Allen Yang

Activity Feed

On November 22, 2015, Allen Yang commented on Palantir – Masters of Big Data :

Good question – I don’t know of a specific threat. But, I would think that various parts of their service offering are being threatened by substitutes. For example, you might be able to cobble together something that looks like a Palantir solution using a combination of AWS for infrastructure + various open-source libraries or even smaller SaaS companies’ services. As commenter JG alluded to above, perhaps over time this means the cost of developing a Palantir-like solution internally will go down (but I still think you need hard-to-find and harder-to-attract engineering talent to do it), and Palantir will need to move “up-market”. Or, perhaps in time somebody like AWS will swoop in and provide a threatening end-to-end alternative to Palantir?

On November 22, 2015, Allen Yang commented on Palantir – Masters of Big Data :

I would guess that it will be a long time before businesses start to build this kind of expertise internally – it seems like it’d take a lot of special human capital to do this.

If anything, I think the biggest “threat” to Palantir is that eventually there are enough open-source mini-solutions that, when cobbled together, would allow a very small number of engineers to build up the same kind of data infrastructure that Palantir can provide.

But in the meantime, you’re right – businesses that don’t harness their data will probably be slowly “left behind”. And that’s great news for Palantir, because it’ll help attract more customers 🙂

On November 20, 2015, Allen Yang commented on SmartMat – quantifying the unquantifiable :

Whoa, really cool idea! I’m impressed by the hardware capabilities they’ve developed. And I did always find it weird that people “learn” yoga basically by being in a large, crowded room with dozens of other people, without receiving much if any personalized guidance or adjustments from a professional. This seems to help address some of that problem.

It seems like there are many parallels to this product in other kinds of sports – lots of companies are trying to insert sensors into different kinds of equipment, from tennis rackets to soccer balls. I wonder why people are going more for this approach (which is hardware intensive) instead of going for the video analysis approach (which would be more software intensive I imagine). For example, I imagine that there are many things which a yoga mat could sense, but also a realm of metrics that a video of somebody doing yoga could also measure. I also imagine that having a machine analyzing video footage of somebody playing tennis or soccer could also generate a lot of recommendations. With both the “embedded hardware” solution and the “video analysis” solution, the problems that people are trying to solve seem to be 1) collect lots and lots of data points then 2) analyze that data. I wonder if the video solution holds as much promise as the hardware solution (people might be working on it already – I’m not very in tune with this space!).

On November 20, 2015, Allen Yang commented on How Ibotta Is Changing the Coupon Game :

Really cool company, thanks for sharing. I’m really intrigued by the concept of giving consumers a bigger discount after participating in market studies. In some sense, maybe the discounted amount represents the economic value of that additional piece of data to the company. Right now this may be in the order of magnitude of tens of cents (I’m making up a number here), but over time perhaps this economic value will go towards 0 as it becomes easier and easier to collect information from end customers.

This does remind me a lot about digital advertising as well. It’s sort of like the video advertising model: in order to see a video, you HAVE to watch this ad. And (as Hulu does), if you tell us which of these 3 ads appeal to you the most, you can just watch ONE long-ish ad to avoid many short ads throughout your TV episode. I wonder why this engage-then-advertise model hasn’t caught on in more instances; perhaps users just really, really dislike it in other contexts?

On November 20, 2015, Allen Yang commented on CVS Health’s ExtraCare Program: Four Feet of Data Analytics :

Cool post, thanks for sharing! This reminds me a lot of what Target did to become known for its data analytics program as well. I love the idea of large chains using customer purchasing habits to improve both their sales and the user experience – seems like a win-win.

It’s interesting that CVS was “firing” its unprofitable customers. I would’ve expected that any coupon CVS decides to create would have its parameters set (i.e. discounted price, quantity of items needed) such that CVS could still guarantee itself some minimal margin. Or, in the case that the coupons actually represent a loss and are meant to encourage a larger basket filled with non-discounted items, I would expect the same effect to exist to some degree even if the customer used multiple coupons in one trip (e.g. the customer would come in for 4 items that he has coupons for, and end up buying 10 items total).

Put another way, I’m surprised that CVS couldn’t optimize its algorithm to take into account the behavior of customers who use “too many” coupons. Theoretically, if the algorithm can predict that the user qualifies for coupons A, B, C, D and E, but that the user is also likely to use all 5 coupons at once, then maybe the “optimal” coupon output would be C, D and E only.

