dwang's Profile

So if I understand your question correctly Melissa, many of the sensors, cameras, and robotics being used are already in place. What DC Water is using AI for here is simply the processing of the data that those cameras already collected. I do agree that IoT will become a powerful tool in scaling up infrastructure support, but I imagine it will be a slow and steady process, and likely largely motivated by shrinking budgets (which will require government to replace more expensive human labor with tech).

It’s a good question, and I think that resource constraints are actually going to be the best driver of AI adoption. As municipal government budgets get stretched, many of them will have no choice but to adopt tech as a way to do their jobs well without materially spending more.

Among sports footwear brands, I think Nike has been a clear industry leader in adopting new uses case of AI-based technology. However, one irony I’ve always found with Nike (and apparel companies in general) has been the long lead times it takes to develop a new product. On one hand, it’s great that AI models can tell you in under a second what shoes are popular…but on the other hand, it takes Nike 18-24 months to actually get those shoes in the hands of consumers. I know that some hyper-fast fashion players (like Shein) are speeding up this process, but do you know if Nike is meaningfully using AI to reduce their manufacturing lead times somehow? That feels like an area where if they can use tech to cut production time by half, would be a true game changer.

Great job Serena! AI-based influencers are becoming a real thing now, and I believe they will exploit an underserved niche of users who crave content that is hyper-customized towards them in a way that no humans can match. I also think the clothes detection feature is absolutely huge…especially if the system can automatically suggest similar looks at different price points. I think a lot of the AI-based use cases you’ve mentioned are being implemented by all the major social media companies…do you think TikTok (or Bytedance) has any unique advantages in this space compared to the other players?

I’m reminded a lot of the old single-player Diablo games, which regenerated new dungeons maps, monsters, and layouts every time you restarted the game. Back then, this was extremely revolutionary, since it gave the games near infinite replay value without it getting stale, and surprisingly, it wasn’t particularly buggy (i.e., no monsters in walls, or unplayable areas emerged). I think one super interesting use of AI in games is personal adaptation of single player games. For example, in an exploration game, depending on how well you are doing, games can dynamically adjust the environment to suit your skill level, to ensure you are always adequately challenged. I know many games already do this to a limited degree, but AI can really push this to the frontier, by crafting monsters customized towards your particular weaknesses, or by crafting new worlds on the fly based on your style. As someone who really prefers the value of single player games, I think this could add a component of dynamism that is missing from the modern gaming landscape.

I love how thoughtful you (and Autodesk) have been in understanding that architects are artists, and that not everyone is open to boiling down every process to the most bare-bones-efficient state. I disagree with you that adding these genAI features means that Autodesk should change (especially increase) their pricing. As a tech company, I believe that thoughtful integration of genAI tools is now table stakes for many design software vendors, and that not doing so is actually more risky for disruption. Autodesk is already the industry leader, and at this point, they have to use their scale and resources to put more space between them and competitors, which means readily incorporating tech tools (that have almost no marginal costs) as part of their base offering. Ultimately, I believe their “success case” here is everyone is using their genAI tools (and not someone else’s), regardless of cost.

Hmm, if the AI is as accurate and independent as they claim, I wonder why they even need to be present at doctor’s offices…could users not just submit an image of their retina online (or through their webcam)? I imagine there might be some special lighting or other adjustments needed for the images, but as the models get better, I can see at-home AI health tools being truly revolutionary. Alternatively, as you mentioned that the marginal costs for each test is effectively $0, I think this could also be a great tool to have at schools, supermarkets, and churches (much like those blood pressure monitors), since healthcare can essentially be treated as a public good (and hopefully government backed).

I’m usually very skeptical of no-code solutions, as the couple that I have tried ended up taking about as much time as more traditional approaches once I factored in the time it took to customize/ fix specific widgets. However, I do like how nicely it plays with existing tools, and wonder if it actually can be an acquisition target for Figma. Like many others, I’m very concerned about the defensibility of products like this, since it essentially is a wrapper on someone else’s LLM (ie. too many layers of abstraction), and I imagine the future of web design will look something more akin to a full-featured solution (like Adobe) than these spot tools, which ironically end up being only useful for select power users.

