Mastering AI Disruption: Identifying Your Industry's "Wave" for Strategic Advantage
This is the Three Data Point Thursday, making your business smarter with data & AI.
Let’s dive in!
P.S. Also, I write the Finish Slime, where I share 6 Data Engineering and Analytics articles I discovered over the week.
Actionable Insights
If you only have a few minutes, here’s what’s going to make your business smarter by knowing what wave (1, 2, or 3) has hit your industry:
Identify Your Industry's Wave. Understanding whether your industry is in Wave 1 (single-step automation), Wave 2 (multi-step integration), or Wave 3 (creation of the new) of AI disruption is crucial. This knowledge will guide your strategic decisions, from product development to market entry strategies.
Embrace Statistical Problem-Solving in Wave 1. For startups and companies in industries undergoing Wave 1 disruption, focus on automating and optimizing a single step of the core industry process using data and AI. Look for tasks where generating educated guesses at scale represents a significant improvement over current methods.
Leverage Existing Resources and Data for Innovation. If you're in a position to influence product direction in a Wave 1 industry, use your unique resources (i.e., your incumbent resources) to solve problems in ways that new entrants cannot easily replicate.
Navigate Wave 2 by Integrating Steps. In industries entering Wave 2, consider how you can integrate multiple steps of the core process for a multiplier effect. Incumbents should be particularly vigilant about leveraging any integration opportunities to stay ahead.
Pursue Creative Solutions in Wave 3. In the Wave 3 phase, focus on generating entirely new products, services, or methods that do not currently exist. This wave allows for the most creative freedom but requires a deep understanding of the emerging needs and gaps in your industry.
Avoid Entering Wave 2 Markets. For startups, Wave 2 industries may present high barriers to entry and low differentiation opportunities. Focus instead on Wave 1 or Wave 3 opportunities where you can create clear value you can also capture.
"It is an old maxim of mine that when you have excluded the impossible, whatever remains, however improbable, must be the truth." - Sherlock Holmes in The Adventure of Beryl Coronet)
In the Sherlock Holmes Adventure of the Dancing Men, Holmes is presented with a strange sequence of stick figures. Those stick figures seem to cause serious distress to his clients’ wives.
Industries get disrupted by AI one by one, data creeps into bio, into agriculture, into everything we know. It rattles everyone but the disrupters themselves, causing turmoil all around us.
"With artificial intelligence, we are summoning the demon." - Elon Musk.
Sherlock Holmes would’ve made for an excellent incumbent CEO. He would not have gotten rattled by such disruption, such chaos. He’s able to find the order in chaos. In The Adventures of the Dancing Men, for instance, Sherlock Holmes is able to look behind the curtain to deduce that these disturbing stick figures are a hidden code - and thus is able to solve the mystery and save his client.
-From fandom.com
And just like Sherlock Holmes, you can find order in the chaos of AI disruption because there is.
AI disruption works in patterns, and one of them I’d like to present to you today is based on a great article by Omri Drory from NfX’s: 3 Waves of AI Disruption in Bio.
You might think that the search engine market and other 100% bit-based sectors are at the forefront of AI disruption, but they are, in fact, not. Bio is, even though it is very much an atom-based industry, and that’s why it poses such a great example.
Why should you care? Because you need to know what kind of waters you're entering to devise a good data strategy. If you’re a product manager and are looking at a wave 1 industry, it’s a vastly different environment than a wave 3 industry. If you’re a startup entering a wave 2 industry, I won’t be on your side to hold your hand when you drown.
Knowing your industry’s state will be the difference between catching a wave and drowning, regardless of your data strategy.
All industries have a core sequential process
We humans are sequential beings. We have physical limitations (a computer does not have), only two hands, a brain that focuses only on one thing intensely at the same time, and inherently linear learning systems. We love to tell stories, Yuval Harari makes our story telling ability out to be the one key thing that made homo sapiens dominate the earth.