Either way, if we extrapolate what data-driven discount generation looks like in the future, maybe we’re heading towards a world where every large merchant has enough data to perfectly understand its consumers, and then can try to adjust prices, discounts, and even what goods are displayed (easy in the case of Amazon) in an attempt to maximize profit from those consumers. The countervailing force would be competition, which would still help keep prices down for consumers. But the counter-countervailing force there is that in a world where data becomes more and more valuable, perhaps consolidation for the sake of aggregating data becomes a more frequent phenomenon.

On November 1, 2015, Allen Yang commented on StackOverflow : Q&A site for programmers :

Love Stack Overflow!

I agree with Austin’s point about programmers jumping around languages often, and thus needing to return to SO. I think SO also benefits from something that other crowdsourced knowledge sites (e.g. Wikipedia or Yelp) does not. With sites like Wikipedia and Yelp, the proportion of users who actually contribute to the knowledge base is very low relative to the number of pure viewers (probably somewhere in the 1-10% range). But, I would imagine that in SO that percentage is actually much higher, helped along by factors such as a stronger community feeling and the fact that a more defined niche of knowledge probably encourages relevant people to contribute more frequently.

What I really find amazing is that anybody answers any questions at all on SO. Many questions are either very niche (thus requiring a lot of work behind the scenes to potentially recreate the original poster’s problem) or broad (thus requiring a well-written response that probably requires explaining more background knowledge). On top of that, the community’s standards for responses seems incredibly high: other users are quick to “reply” to a response to the original question complaining that the respondent (who is purely trying to be helpful!) didn’t do X, Y or Z best practice in their response.

I find this last point particularly interesting because it breeds a community culture that is much more confrontational and, frankly, somewhat arrogant than other crowdsourced communities I’ve seen. Yet, it seems to work for SO. Perhaps this goes to show the importance of developing a fairly unique community culture to reflect the niche knowledge domain you’re targeting. And, perhaps understanding this is why SO’s parent company, Stack Exchange, has been able to move into so many other knowledge domains with the same type of Q&A forum (see:

On November 1, 2015, Allen Yang commented on Finding the Next Unicorn with Product Hunt :

I really enjoyed using Product Hunt during its earlier days and still check it daily, but recently something’s felt a bit off about the listings. It’s as if posting a product on Product Hunt is now an activity that can be “maximized” through a set of PR and marketing checkboxes. This was probably inevitable, but to me makes the website feel much less authentic, and makes me worried about its future.

This problem doesn’t just affect PH, but also any site that can be used as publicity for a new product (mostly in the tech space). Another prominent example is Kickstarter – today, being able to launch a “successful” Kickstarter campaign is really more of a marketing exercise. If you do X, Y and Z, the chances of your campaign succeeding are greatly magnified. There are now similar “best practices” for PH (e.g. posting at a specific time of day or day of week, having a certain kind of tagline, etc.). My worry here is that this detracts from the site’s ability to reliably signal what it’s trying to promote: Kickstarter for the “crazy idea” that people actually demand, and Product Hunt for the “cool new” product that you otherwise would not have heard of.

PH has recently gone into a couple new “types” of products such as Books and Games. What’s interesting to note is that these do not seem like they’ve caught on yet – the amount of activity on those lists are orders of magnitude smaller than for the core Tech list. Part of me wonders if this is because these new lists don’t suffer as much from the problem of “best practices” that affects Tech.

This decrease in authenticity worries me, but it does solve another problem: how to differentiate the interesting “signal” products from the much less interesting “noise” products. During my daily check of PH, I’d guess that 95%+ of the products listed are not that interesting. But, the products that are interesting are usually those that have the most upvotes, which I’d guess are also the group of products that have used the “best practices” mentioned above. So perhaps this speaks to some balance that needs to be struck on sites like these, a balance between best practices that legitimately signal quality vs. not allowing these best practices to have so much influence that they can completely make a low-quality product look like a high-quality one.

On November 1, 2015, Allen Yang commented on Crowdsourcing At The U.S. Army :

Cool post – I had never thought about the government leveraging crowdsourcing before.

I completely agree with your point that cybersecurity is going to be one huge obstacle in the way of more crowdsourcing by the government. This seems to apply at a few different levels. First, there’s the obvious concern that if the government crowdsources the development of an app, developers could theoretically insert “backdoors” that can be used to compromise the app in the future (these backdoors could be intentional, or even the result of oversight in development). Second, there’s the concern that if you rely on a third party’s API (e.g. Google for Maps), you’re essentially giving Google some of the usage data for the apps, so perhaps there’s a concern of security leakage there. Third, perhaps there’s also the concern that posting a prompt for the crowdsourced development of an app is itself a leakage of information about what the military is working on and what strategy it’s using.