Great job Aliza! Obviously, using LLMs for anything nowadays is super trendy, as well as any counterarguments on safety and bias. I love the fact that they’ve limited their user pool, since I imagine the downside of opening this app to the general public is just much too great (I can already see the stories that get posted if people are able to game the software to produce scary responses). I actually think there’s a very valuable use case here for diagnosis, as opposed to treatment. It makes a ton of sense that an AI conversation buddy can create a reasonable inference on what types of issues are troubling someone before handing them off to a trained therapist…the treatment portion makes me uneasy, since I genuinely do believe that it requires human empathy. While Woebot may be able to simulate it, if the illusion cracks at any point, the patient might end up in an even worse state.

Hi Sreeni! Can you talk a bit more about how Fiverr can guarantee product quality and transparency to their customer? Is it because these tasks are so unsophisticated that there’s not much room for error? Or is it that at a $5 price point, buyers just aren’t as choosy, and if a seller disappoints with their service, that seller’s reputation suffers and the buyer can just spend another $5 with another vendor?

I agree with your point that for some of these simpler tasks, even $5 might be too much considering the relative cost of GenAI to reproduce this. As the company has been growing, have they noticed any specific niches that they could really own (much like how ZBJ disintermediated the copyright space)? These kinds of platforms catering to micro-services generally make me nervous because I always imagine that if their buyers do well, eventually they will upgrade to more professional vendors, and so Fiverr always sees their “best customers” churn.

Thanks Sam! Do you believe that Weee!’s biggest competitive advantage has been their supply chain management? Since you mention that they mostly cater to audiences that typically don’t have access to local ethnic groceries, I imagine they must manage a complex system of delivery operations that would be hard to centralize across a wider (and likely less dense) geographic area. However, as with most networks, once they get to a certain scale, it actually allows them to attack the incumbents (in this case, more traditional grocery stores in cities), so perhaps they’ve simply outgrown this problem. Do you know if they are still leveraging small, third party delivery services? As the company has scaled significantly, I imagine that it would be hard for them to find vendors to keep up; but while it may make sense to operate an in-house delivery team, that also dramatically increases their expenditures and operations.

It’s an interesting company, and I remember that I almost used them when I used to live in NYC, but the negative reviews threw me off. The service that they provide is indeed valuable, especially in dense urban areas where transporting the furniture may actually be the most difficult part of the process. However, to your point, to truly leverage their network, they need to make the inventory available nationwide, and that just feels like such a heavy undertaking, especially for a smaller player in the space. I imagine quality assurance might also be a big problem with them- most of the used furniture I’ve bought have had chips, cracks, or defects in some way, and while many platforms allow you to inspect the piece in person before finalizing the deal, AptDeco essentially asks you to buy the item sight unseen, and once it’s delivered, it’s probably too late to price adjust. If done right, it can deliver fantastic value to all parties on the platform (just look at how Craigslist has persevered, even though they don’t really have any value add services), but in some ways, it just feels like a tough market to play in because of the unpredictable nature of the underlying goods.

Great job Wabantu! I’ve been an early adopter of Robinhood, and to this day, feel like it has maintained its status as the most well presented and easy-to-use trading platform on the market. However, one concern that I have had is that (similar to what others may be mentioning) Robinhood seems to incentivize trade activity, preferring actively managed investors, rather than passive ones. Obviously, this makes sense since trade (and trade adjacent) activities are what ultimately drives their revenue, but it does call into question the obvious conflict of interest, since it has been proven that actively managed money tends to perform worse than passively managed. While it is clear that these decisions are made by willing adults who are not forced to do this, I do worry that the simplification (perhaps gamification?) of trading actually induces unwanted behaviors. While we cannot prevent people from doing this, perhaps there is a public responsibility to not make it seem too easy or attractive (like adding bitterants to bleach, rather than making them orange flavored). Simply put, perhaps this is a situation where some friction may actually serve society well.

Thanks Abdoulaye! Can you talk a bit more what it is in particular that allowed Jumia to overcome some of the daunting challenges in Africa that its competitors couldn’t? In particular, my understanding has been that in environments where there are low levels of infrastructure, local players tend to win because of their ability to better understand the terrain and focus on being more precise in managing operations. In these cases, you have a very fragmented vendor ecosystem without any dominant players. Secondly, given these infrastructure woes, I wonder if Jumia has received any government assistance in helping create infrastructure that benefits both the commercial and community interests; for example, I’ve read in other cases where local governments have collaborated with large companies to build roads, dams, and ports, all to support the growth of a particular industry (or company) there.