Yet stories are linear in nature. We’re linear, sequential. And so, all industries, no matter their nature, evolve one key sequential process that represents the core of it.
The search engine market: Customers demand answers to their questions => They put that question into a sequence of texts => They type those into search engines => they get the answers they want (supply)
All markets have some linear sequence at their core, linking demand to supply via the participants of the market.
So, keeping this idea of a key sequence in mind, let’s analyze the three waves a little bit closer.
Wave 1: Multiple start-ups pick different steps of the sequence and automate them via ML/AI.
In Pharma, some steps are easy, and some are hard. Some are expensive, and some are cheap. If you want to develop a new drug to aid an illness, you need to figure out the core ingredients, the compounds, you then need to make sure they work on humans by running preclinical and clinical trials. Only then will you get FDA approval to sell your drug.
As you might imagine, experimenting on humans is by far the most expensive thing you can do! Clinical trials and preclinical trials are super expensive and time-consuming. That’s a killer combination. So to make it in the pharma industry, you’re core objective is to spend as little time as possible in these phases, by making sure you got good compounds to start with.
Previously, this involved tons of physical experimentation inside labs. However, the pioneers realized that compound screening is a great place to start applying data & AI. By now, there are tons of applications of data & AI inside compound screening, all based on two ideas:
Virtual screening (instead of the atom-bound physical experimental process)
Optimization of compounds (again, instead of the manual process)
The key lever in wave 1: Wave 1 disruptions are simple - they focus on one single of the many sequential steps inside the sequence and optimize the hell out of them using data & AI.
An example from today is coding. GitHub Copilot takes part of what it takes to write a complete piece of software and starts to automate that.
Notice, while the Copilot suggestions are great, they are not perfect. And they don’t need to; they just need to be good enough. For the average coder, they are much better than the alternative of searching and guessing the right solution.
GitHub isn’t trying to write a complete application automatically, test it and deploy it. Rather, it focuses on one step they think they can solve.
Interestingly, they weren’t the first ones to try this, but rather the first ones to nail it.
Navigating Wave 1 Disruptions
There’s a reason wave one disrupters in bio picked compound screening. They could’ve also picked the much more expensive preclinical and clinical trial phase, but they decided against that. Compound screening was a perfect candidate because it was about generating ideas and taking educated guesses at potential compounds, not about getting them right.
People still have to look at them, but because the process is statistical (people were trying out stuff before!), using data & AI is a huge step forward. So if you want to be a wave 1 disrupter, I recommend you pick a statistical problem, one that has no clear outcomes, no perfect solution, one where generating educated guesses at scale is a 10x improvement.
If you look carefully at the GitHub CoPilot, that’s exactly what it is: an educated guessing machine for writing code. But GitHub has something else going for it with the CoPilot. Other companies have tried to solve the same problem before but have failed. IMHO what made GitHub succeed is their ability to choose a problem, they had superior resources to solve. They used their immense database of code and integration into their existing product to make sure they could truly provide a solution to this problem, unlike the companies that came before them.
For bio it was very much the same, the company behind Atomwise the first convolutional network combined their expert machine learning knowledge with a database that was twice the size of any big pharma company to create their breakthrough drug compound screening solution.
Finally, for wave 1 disrupters, I suggest you pick something humans suck at, like driving. In fact, one of the automatic driving disruptions I’ve been most impressed with is automatic lane control for agricultural machines. 5 years ago, the very expert drivers, the 1% of the drivers were able to hold lanes such that the wasted effort for the machine was minimal, now every single John Deere machine can do it, better. It’s only one simple step in the whole sequence, but it’s one of the things that made John Deere a market leader and a true data company.
I know many software engineers will hate this idea, but I remember one of the best books on software writing (The Pragmatic Programmer) explaining why humans are fundamentally not made to code: Because code needs to be perfect, and humans are made to get things to work - in nature that never involves “perfect” but “good enough.”