I think another major obstacle to using crowdsourcing to develop an app is the question of maintenance. Who is the best party to make future updates and fixes to a crowdsourced app? Locking the original developer into a long-term contract doesn’t seem to be the right answer. But, it might also be unrealistic to expect the government to develop these capabilities just to maintain a handful of crowdsourced apps. Perhaps future work can be crowdsourced again, but I imagine that would really compound the cybersecurity concerns above.

On October 2, 2015, Allen Yang commented on Ola Cabs: The “Copy-Cat” beating Uber at its own game in India. :

Hooray for network effect-driven competition benefitting the user!

Your points about why Ola was able to beat Uber were very interesting, and highlighted to me the importance of context-specific platform investments. I think people generally think that “platform R&D” (to borrow the term from the simulation) is always a generic push in the same direction. However, with Ola it seems like they’ve made R&D investments specifically catered towards the Indian context, whereas Uber has tried to adapt better to a credit card- and mobile-heavy userbase. Consequently, Ola has become more entrenched in India, while Uber has become more entrenched in places like the US.

So given that Ola’s model clearly works in India, it’s surprising that Uber is unable to just copy the same India-specific R&D investments to have better results against Ola, especially given that India is a huge priority for Uber and Uber has a lot more cash to work with. The problem, however, doesn’t seem to be “symmetric”: if we think that over time India will converge towards the US in terms of credit card and mobile penetration, I don’t think Ola would have a hard time mimicking Uber’s R&D investments to better cater towards that crowd. Perhaps it is a general fact that as long as you a) get the customers first, and b) continue changing your product based on your customers’ changing needs, you’ll be able to continue benefitting from the network effects.

This is interesting because it perhaps speaks to the cross-country limitations of network effects. Maybe this is to be expected for products that deal with people and regional logistics issues (e.g. Uber/Ola, maybe something like parcel delivery services). But, it also seems to be somewhat applicable to mostly software- or knowledge-based products. For example, Yelp seems to be having trouble expanding to certain non-Western geographies, probably because the customs and needs in other countries are different and not completely compatible with the R&D investments that it has made thus far.

This is an interesting example of how important it is to consider how direct network effects map onto users’ social “graphs”, and of how different products can have different critical thresholds for their network effects.

To the first point, as you mention, Fitbit tries to use competition with friends and family as a key motivator for people to buy their product and continue interacting with it. However, individual users are unlikely to feel the direct network effects until they actually have friends and family using the product. In other words, if 10 random people in the country are using Fitbit, they’re unlikely to be my connections, so I have little motivation there to adopt the product. Even at 10,000 random people, I might still have no motivation if my friends and family aren’t represented in that pool, but perhaps connections that are 1 or 2 degrees away from me have adopted the product and are slowly spreading the network effects towards my networks. So at this point, it’s entirely possible that the direct network effects have zero impact on my decision to buy, while another user might see a disproportionate number of their network using the product, and thus feel very strong direct network effects. This parallels the case with Facebook’s growth: when Facebook was just a Harvard undergrad product, there was no reason for somebody from Stanford to adopt it.

This also influences the second point, which is that different products have different thresholds for their direct network effects. For crowd-sourced knowledge platforms, the impact of an additional user who contributes an additional article or review (e.g. on Wikipedia or Yelp) directly benefits everyone else. For something like the telephone, the effect of a random additional user joining is weaker but still potentially valuable, to the extent that I sometimes have to reach out to “random” people or businesses. And finally, for something like Fitbit or Facebook, a random new user probably has close to zero effect on most people outside of that user’s immediate network, and so the threshold needed for direct network effects to occur is probably higher.

On October 2, 2015, Allen Yang commented on Coursera: using network effects to build access to education :

I completely agree that Coursera and the other MOOCs try to build in direct network effects by encouraging discussions and peer work. I wonder, though, if there are actually two kinds of users, the “casual” learners and the “serious” learners, and whether a MOOC actually wants to encourage the growth of both groups.

My experience with MOOC courses has generally been that the student discussions are a bit chaotic and all over the place in terms of background knowledge and understanding of the course materials. This might be caused by or compounded by the fact that the standards for “passing” a course are relatively low (a bit higher for certification, but certificate-track students and non-track students are generally grouped together for discussion and work, I believe), which leads to more “casual” learners. While I think discriminating against these kinds of students would go against the core principles of a MOOC, I do think their presence somewhat detracts from the experience of the “serious” learners, who have a harder time getting high-quality and equally-interest-matched feedback from other “serious” learner peers.