Thanks for the post Serena! Little Red Book seems like a great product, and I love how it avoids many inherent conflicts of interest by staying away from a traditional advertising model. It appears that it built its initial popularity by catering to people in 2nd tier-and-below cities, who were not the typical audiences for social apps, and since then, has become the dominant player through a tremendous feat of user land grab.

My concern with all super-apps is that for more professional use cases (such as house buying or trip planning), whether they can continue to serve those use cases appropriately without diluting their value for all parties. For example, if I want to buy a house, perhaps LRB can be a good place to initially browse, but can the app be disciplined and sophisticated enough to cover all the nuances and details that go into the house-buying experience. We often learn that to truly create a differentiated product, you need to be obsessed with a niche, and while theoretically nothing is stopping any company from doing that, the reality is that when it comes to resource prioritization, it’s impossible to prioritize everything. Given LRB’s success thus far, I’d be interested in learning more about whether it was just due to their scale, or if they’ve developed techniques to empower their teams to deeply prioritize many projects simultaneously.

Toast has been a rocketship in the Boston tech scene, and I’m glad to seem them featured here! You did a great job breaking down all the functionalities that Toast supports for small restaurants, but noting how that comes with a price as businesses inherently face steeper switching costs as they utilize more of the Toast ecosystem.

Do you have a sense of what their NPS is? Anecdotally, I’ve talked to a few restaurants who use Toast, and their reactions were…lukewarm. It seems that Toast was a first mover in a space that deeply craved “good enough” technology, but I’d be curious if their product is genuinely meeting restaurants needs, or if it’s just clearing the low bar of paper/ pencil solutions. Finally, I didn’t know that they were making loans part of their business model, but it is very smart. At the very least, I wonder if they would be willing to share some of this data with third party lenders (like banks), to help scale up their lending volumes.

Thanks Carmen! I’ve looked into Turo several times for personal use, and each time, the fees were what ended up making me choose an alternative. The sample illustration you provided was very helpful, and I wouldn’t be surprised if Turo was indeed taking a 50% margin on each rental. In my case, Turo usually had a miles limit (which traditional car rentals didn’t have) that was the dealbreaker, since I only rent cars for long distances; once I factored in everything, it appeared that traditional car rentals were actually cheaper.

I would be curious in understanding whether Turo’s large fee percentage is justifiable in that it is required for operations, or if they feel like they can take such a margin because they can price match traditional car rentals. If the latter, I would expect there to be an increase in lower cost digital players entering the market, and driving down the fee structures towards something a bit more sustainable.

Great post! I think Bumble has done a great job in experimenting with ways to leverage their existing network (and I’m sure part of their intent is also to “cross-sell” their networks on the additional relationship types). Do you know if there’s a breakdown of their volumes of engagement across these 3 product types? In my mind, each of these relationship types (romance, professional, and personal) are fundamentally different, and I’m not sure if they are easily conflated into a single superapp. I think one of the tricky things here is that by scaling up the user base, Bumble is inherently diluting the quality of candidates across their app, and in a scenario where they don’t have a material way to vet/ ensure quality control, growth seems to go in direct opposition to quality. While there might be user growth in the short-medium term, eventually, the decreased quality catches up, the most fervent users leave, and the app suffers.

It’s very meta of Samsung to use AI to power…making better tools for AI, but then again, I guess that’s the virtuous cycle that technologists dream of 🙂

Going up against TSMC is a monumental task, and I think Samsung truly has their work cut out for them; my understanding is that in this industry, the technical knowledge is perhaps even more vital then the physical materials, and that if you gave Samsung TSMC’s factory, Samsung wouldn’t even know to operationalize it (and vice versa). My concern is that in a space that does indeed require such a wealth of technical knowledge, if gen AI can truly deliver meaningful insights on process and cost improvements. I believe that gen AI is exceptional at summarizing large volumes of human-curated, clear data, but in this case, we are talking about super niche and opaque processes that humans would struggle articulating. Hence, I worry that getting the right training data into the AI models would be a deal breaker. However, I am all for cheaper processors, so I wish Samsung the best of luck!