Wave 2: Businesses realize that if one step is possible, why not go further and integrate multiple steps? Use automation in between to get 5x of the results.
In his article, Omri describes how, in 2016, a shift happened in the bio-industry, pushing it into a second wave. The technology went from simply 10x the experimentation capabilities of compound screening to actually being able to tell what compounds could work to reverse a certain disease and which don’t.
Previously that’s been step 2 in the sequence of drug discovery, but in a wave 2 disruption, companies decide to go a step further and integrate multiple steps together, use automation between those steps to get a multiplier effect working.
Great examples of this in action today are provided by Google. If you think through the process of searching you usually imagine typing in a question, and then clicking on links. The wave 1 disruption in searching was really the search engine itself as a place to provide links.
The wave 2 disruption came with Google Quick Answers, provided by all search engines today, they still let you ask a question, but they sometimes provide an answer right away.
There’s also another step integrated with Google Instant Search, predicting what you want to ask, thereby integrating into more parts of the sequence (now we just need a brain link to complete it.)
Navigating Wave 2 Disruptions
The waves of disruption can be turbulent, they will capside your boat wether you’re an incumbent or already inside the industry.
Google already was in a wave 1 disrupted industry, in fact, they themselves were one of the wave 1 disrupters inside the knowledge search space. Wave 2 disruptions are peculiar in that way - as you probably know, all search engines today feature both Instant Answers and Instant Search. You get them across Bing, yahoo, and all the other marginalized search engines.
That interestingly makes these wave 2 disruptions after a wave 1 disruption only valuable to the user! Not the company. That doesn’t mean if you’re confronted with a potential wave 2 product you shouldn’t build it, quite the opposite, inside “How Google Works” Eric Schmidt tells the story of how the Instant Search was discussed internally. From a profit perspective, it seemed to not make sense. And yet Google very well knew they needed to be the first mover here. So they built it, even with a potential negative impact on profitability. If you’re an incumbent company after wave 1, you need to jump at wave two opportunities.
On the other hand, there is not a single search engine that launched into the wave 2 disruption. There’s a reason for that, the dynamics of wave 2 don’t make sense for new companies, who would launch a company to not gain any value? For startups, the best advice is to stay away from wave 1.5 industries - the ones close to being disrupted by a wave 2.
However, if the industry is bare-bones, it’s a different story. There is no reason you need to go for wave one first. A lot of industries have what is called “chain link” properties - meaning you only provide a meaningful improvement of value to customers if you provide a 10x in more than one “chain.” In those industries, make sure you pick a combination of steps that link well together. Where you can not only provide an individual 10x per step but rather let the value flow through both steps and create a multiplier effect.
I recommend using a value chain analysis to do so. There are a few good methods in my article on deriving good data strategies.
If you feel like the combination of steps is too deep, you still can pivot into a sub-niche, a meaningful industry segment of good size that is more susceptible to chain links.
Wave 3: Get creative and provide something new out of that sequence!
Omri calls the current wave of disruption inside bio wave 3. It’s the wave where we go away from working with the existing steps and start to generate something entirely new! That’s an exciting place to be because if you generate new things, you’re not bound by current demand and supply, you can tap into huge new subsegments of the market.
-From Omri Drory, The 3 Waves of AI in Bio
It also means your solutions are paradoxically not held to the tight quality standard of wave 1 or 2 disruptions, because they simply are completely green field.
“The combination of these factors has paved the way for our new ability to generate new things. We have models that “understand” the core of protein-ness, from the biological language that makes them up to the structures that facilitate function. And we are developing the tools to design them from scratch.” - Omri Drory
A great example of a small isolated current wave 3 disruption I personally enjoy is Spotify’s new personalized niche mixes - the Made for You Hub. Previously, Spotify relied on humans to label and organize playlists that were at first generated by humans (you and me), later also to label machine-generated lists, but now, Spotify automatically generates everything for me:
A playlist I might like (and they get it right a lot!)