In essence, I think the direct network effects highlighted in your post actually hurt MOOCs, and I wonder if there’s a way or a precedent for a product to foster self-selection and network effects within just a subset of its user base. I don’t know if players like Coursera or EdX would do this anytime soon – they are currently focused on growth in users and revenue, not necessarily providing the best “serious” learner environment.

On September 13, 2015, Allen Yang commented on Get a Slice of This: Domino’s Dominates Pizza through Digital :

I’m also a big beneficiary of this service!

It’s funny because when Domino’s started releasing some of these features, it all seemed like a marketing gimmick (e.g. ordering via tweet – who actually does that?). But now, as you said, it’s clear that Domino’s is prioritizing their digital experience, and now it’s basically unfathomable to me to order pizza from anybody but Domino’s (unless you’re going for more “artisanal” pizza, but that’s really a different product I’d say).

I agree with Rob’s worry about their delivery asset utilization. My bigger curiosity is why Domino’s maintains its own delivery force at all – from companies like Seamless and Caviar, we’ve seen that restaurants can somewhat pool together their delivery assets, which saves all the partners some money. In some sense, platforms like Seamless and Caviar are similar in that they allow restaurants to “outsource” their online ordering systems – now a restaurant doesn’t have to develop its own system, but rather pays some fee to use Seamless’s.

So an interesting question is why Domino’s decided to build its own online ordering system. Yes, it gives them greater control over the experience and allows for interesting customizations, but is it worth the long-term cost to upkeep when companies like Seamless are offering basically the same functionality?

Because Domino’s decided to build its own solution, it again makes me think that a lot of this could almost be viewed as a marketing expense – the dedicated ordering user experience is a way to draw in and retain users. I’d be very interested to see, in the long run, whether its projects in this arena can help out operationally or in some other way – there’s much more to a “digital revolution” than marketing.

On September 13, 2015, Allen Yang commented on Digital Innovation in Online & Mobile Banking at J.P. Morgan Chase :

I’m also fascinated by how the consumer financial industry is going through a digital revolution. I’ve been a user of Ally Bank for a while now, so it’s interesting to hear about what Chase offers.

One thing that’s struck me about online banking is that it’s actually still far from being a full-featured “port” of brick and mortar banking onto the digital space. In other words, there are still a number of banking actions that can only be done at a branch with a real person. These banking actions tend to be rare but very important in life, actions such as sending wires, obtaining loans, etc. In the long run, this could mean a couple different things. One outcome is that online banking develops all of these capabilities, and brick and mortar branches will no longer need to exist period. A more likely outcome, I think, is that the number of physical branches decreases significantly, but certain, complex banking “features” will still only be available in person.

I think this second outcome is actually potentially bad for consumers – it would make it a lot more inconvenient for consumers to do these rare but important banking activities, and it would probably inhibit the adoption of online banking (because consumers “have to” have a brick and mortar account for certain important tasks, so why even sign up for a second bank account?).

There are certainly companies that are trying to digitize the entire banking experience, and now we see the rise of startups in the lending space that can give people loans without face-to-face interaction. However, these are generally separate from the online banks doing deposit-keeping functions. What will happen to this space in the future? Will every consumer have to interact with a “constellation” of consumer fintech startups, each specializing in a different banking activity? Will one online bank start acquiring these startups and form the first super-powerful completely-online bank? Or wiill a brick-and-mortar bank like Chase push to digitize all their actions and make their own physical branches obsolete?

On September 13, 2015, Allen Yang commented on Wealthfront: Winning at Wealth Management :

Totally agreed that Wealthfront and its competitors like Betterment are doing a great job competing against the “brick and mortar” incumbents. This is especially evident in the user-facing part of the wealth management experience. But, I wonder how much is different between Wealthfront and traditional players in the “back-end”, namely the algorithms and portfolio decisions that translate a user’s preferences into actual trades and portfolio allocations.

In some sense, if Wealthfront/Betterment replicate the same portfolio decisions as Charles Schwab, then perhaps what they’re doing is just a new marketing play: they’re using a digital platform to reach more users and “democratize” a product that used to be targeted more towards the wealthy. However, this makes me nervous that competition against Wealthfront and Betterment will become fierce and that “digital wealth management” will become somewhat like a commodity – the actual financial performance of Wealthfront and Betterment won’t be any different, and instead it’s all about who has the most marketing (i.e. VC) dollars.

I think this speaks to a powerful consequence of a digital revolution. Whereas previously geography and brick and mortar locations used to be a source of advantage, a digital product knows no geographic bounds. It might become harder for companies in certain industries to differentiate themselves against competitors, but this could mean better outcomes (e.g. lower wealth management fees) for consumers as certain products become commoditized.