To stay in touch, you can find us here: Gabrielle's Twitter and Shan’s Twitter
Hi friends, we back! The last two weeks have been quite a whirlwind as we were trying to navigate through the world on fire, which seems to only continue with yesterday’s news around UBS/ Credit Suisse acquisition and SVB bankruptcy reorganization. Despite covering the Financials Sector for many years since my public market days, neither Gab nor I would claim ourselves as true banking experts. So at some point, we should come back to this series of events and further unravel them to the best of our understanding to our audience. But before we go back to the routine public / private market pieces, I’d like to use this week to discuss one subject matter that has been on my mind for some time - this piece is almost a bit more philosophical than pure investment sense, but, as always, would love to hear y’all’s thoughts.
Proliferation of Recommendation Engines: A bit of History
Recommendation Engines, are defined (source) as advanced data filtering systems that use behavior data, computer learning, and statistical modeling to predict the content, product, or services customers will like. The first recommender system was created by Elaine Rich, a female American computer scientist, in 1979. The system was designed to figure out a way to recommend a user a book she might like. The idea was to create a system that asks the user specific questions and assigns stereotypes depending on the user's answers. In the modern business world, the Marketing industry is likely the vanguard in adopting recommender systems. Brands, CPG companies, and Marketers would intentionally put users into buckets of stereotypes even before any products are released to the general market and then design recommendation marketing campaigns accordingly. Media and Entertainment soon took this to the next level with the amazing rise of Netflix, TikTok, DouYin, and many more. Nowadays, it’s tough to find an area where recommendation engines are not at work - from B2C to B2B alike. So what does the proliferation imply for our society and the human race? Our current thoughts lie below:
Recommendation as Catalysts - B2C: Commercialize Your Thoughts
Many people’s first user experience with Bytedance started with 今日头条 (Toutiao). Toutiao was founded in 2012 as a news and media content platform that pushes tailored feed-list for individual users. It soon gained a lot of popularity and received $100M Series C funding with Sequoia Capita in 2014, almost 10 yrs ago. Using NLP (natural language processing) and computer vision, Toutiao extracts entities and keywords as features from each piece of content. As users interact with the app, Toutiao’s algorithm captures detailed nuances of the platform-user interactions through metrics like time spent per content, likes, comments, open rate, etc, to fine-tune the algorithm and models further to suit users’ preferences.
On one side, this is an impressive application for the many years of technological advancement in the NLP & Machine Learning academia space. On the other side, it inevitably started to induce more “addictive” consumer behavior. People oftentimes find themselves spending hours on the screen with news, short-form videos, and live-streaming sessions (mainly in China today). While Bytedance has certainly done a great job perfecting its recommendation algo and pushing it towards multiple presentation formats, Bytedance is NOT the only one catalyzing the proliferation of recommendation engines. Almost all consumer-facing companies utilize this tech to some extent - from your “recommended movies list” on Netflix, to “guess you will like” shopping cart page in various brands' storefronts, to “rent a car with your flight” on travel agency platforms - recommendation engines are now everywhere in consumers’ daily life.
The powerful recommendation engines have created more enjoyable and curated consumer experiences. Users are able to find products that fit better with their preferences, services that meet their needs, and experiences that cater best to their desire. On the merchants’ side - with an increased willingness to purchase and likely a larger basket size - LTV (Lifetime Value) of these consumers is hugely lifted while the CAC (cost of acquisition) remains relatively stable. Businesses can have a better ROI using the according marketing and content production dollar to specifically target the user profiles that fit best with the companies’ mission and style. Businesses are therefore turned more profitable and sustainable.
Recommendation as Catalysts - B2B: Automate Your Workflows
Today, most recommendation engines are powering consumer-facing products or content-based systems. Albeit earlier in the adoption curve, recommendation engines within the software & B2B workflow space could also see huge potential. Vertical-focused software and marketplaces could be the best experiment grounds for the following reasons:
Record of Truth: many vertical-focused software and marketplaces start off with one edge feature that generates sales upside or savings for the company. But eventually, they expand into a platform format that holds records of transactions, relationships, and operations for many facets of the business. With records, recommendation engines can systematically study patterns and categorize interactions of the business with customer accounts, other businesses, and internal department functions.
