Ask Me Anything (AMA) with Jeffrey Wang, Co-founder and Chief Architect at Amplitude

Ask Me Anything (AMA) with Jeffrey Wang, Co-founder and Chief Architect at Amplitude
Userlevel 4
Badge +2

Hi fellow data monsters,

I’m Jeffrey, one of Amplitude’s co-founders and chief architect. I’m hosting an Ask Me Anything on March 31 in which I’ll be answering questions directly from you on anything relating to Amplitude—whether it’s technical questions about the product, my journey in becoming a co-founder to where we are today, where I see Amplitude headed in the future or anything else that comes to mind! I’m excited to leverage our new community as a way to connect with all of you. Here are a few tips to help you get the most out of this event:

  • If you have not yet signed up for the community, click Login in the top-right of the Amplitude Community and create an account using the email you use to log in to Amplitude
  • Once you’re logged in, you’ll be able to ask your question as a reply in this post
  • “Like” other questions that others have asked that you want to be answered
  • We’ll collect questions from now until the day of the event (March 31) - which I’ll then go through and answer as many questions as I can!
  • Our community team will be moderating questions and making sure they keep to our guidelines

Thanks and looking forward to connecting with all of you!


This topic has been closed for comments

23 replies

Userlevel 2
Badge

Hi Jeffrey,

One thing that interests me a lot is how you guys think about uncovering causal relationships versus finding correlation. In Amplitude, a lot of the tools (like Compass and Conversion Drivers) are really tools that discover correlations, and analysts – like myself – are left to make our own interpretations about whether there is in fact a causal relationship (and its direction). 

Most people in my industry use some form of randomized controlled trial (through A/B or multivariate tests) to get around this problem. However, running RCTs is both time consuming and there is a limit to how many you can run simultaneously (unless you have millions of users).  

I don’t know if you are familiar with the work of Judea Pearl on causal diagrams, but there are ways – given that you take the time to build a causal model – to understand how you can control for the right confounders to uncover actual causal relationships through observational studies. 

Have you given any thought to how Amplitude can help us analysts get around the old adage “"correlation does not imply causation" “? 

Thanks!

Userlevel 4
Badge +3

Hey Jeffrey - big fan here 🙂 What’s been the biggest positive change you’ve seen in your tenure at Amplitude?

Badge

Hey Jeffrey - what first drew you into the analytics space? What are some unique and interesting ways you’ve seen users use Amplitude for?

Userlevel 7
Badge +10

Hey Jeffrey,

What were the earliest pain points which the team tried solving which eventually led to building Amplitude?

Can you talk about Amplitude's first minimum viable product? If you can share early product wireframes/screenshots, that'd be really cool  :relaxed:

Userlevel 1
Badge

Hi Jeffrey, 

 

What do you think is an Amplitude feature users don’t utilize as much as they should, but is truly a great advantage to learning more about their business? 

Badge

Hi Jeffrey, 

 

For those thinking of joining a start-up, what’s been the most difficult thing about being a co-founder? What do you enjoy most?

 

Vince

Badge

Hi Jeffrey!

How do you divide your responsibilities clearly between yourself, CTO and VP of Engineering, and how did that come to be?

 

 

Badge

Hello, Jeffrey!

I am curious if you guys are planning to release any SDKs for some of the top gaming engines aside from Unreal and Unity ones?

 

Badge

Hi Jeffrey!

Wondering if the Amplitude team has thought about limiting read/write access with your current API Key architecture?

Badge

Hi Jeffrey, 

 

Where do you see Amplitude in 5 years in terms of the business and how that impacts the Tech Industry? How do you see your customers coming along that 5 year journey with Amplitude? Additionally what do you think will change within Amplitude i.e. the culture in the 5 years? 

Badge

Hi Jeffrey,

 

I worked in 2 different companies that are using Amplitude for almost 2.5 years and I really like it! 

However, I have a pain point that is really hard to solve: user properties can’t be backfilled in Amplitude. 

The common use case is: we have a lot of user properties based on business logics in the data warehouse and we send to Amplitude once per day. For example: acquisition channels,  AB test groups, etc. Because we can’t backfilled user properties, users won’t have the correct user properties in the events they triggered. We therefore need to use other tools to make the analysis that are related to these user properties. 

A way to work around to to send the user properties in the event as event properties, but it’s not always feasible depending on if the event can pick up the business logic. 

Is there a plan to handle this problem in Amplitude?

 

Userlevel 4
Badge +2

Hi Jeffrey,

One thing that interests me a lot is how you guys think about uncovering causal relationships versus finding correlation. In Amplitude, a lot of the tools (like Compass and Conversion Drivers) are really tools that discover correlations, and analysts – like myself – are left to make our own interpretations about whether there is in fact a causal relationship (and its direction). 