Wow Sam, great minds think alike!

I think one thing I’ve always found interesting about Tesla is their reluctance to use LiDAR devices to augment their sensor offerings. Historically, they’ve made the argument that it is unnecessary on technical grounds (i.e. their cameras and other sensors work just as well), but I’ve always believed that it was because of cost. But as the prices of LiDAR decrease steeply (not unlike microprocessors in the past decades), I’m curious if Tesla will reverse course and adopt it. LiDAR is still regarded as a superior technology to traditional cameras, and as regulatory scrutiny increases, it would almost be irresponsible of Tesla not to at least offer it to quell public concerns.

Splunk is a great business, and their recent successful acquisition notwithstanding, they’ve really managed to differentiate themselves in the crowded “data dashboard-as-a-service” space that has been aggressively targeting internal company operations over the last few decades. I think a constant tension facing Splunk (and perhaps the space in general) is whether or not they want their end users to be active, or passive, users of their product.

On one hand, data analytics is great, and Splunk would love nothing more than for people to use Splunk to build dashboards, analyze the data, gain insights, translate them into actions, and repeat the process. However, their customers only have limited bandwidth, and very often, data is ingested…but no one actually looks at them (despite paying for the service).

On the other hand, Splunk can make the argument that it is able to passively monitor the data for anomalies (thus saving companies labor hours), but over time, as the data gets stale and customers are no longer creating new dashboards, Splunk’s value prop also decreases.

I’m hopeful that Splunk (or now Cisco) can use their internal data analytics to figure out a way to walk this tightrope effectively, but in any case, great writeup on a company that truly uses big data at the very core of the company.

This is awesome! Healthcare is simultaneously a big data player’s wildest dream and worst nightmare. The pure volume of data available, alongside the mission criticality of its use cases means there is so much opportunity to build rewarding (both intrinsically and financially) products, which is something that CVS appears to have accomplished here. However, the fragmentation of the space in general means data collection and usage is extremely difficult, and the albatross of government regulations hang over every step of the value chain.

I know many digital health startups have tried making a business out of helping people design custom digital health plans, but customer acquisition was always the hard part, especially since many of the end users were either unused to these technologies, or could not bear the monetary or labor burden of adding another service to their mix. CVS is in a great position to offer this as a value-add service. However, given recent market conditions that has put greater pricing pressures on traditional pharmacies (ex. California recently announced it would not longer use CVS), I do worry that these data-fueled services will get even more fragmented, and undermine their existing (and somewhat tenuous) viability.

Poor stock performance aside, I think one core issue with StitchFix has been their desire to use data to create what appears to be a human-to-human relationship, but ultimately feels empty once issues arise. Having a stylist is both a practical and an emotional relationship, and professional stylists often share an intimate connection with their customers in terms of understanding and choosing their “look”. StitchFix is trying to use data to replicate this at scale, but once something goes wrong, there’s no human there to physically apologize or learn, and so the customer becomes very (quickly) aware that at the end of the day, they are talking to an algorithm. The irony is that digital and human stylists may make the same mistakes over and over again, but the emotional ties of working with a person means customers may be more forgiving of the latter. I sense a parallel to ChatGPT here: when it’s working great, everything is awesome; but when it falls short, it can feel acutely dehumanizing.

The obvious elephant in the streaming room is how Max can hope to compete with Netflix, who has been pushing the envelope in many areas of data analytics and recommendations for decades now. Obviously, Max wants to believe that their differentiated content offerings will encourage customers to choose it instead of (or perhaps in addition to) Netflix, but to your point, the recent industry-wide focus of profitability means that more content many not be most viable path forward. But then again, maybe Max’s goal is not to be the #1 streamer, but simply obtain enough market share to sustain itself as a viable business. In any case, great work on breaking down Max’s data analytics usage! I’m curious to see if Max can find ways to use data to truly separate itself from the ruthless competitive landscape, since I’ve always been a huge fan of the prestige programming on classic HBO.