Chokepoint for Transaction: only historical records are not enough, and the recommendation engines are only powerful as the next iteration of user feedback. Sitting at the center of transactions, these vertical-focused softwares and marketplaces can receive feedback on interactions in the formats of orders, booking frequency, and times spent, just like how we click on the Movies in the “You may also like” list on the Netflix screen. Consistent and forward-tracking interaction will allow recommendation engines to fit better and cater to the transition and growth of businesses.
Upstream to Downstream - Full Workstream: in their mature format, these systems have the potential to run full stack from procurement, inventory management, order management, CRM, and financial payments. While we don’t believe recommendation engines will be powering the business’ operations end-to-end, acknowledging the information flow and transaction record changes would enable more curated recommendations.
For example, the operating system for a spa service could pop up a message that goes like “based on the last purchase time and # of serviced booked in the past few weeks, would you like to place an order for the next batch of shampoo inventories?”
Recommendation as Inhibitor: Extremism, Bifurcation, and Rashomon Effect
The above image is a few years old - but it lived in my brain rent-free for many years. One critical negative factor of a perfectly built recommendation engine, aside from the addictive consumer behavior mentioned above, is a very narrowed worldview.
Do the experiment as below: within your own friend group, ask to screenshot the Twitter feed of a very “anti-crypto” friend and the Twitter feed of a very “pro-crypto” friend - I assure you that you will see a very different worldview on how “crypto” is presented. I have done so, and the result is quite alarming.
The SVB fallout actually accelerated this train of thought of mine quite a lot over the past few days - I have seen people screaming “bank runs are the exact reasons why Defi is the ultimate answer,” and people who deeply believe all money should be moved into the most deep-pocketed 4 banks in the U.S. Recommendation engines may only further accelerate bifurcation of thoughts and minds - we as human beings are already born with “confirmation bias” and love to see our opinion & thoughts seconded by many others. Recommendation engines could function as a machine that leads to people seeing videos, news, and tweets that only lead them to believe what they think is right.
Rashomon, a famous movie created by Akira Kurosawa in 1950, basically depicted a story where one event can be presented from many different angles and interpretations that, at the end of the day, nobody knows what’s the “true story.” We could risk having very bifurcated worldviews and hence “history” of what really happened and will happen in our world if extremism starts to grow with more content feeding these growing narrowed views. So are recommendation engines the devil or angel? I have no ultimate answer - I deeply believe that technological advancement always comes in a neutral form, and then it's the people who utilize it to bestow meaning to it.
Chart of the Week in the Public Market:
Regional bank crisis intensified fear of a broader financial crisis. All major indices sold off over the past two weeks as investors turned to “risk-off” mode. The market is now anticipating a recession + Fed pivot in June. If you have been following us for a while, you will know that recession has always been our base case. We think over the next few weeks the macro narrative will continue to change between “soft landing”, “ mild recession with unemployment rate touches 4.5%” and “hard landing” and biased toward downside in S&P. As the consensus shift toward recession, we expect tech to outperform while value sector like financials lag.
Fintech index slid from 8.4x to 7.7x - B2B Payment as a sector took a huge hit with the -20% drop with Bill.com. See our read from the past earnings report here. Consumer Fintech held up the best with almost +50% YTD growth at 4.3x. InsurTech, which was most depressed during the decline, also inflected up. Most other sectors remain at roughly flat YTD.
Index remains roughly stable from two weeks ago’s 2.8x to 2.7x - Gaming continues to lead with 4.6x EV/NTM Rev premium whereas Subscription category also came up.
(Market data as of 3/17/2023, source: Bloomberg, CapIQ. See index composition at the bottom)
Chart of the Week in Private Market
Quite a few major news in the private market over the past 2 weeks - Stripe led the pack with a big $6.5B raise (news), which lifted the whole index to be almost the same as in the beginning of year high (post-winter lows). However, we also tried to take out of the Stripe impact - as shown in the graph below, Stripe aside the whole industry is still at a pretty contained pace.
(Deal data as of 3/19/2022, source: Pitchbook. Defined as - Series B+ global growth stage deals)
Sources: Software Index: over 200+ public companies / Fintech Index: V, MA, PYPL, SQ, BILL, ADYEN, SHOP, LSPD / Consumer Index: ABNB, BMBL, CHWY, CVNA, DASH, DHER, DKNG, DUOL, ETSY, FB, FTCH, GDRX, GOOGL, MTCH, NFLX, OPEN, PINS, POSH, PTON, ROKU, SFIX, SNAP, SPOT, UBER, W. Please feel free to ping us for further detailed breakdown