Most people in my industry use some form of randomized controlled trial (through A/B or multivariate tests) to get around this problem. However, running RCTs is both time consuming and there is a limit to how many you can run simultaneously (unless you have millions of users).  

I don’t know if you are familiar with the work of Judea Pearl on causal diagrams, but there are ways – given that you take the time to build a causal model – to understand how you can control for the right confounders to uncover actual causal relationships through observational studies. 

Have you given any thought to how Amplitude can help us analysts get around the old adage “"correlation does not imply causation" “? 

Thanks!

 

Hi @morvik, thanks for asking! This topic is top of mind for a couple of teams at Amplitude right now, and in many ways I consider causal analysis the “holy grail” of product analytics. We’re familiar with the techniques and literature in this area, which has advanced quite a bit in the last few years, and the primary challenge is figuring out how to turn the ideas into product.

We’ve already found success building causal graphs and doing one-off analyses to deduce causal relationships on top of our data. Now the team is exploring ways to expose this in our product in a self-service way that can generate reliable results. Amplitude has always been about bringing powerful analyses to non-technical users, so we hope to do that with causal this year!

Userlevel 4
Badge +2

Hi Jeffrey, 

 

What do you think is an Amplitude feature users don’t utilize as much as they should, but is truly a great advantage to learning more about their business? 

 

Hi @dsax, great question. There are so many things I could talk about here, but I’ll give you two takes on it. The first is the Engagement Matrix chart, which in my opinion is the best high-level overview of your product. It lets you quickly distill core actions, power user functionality, dead features, etc. You might only look at it once a month, but I’d recommend everyone periodically check in on their engagement matrix.

The second is related to everyone’s favorite problem, instrumentation and taxonomy. The #1 reason we see people fail to find value in Amplitude is poor instrumentation. We’ve spent a lot of time building out our Govern tool to help people do this, but at the heart of it is a cultural problem, not a technical one. The teams we work with that embrace the idea that instrumenting well IS doing good product analytics go far with their data!

Userlevel 4
Badge +2

Hey Jeffrey - big fan here 🙂 What’s been the biggest positive change you’ve seen in your tenure at Amplitude?

 

In the industry -- one of the phrases that my cofounder Curtis likes to use is that we’re “bringing science to the art of product development.” Not that inspiration and creativity aren’t key to building good products, but it’s been amazing to see how people have progressed over the last decade building digital products. We also liken it to the Moneyball trend in baseball, it’s about elevating the entire playing field, and in our case we hope that the products that everyone interacts with continue to get better as a result!

Userlevel 4
Badge +2

Hey Jeffrey - what first drew you into the analytics space? What are some unique and interesting ways you’ve seen users use Amplitude for?

 

Hey @JCordero, thanks for asking! Personally, I love solving hard engineering problems, and distributed systems has been an area where I get to work on a wealth of those. Analytics is particularly interesting because in my opinion it is one of the domains where strong technology is a key factor in business success, e.g. our ability to build Nova has opened up so many doors for the business/product. That combined with the fact that there was a new wave of interactive apps (especially mobile) in the early 2010s made it an exciting space to work on.

As for unique ways to use Amplitude, I think engineers typically are the most creative ones. Amplitude is neither a performance monitoring nor a crash reporting tool, but suffice it to say that plenty of people use it for that. At the end of the day, we record rows and let you perform aggregations on the data, so you can do quite a bit with that :slight_smile:

Userlevel 4
Badge +2

Hi Jeffrey!

How do you divide your responsibilities clearly between yourself, CTO and VP of Engineering, and how did that come to be?

 

 

 

Hey @yjwong, it’s definitely subtle so I get this question quite a bit. In our case, it actually ended up being pretty easy. First of all, neither I nor Curtis (my cofounder and CTO) ended up wanting to go into management long-term, so after the first ~4 years we stopped managing and made sure we hired an amazing engineering management.

The split between Chief Architect and CTO is kind of arbitrary, but what happens in practice at Amplitude is that I spend more time with the core team and infrastructure, while Curtis works more on R&D efforts to explore new products. There are no hard lines drawn, though, so it’s also dependent on what the priorities of the company are at any point in time.

Userlevel 4
Badge +2

Hey Jeffrey,

What were the earliest pain points which the team tried solving which eventually led to building Amplitude?

Can you talk about Amplitude's first minimum viable product? If you can share early product wireframes/screenshots, that'd be really cool  :relaxed:

 

Hey @Saish Redkar, this question really takes me back! The first folks we talked to about this were mostly ex-Facebook and ex-Zynga PMs who had started their own companies. Those two companies were the pioneers of product analytics, and the PMs who left really wanted tools that could match up with what Facebook/Zynga had internally. We spent all of the early days understanding that pain and how it wasn’t addressed by anything on the market, and then building with nonstop feedback.

