on Postgres, AWS Redshift, Python, and R
“All of our production data is in Postgres, and we will read from the secondaries in AWS. We push all of the Postgres data into Redshift and do a lot of more intensive batch computations off of Redshift. The data science teams are using either Python or R we do…”
Jeremy Stanley Sep 30, 2016 ▶ 17:10
Insight
Startups should not hire data scientists before reaching an MVP
“Basically my advice is not to start a startup with a data scientist. You know, focus on getting to MVP, getting some traction, generating some real data, make good decisions early on, and then hire data scientists as you scale.”
Jeremy Stanley Sep 30, 2016 ▶ 22:26
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.

Try the podcast clipper Learn more