Discover more from Tanay’s Newsletter
Startups vs Incumbents in AI
The benefits and downsides startups face and what they should do vs incumbents in using AI
This is a weekly newsletter about the business of the technology industry. To receive Tanay’s Newsletter in your inbox, subscribe here for free:
The use of foundation models and AI more generally is reshaping the business landscape. As AI technology makes its way into every application, there has been a lot of discussion about who will capture its value between startups and incumbent applications. Will all of it accrue to incumbents or will startups will have a chance to succeed?
I’ve touched on this topic before a bit, but in this piece, I will dig into the benefits that incumbents and startups have in the AI race, and discuss what startups can do to compete with incumbents.
I’ll touch on:
Benefits Incumbents have
Benefits Startups have
What Startups should do
Incumbents have existing distribution and already reach millions of customers. This advantage allows them to integrate LLMs into their products and immediately reach their entire user base, rendering the need for that user base to seek other solutions less important.
We’ve already seen companies such as Microsoft, Adobe, Google, Salesforce, Notion, and Intercom quickly announce or launch Generative AI features into their core products – design tools, CRMs, notetaking, email, productivity, spreadsheets, support tools, and more.
Will most of those hundreds of millions of users who get these tools bother seeking out a new product just on the basis of AI? Likely not.
We’ve seen the distribution advantage for Incumbents play out a number of times before in software, with Microsoft Teams vs Slack, just one example
2. Proprietary Data
ChatGPT and other models by themselves are good, but they become great when paired with relevant data from the user/organization.
Incumbents have access to large amounts of this proprietary data which they can provide as context to the models or use to fine-tune them. This data is typically collected from their existing customer base and gives them a head start over startups.
A great example is Microsoft, which in their copilot announcement discussed their use of “customer’s calendar, emails, chats, documents, meetings and contacts” in their product.
Incumbents typically have more access to capital than startups, even despite the large appetite from venture funds to fund startups in this space.
Now, in some ways, capital is needed less at the application layer since the LLMs are more expensive and can be used as an API relatively cheaply.
But capital is always somewhat of an advantage, and incumbents can use it to forge proprietary partnerships, hire talent, or subsidize costs.
Microsoft again is a great example, with its partnership and investment in OpenAI that gave them access to GPT-4 in Bing slightly before launch and their ability to eat some of the inference costs for a product like Bing or Office copilot when it launches.
Incumbents have an advantage in already having talented engineers, designers, and AI researchers on their team and in potentially attracting more talent. This only goes so far however since ultimately this talent has to be given the freedom to actually ship.
Google is a good example – while researchers who worked there in some ways started this whole movement through their 2017 paper, it was OpenAI (not an incumbent at least at the time and relative to Google), that brought it to market.
In addition, in some ways, if you’re building an application, the need for AI-specific talent goes down a lot given the use of LLMs via APIs – more of the work might be in prompt engineering, building workflows, design, etc. While incumbents have these on hand, they aren’t a cornered resource per se, compared to say AI researchers which are more concentrated at large companies.
If you don’t yet receive Tanay's newsletter in your email inbox, please join the 5,000+ subscribers who do:
Many new startups recognize that LLMs are a compelling “Why now” and so can focus on leveraging them well and getting the full value out of them for the use case they are tackling.
Many incumbents might create a new AI team and expect them to tack on the GenAI features to the core product, but it may or may not be a priority. We’ve seen that some incumbents have definitely made it one, but in other areas, the incumbents may not, giving startups an opportunity merely by focusing.
Speed is another advantage that goes hand in hand with focus. Startups can move quickly, experiment, and pivot if needed, without the bureaucracy and red tape that may exist in larger companies.
At large companies, an idea may need to go through multiple chains and either be killed or at least be pushed out for multiple months. At a startup, it can go from being developed to being launched in a day.
"Let's say a junior executive comes up with a new idea that they want to try. They have to convince their boss, their boss's boss, their boss's boss's boss and so on—any 'no' in that chain can kill the whole idea." – Jeff Bezos
3. No existing business to defend
Startups have no existing business lines or products to defend, which can be an advantage when it comes to disruption. They can focus solely on creating new AI products and services, without worrying about having an existing business to cannibalize.
Incumbents may face the “burden” of how things are currently done / their core business, and have boundaries within which they need to operate which can end up constraining them.
Google is a good example here with its search business. They could offer LLMs in search, but that could hurt their search business economics (adding inference costs to every search query), and so they have to do so in a way that is potentially separate unless forced to. Microsoft (an incumbent in most markets but not in search), Perplexity, and others don’t have that issue, and so can focus on offering the best experience to customers.
