Intercom and Microsoft both made big AI feature announcements this week. BuzzFeed’s stock price soared after it announced that it will use AI to generate content. Incumbents are integrating AI rapidly, and this should come as no surprise.

Given I’m building in AI, this question is front and centre for me. I’ve been thinking a lot about how startups might compete given the existing distribution that incumbents have.

Incumbents have three key advantages: distribution, proprietary data and capital.

Incumbents have distribution.

There’s no better example of this than Microsoft, and there’s no better graph that the one below to tell the story. Slack had a superior product, but did what it does best: leveraged its distribution.

Microsoft announced a $7/month premium plan that leverages OpenAI’s tech to take notes for you. Tough market for anyone operating in the productivity space. For starters the price point is very low. More importantly, if Teams is already installed, you can have automatic meeting minutes with the click of a button.

Incumbents have built large datasets which gives them an advantage with AI.

Intercom this week announced a bunch of AI features: summarisation of chats, adjusting tone, rephrase, expanding text and generation.

Intercom has a tonne of data on what customers asked, what response was given and whether this was satisfactory for the customer.

From my building in AI, I can tell you that the output is several magnitudes better if you can provide examples or context. If we assume that at the start, Intercom will use off the shelf APIs, they can enrich each request with relevant context and get a better response.

If they ever wanted to do, they can spin up their own model using all the data they have. Incidentally, Intercom already has a neural net that generates responses for questions from customers.

Incumbents have capital and this a BIG advantage in AI. Microsoft has $13 billion in cash on its balance sheet. Hard to compete with that.

There are two costs in AI: training and inference. The money goes a long way in the AI world. Incumbents have the ability to price lower and subsidise the market. They can run more experiments. They can afford to train their own models.

I’m realising this essay is sounding mostly like doom and gloom, but it’s not!

Startups can and always will have the ability to compete, but it’s going to be hard. Some folks believe that incumbents will capture all the value.

Note the nuance in Sam’s point: he isn’t saying that incumbents will always capture value of a platform shift. He’s saying that generative AI isn’t actually a platform shift.

I disagree and believe that it is a platform shift. I believe this because of AI does two things: a) it increases the output from a person when applied to the right problem, and b) it allows someone to do something that they could not otherwise do (e.g. create digital content).

Based on the above, here are some ways in which a startup can compete with an incumbent:

Target enterprise customers. Many enterprise customers will not want to share their data with OpenAI. Can you help them leverage AI on their own servers without sending data outside? Imagine wanting to leverage AI for customer service. No one is going to want to send personally identifiable information to GPT3. Is that even allowed?

Typical SaaS advice: pick an industry and go vertical. If you become the platform for e-commerce and build your product purely for that buyer persona. You will have an advantage over the general products from incumbents. Pick a big market, find inefficiencies and start building.

There’s opportunity to rethink the experience from first principles. This is a long-term, high-risk, high-reward bet.

Consider the Teams feature that lets you record notes and meetings. The overarching outcome here is for people within a company to reach consensus on something and take action. Is that the best way to achieve it? What if you could build a product that helped avoid the meeting entirely?

Startups are not tied to an existing customer base or experience, and have a clear advantage here.

Finally, moving fast is a clear advantage. Most incumbents, though not all of them, will move slowly because of organisational inertia, risk appetite or some mix of the two. Use this to your advantage and get stuff out of the door quickly. On a tangential note, Microsoft has really found the perfect balance here with their OpenAI partnership. More on that here.

Someone once told me that to build something valuable you have to have the perfect blend of optimism and pessimism. Without enough optimism, you give up too easily. Without enough pessimism, you fall in love with your idea even if it’s not working.

Be wary of incumbents but don’t let it stop you. A nice way to think about it is it takes a while to build something great. If it doesn’t take very long, someone else is going to build it and you’re going to have to fight really hard to win in that market.