I can see how the Playstation team uses data analytics to increase the operational efficiencies of their manufacturing processes, but one thing I’ve always been curious about is how they can better use the data from their gamers to create better games (or more curated games). Since most video game platforms now host their games digitally, and track players’ playing time or engagement (ex. Steam or the PS Store), I’ve always been keen to see if that data helps them discover player niches, which they then translate into games. Netflix has had great success in this area, by discovering undiscovered segments (i.e., people like to watch small-budget Korean horror movies), and then creating content to fill that void, in a way that large production studios typically cannot afford to, or unaware of. A rise in indie game that fills these targeted audiences would be great for the market, and may lower the barriers to entry for many smaller game studios.

I always have deep concerns when two entrenched incumbents (in this case Google and Ford) come together and believe that “if we put our two existing technologies together, everything will go great!”, specifically in the case of in-car optimizations. I think one of the reasons why Tesla has been so successful with its vehicles is because it has aggressively vertically integrated its hardware and software stack, so that the end product is genuinely reflective of a single design vision. In Ford’s case, because Ford already outsources many of its subcomponents, its vehicles are essentially a collection of technical pieces manufactured by multiple OEMs. Coordinating those pieces took decades, and adding Google technology in a way that requires it to interact seamless with all of the existing pieces is a tall task, and not one that can be easily undertaken by a company with limited history in software integration (Ford). While I hope this venture succeeds (since competition benefits the end consumer), I am skeptical if Ford can truly make their vehicles play nice with Google, or if this will end up feeling like tacked-on software that doesn’t deliver on meaningful customer value.

It’s really awesome that Carnival is being so forward thinking with adopting this new technology. A cruise ship is the perfect place to run experiments on new offerings, and I can see it being a valuable experimentation engine for the company going forward. While I understand the privacy concerns, I would also say that a cruise ship should not be the place that people go to if they are looking for privacy 🙂 Besides, they could always just opt out of wearing the tracker.

This idea of using the Medallion to reduce labor hours is very interesting; while I’m not sure how they were able to reduce pre-departure time so drastically, I can imagine a world where this technology can offer helpful directions that can reduce the need for human guidance even further, and at ever step of the cruise process. It also makes it easier to highlight bad actors, or unwanted behaviors, on the ship, and I can see it having value in helping with contact tracing, or otherwise being able to detect anomalous behavior.

Wow, I didn’t realize the sophistication behind AccuWeather’s forecasts! I always thought they were “yet another measurement platform” , and I’m impressed that they were able to negotiate so many contracts with non-traditional weather groups to get a final estimate. I can imagine that they are leveraging the “wisdom of the crowd” to get a more accurate reading, but it would also be cool to see just how much more accurate they are than other sources, and that when they are wrong, is it only because they are seeing historically unprecedented events (ex. global warming related surges), or is it that there are consistent blind spots when it comes to humans measuring weather. I also wonder if this system of gathering such a varied dataset and collating it into a single graph has other uses as well, such as tracking the spread of disease vectors.

Thanks for writing this Ankit! Fun fact, one of the founders of Ramp was my manager at the consulting firm I used to work at.

Ramp’s product vision has always been one step ahead of its competitors, and I’m glad you are highlighting how powerful it is to have peer-based benchmarks for these SMBs that may typically lack insight into what their best practices should be. I imagine there must be a world down the road where Ramp can even provide loans based on a customer’s spending characteristics, and how close to “ideal” it might be. I hadn’t considered before how powerful ChatGPT could be when layered upon such an expansive dataset, but it really does yield some very compelling use cases. One concern would be since B2B contracts tend to have a variety of conditions (early exit, termination/ integration fees, etc.), just how accurate are Ramp’s benchmarked prices, and if they may be misrepresenting the nuances of some of these vendor prices.

Great minds think alike haha! I actually wrote about something similar in my Tesla post, given that Tesla has now started offering insurance in select states off of what is essentially built-in telematics data.

I like your analysis of the self-selective behavioral issues that arise from opt-in telematics, and I have to say that I’ll be surprised if telematics are not made mandatory at some point. After all, current insurance premiums are priced based on age, gender, and other demographics data- it’s imperfect and contains a fair amount of bias. What better system to gauge driver risk profile than to literally observer the driver actually driving? It’s true that the premiums for bad drivers will go up…but in an accurately priced insurance system, they should be paying more. The fact that they are currently being subsidized by better drivers in the present scheme is unsustainable, and I hope that this new trend can actually cause drivers to drive more carefully because they have no other choice if they do not want to be labeled a more dangerous driver.