For analytics, you actually have to have quite a bit to get to an MVP, e.g. an SDK for collecting data, a data pipeline for processing it, a query system to analyze it, and a UI to display it. Those are pretty much all the same pieces as we have today, but the first iteration was incredibly hacky on all fronts. As a fun little tidbit, we used a LOT of postgres, both as a queue and a query system. Sadly, documentation from the early days hasn’t really survived, but you can see some 2014 screenshots in this TechCrunch article.

Userlevel 4
Badge +2

Hi Jeffrey, 

 

For those thinking of joining a start-up, what’s been the most difficult thing about being a co-founder? What do you enjoy most?

 

Vince

 

Hi @Vsant1990, doing a start-up comes with all sorts of extremes. For me, the thing I enjoy most out of everything is working with a group of incredibly motivated and capable people. If you do it right, you can create an amazing work environment that brings out the best in everyone and is also really fun! That’s what I’m most proud of doing at Amplitude and something we’ve tried to maintain as we scale.

As for the most difficult thing, it’s probably the fact that it’s an absolute grind. If you want to start a company, don’t go into it thinking it’ll be hard for a year or two and then get easier. If anything, it consistently got harder for the first 5 years. Expectations increased, the demand for excellence increased, the complexity of the business increased, and over and over we had to deal with situations that none of us had ever seen before (not to mention COVID!). That’s not to say it wasn’t fun and rewarding, but without some real mental fortitude and the support of my cofounders, there’s no way I would’ve made it to today.

Userlevel 4
Badge +2

Hello, Jeffrey!

I am curious if you guys are planning to release any SDKs for some of the top gaming engines aside from Unreal and Unity ones?

 

 

Hey @Elena.Tudu, thanks for asking. Gaming has always been a big vertical for us since gaming companies are often at the forefront of product development and technological prowess. Is there a particular SDK you have in mind? I’m sure our team would love to learn more about where you’re interested in using Amplitude.

Userlevel 4
Badge +2

Hi Jeffrey!

Wondering if the Amplitude team has thought about limiting read/write access with your current API Key architecture?

 

Hey @floatingbui, we’ve heard that request a number of times, but have yet to prioritize it. Access control is becoming a bigger topic, though, so we’d love to understand your pain better. Feel free to reach out to me via email (jeffrey@) if you’re interested in sharing!

Userlevel 4
Badge +2

Hi Jeffrey, 

 

Where do you see Amplitude in 5 years in terms of the business and how that impacts the Tech Industry? How do you see your customers coming along that 5 year journey with Amplitude? Additionally what do you think will change within Amplitude i.e. the culture in the 5 years? 

 

Hey @pdwang, great question, and something I think about quite a bit. First of all, our goal is definitely long-term impact and also not limited to the tech industry. We believe all companies are becoming digital product companies, and so they’ll all need something like Amplitude to build great products.

Today, our customers rely on us to answer questions and learn about their product/users, but ultimately it’s up to them to make all the decisions about how to change their product. In 5 years, I’d love to see Amplitude do a much better job of “closing the loop,” i.e. turning insights into better products in a more automated way. This means being better integrated into how companies build product and make decisions. We have customers asking us for this stuff, so we’re excited to help them in even more ways than today.

As for within Amplitude, my hope is that we’re able to set up a culture of amazing distributed innovation as we grow. We have one of the largest and most diverse behavioral datasets in the world, and we believe that it contains the key to building amazing products. But we don’t know how to do that yet today, so I’ll be counting on all the future Ampliteers to figure this out, and setting up a culture for doing that is key.

Userlevel 4
Badge +2

Hi Jeffrey,

 

I worked in 2 different companies that are using Amplitude for almost 2.5 years and I really like it! 

However, I have a pain point that is really hard to solve: user properties can’t be backfilled in Amplitude. 

The common use case is: we have a lot of user properties based on business logics in the data warehouse and we send to Amplitude once per day. For example: acquisition channels,  AB test groups, etc. Because we can’t backfilled user properties, users won’t have the correct user properties in the events they triggered. We therefore need to use other tools to make the analysis that are related to these user properties. 

A way to work around to to send the user properties in the event as event properties, but it’s not always feasible depending on if the event can pick up the business logic. 

Is there a plan to handle this problem in Amplitude?

 

 

Hey @SiyunLi, glad to hear you are a happy user! And your pain point is very much heard, this has been a recurring theme of users in Amplitude. In the short term, I recommend building cohorts of the users satisfying your criteria and then doing analyses on top of those cohorts. That essentially joins your new user properties with the old events.

But that’s a bit of a hassle to do, and so I’m excited to let you know that we’re in the middle of building our the underlying infrastructure that would enable what you’re talking about. Use cases like current user properties and computed properties (aggregations) are on the horizon, so keep an eye out for this!

Userlevel 6
Badge +9

@paladin8 thank you Jeffery for taking the time to answer all these questions! I hope everyone has received the answers they were looking for.

I will now close this topic for comments, and thank you all again for taking the time to reach out to us!

Best,

Daniel