Imagery that works for me (I got a Van Life playlist that features a nature scene)
A description that fits the feeling of the music
And it works, I listen to new music and broaden my experience, because it is something entirely new - a wave 3 disruption. This is a rare instance where the label generative AI fits the purpose. Wave 3 is truly about generating something new using AI.
-Spotify - Mixes for the Moment
Navigating Wave 3 Disruptions
Wave 3 tools rapidly become the new normal inside an industry. These disruptions tend to be a lot like wave 2 disruptions in the sense that everyone will use the same tools quickly, but because the applications are green field, they do provide a sustainable competitive edge and create extensive value for the companies.
So whether you’re an incumbent or a newcomer, in a wave 3, you’ll need to pick up the new generative tools quickly. If there’s a ChatGPT-like fundamental tool, an AlphaFold API, or whatever it is in your industry, get your hands dirty.
Go for something that is big! That usually involves finding the right set of users that have a big problem! I’m a perfect example because I’m musically illiterate, that’s why Spotify mixes ar perfect for me. For people with a more advanced music taste, I assume that finding good music isn’t a “big problem,” but it is for me.
To create new things, you’ll need to pick something no one else is doing. Go for subniches, subproblems, submarket segments. Don’t just chase the latest fad. There’s a reason GitHub Copilot is one of the most successful applications of generative AI, and most of the other 100,000 ChatGPT-powered coding tools suck.
Be careful what wave you’re starting in
“Being in the middle rarely works. The second generation of AI for drug discovery found it hard to break out. There was too much noise.” - Omri Drory, Ph.D. The 3 Waves of AI in Bio
Google is a great wave 2 example because they were in wave 1 as well. But you might also notice no other search engine started out in wave 2. And almost all search engines now feature the same advances that Google does. The advances of wave 2 were mostly beneficial to the user, not to any individual company.
There’s no space for you in wave two if you’re starting out, make sure to stay away from it, try to go for the wave one or wave 3 stage industries.
For incumbents, you’ll need to look to surf all the waves out there, but you can be pretty relaxed in wave 2.
Applied Waves
The business intelligence industry is estimated to be around $30b and growing. And despite every single vendor claiming they have entered the age of generative AI, it is pretty clear the industry is still searching for its first big disruption. To me the state of the BI industry is clear: A wave 1 is coming, but no one knows from which direction.
Companies like mode focus on writing SQL queries with AI help, Tableau tries to democratize data with AI by getting personalized insights into the user flow. Google Analytics has been using AI to detect anomalies and more for some time. But I think it’s safe to say, no wave has yet hit us that carried us all away.
So what can you do? If you’re a startup founder, figure out good statistical problems inside the space and nail them. If you’re an incumbent company PM, use your resources! Use the data you have to automate something that no startup could automate.
If you’re looking for a wave 2 industry, I’d be looking at the email application industry, not to be confused with the marketing (HubSpot) or encryption email market. Even though all the popular clients are free, the market has a size of $3.8b. Interestingly, the growth is mostly driven by wave 2 disruptions, focusing on usability, which is also present in free solutions like Gmail.
Algorithmic spam filtering has been around for some time, but only recently did email application developers start automatically labeling your mail as junk, marketing, or useful. Only recently did they start suggesting potential replies and correcting your grammar. It’s the combination of steps that makes these tools so powerful; they integrate to make working with email fast and to clean out your inbox fast.
If you want to enter this industry, I’d say don’t. It will be fast-moving and won’t generate any value for the companies owning these solutions for quite some time.
If you’re looking for a wave 3 industry, it’s bio! And it’s a great place to be.
Here are some special goodies for my readers
👉 The Data-Heavy Product Idea Checklist - Got a new product idea for a data-heavy product? Then, use this 26-point checklist to see whether it’s good!
The Practical Alternative Data Brief - The exec. Summary on (our perspective on) alternative data - filled with exercises to play around with.
The Alt Data Inspiration List - Dozens of ideas on alternative data sources to inspire you!