4. Data being centralized within companies
As touched on above, many incumbents have access to proprietary data which is created in many cases within their product. For example, Salesforce with Sales pipeline info, Intercom/Zendesk with Customer Support information, etc. This information would improve the use of LLMs.
But one benefit for startups is that there has been a trend at companies towards centralizing data from all the various tools into the data warehouse.So in some ways, less of the information is inaccessible, as long as the startup is able to sell into the company and gain access to the warehouse as part of the deployment.
Then, it can be on almost equal footing with incumbents in terms of having the organization’s data. In fact, it appears that at least initially incumbents are not leveraging the whole warehouse but just their internal data, so it is possible that startups may even have a data advantage.
5. Less reputational risk
Startups have less reputational risk than incumbents. Incumbents may have to worry about their AI going wrong in various ways, hurting customer sentiment or damaging their brand and reputation. Startups, on the other hand, have less to lose in terms of reputation and can take more risks in terms of using AI more fully (i.e., automating tasks, getting the wrong output once in a while, etc).
We saw this play out a little bit with Google. While OpenAI put ChatGPT out there, Google hesitated, because they had more risk and more to lose, as the below quote from Jeff Dean, head of Google AI, illustrates:
Dean told employees, emphasizing that the company has much more “reputational risk” and is moving “more conservatively than a small startup.”
“We are absolutely looking to get these things out into real products and into things that are more prominently featuring the language model rather than under the covers, which is where we’ve been using them to date,” Dean said. “But, it’s super important we get this right.”
What should startups do?
There’s a common adage, first popularized by Adam Rampell, that the battle between every startup and incumbent comes down to whether the startup gets distribution before the incumbent gets innovation. It seems that in this battle, the incumbents are getting the basic innovation done pretty quickly. So what can startups do?
To compete with incumbents in the AI race, startups must focus on their advantages and leverage them to their full potential. Here are a few strategies startups can use:
1. Focus on specific verticals/problems
For many horizontal-type use cases, if AI is the only differentiator that a startup is seeking to provide, it is likely that it will lose out to the incumbents on horizontal use cases, especially if it can just be tacked onto a product.
Startups should aim to hone in on specific problems or vertical-use cases and build solutions for those. It’s less likely that incumbents will tackle these specifically and there may be more room to innovate by rethinking the workflows and processes for specific problems.
2. Leverage Proprietary Datasets
It’s critical that startups build more than a simple wrapper around the foundation models. One way to do so is by leveraging additional data, for which startups have a few options:
Get data from the company/customer via integrations with data sources and the warehouse
Create proprietary data in your application via usage over time (make it some form of a system of record)
Purchase external data relevant to the use case
Convert some of the unstructured data in the company (images/videos/handwritten forms) into structured data which you store
3. Think AI-Native
It’s likely that no matter what category you’re in, the incumbent in your category will bolt on using the models in their application in “the obvious way” within a year. To differentiate, startups have to use the fact that they’re starting from scratch and use that to their benefit, and rethink what the problem they’re solving might look like with AI to support.
Startups should be thinking about what the end-state workflows could be, and how to iteratively work towards that. It’s likely that they’ll have to insert themselves into existing workflows as opposed to completely reimagining them immediately, but it’s important to always think about the bigger picture.
4. Leverage New Demand Channels
Some of the incumbents may be more hesitant to build things like ChatGPT plugins or leverage those potential forms of demand. As a startup, it’s important to use any distribution benefits to your advantage. If customers are there, then you should be building for those channels, even if it may mean the risk of not fully owning customers in some cases.
5. Innovate on User Interfaces
The ideal user interfaces to use these AI products are likely not fully developed. Incumbents may be more hesitant to completely redesign their product especially early on, but startups have more of that luxury since they are starting from scratch.
Whether it’s using voice, a chat interface that can search or directly take actions or a command bar combined with search, there’s a lot to explore around what the future of interfaces for many applications looks like and tying the AI in very tightly with the application.
Thanks for reading! If you liked this post, give it a heart up above to help others find it or share it with your friends.
If you have any comments or thoughts, feel free to tweet at me.
If you’re not a subscriber, you can subscribe for free below. I write about things related to technology and business once a week on Mondays.
Note that I focus more on AI applications rather than building foundational models or infrastructure for example
This point is less true in B2C contexts, where most of the data still resides with Amazon, Google, Facebook, etc, and is rarely centralized by the user