How to create and use technical wedges strategically as a software business. Lenny Rachitsky of Lenny's Newsletter wrote a piece on finding a wedge. He defines a wedge as follows,
A wedge is simply a strategy to win a large market by initially capturing (1) a tiny part of a larger market or (2) a large part of a small adjacent market.
He goes on to give examples across different industries that have been impacted by software: Doordash, Stripe, Uber, Shopify, and more. He lists APIs like Twilio and Stripe as wedges, so I thought I'd expand a little more on two types of technical wedges I've seen.
Lenny explicitly calls out a quote from Datadog CEO Oliver Pomel, saying that they didn't have a wedge. I'd argue, the observability agent is the wedge. Like an API acting as a wedge for Stripe or Twilio, the agent is a lightweight way to collect metrics and start seeing value from a observability pane. Better yet, Datadog based its agent on open standards, making it easier for companies to adopt. I think that this fits into one of Lenny's criteria for a good wedge,
Educates the market—sometimes customers aren’t ready for the bigger transformation, and need to start with a small dose
As a more general rule, you can sometimes use a data plane as a wedge. The data plane can often operate without a control plane, but not as well. For example, imagine a logging agent with a control plane – it's still collecting logs, and you may aggregate them manually or simply use them to debug. Likewise, a kubelet (or Docker daemon) is useful on it's own to run workloads, but isn't sophisticated enough to make up a multi-node distributed system.
A remix of Chris Dixon's idea: come for the tool, stay for the platform.
Scale is great example of a company creating a wedge into ML Ops. Scale started with data labelling tools – before companies can use their data, it needs to be cleaned. I'll call this the dataflow wedge. It satisfies two of Lenny's criteria for a good wedge,
Builds momentum—can be sold quickly and keeps customers coming back
Naturally extends into a much bigger opportunity—more product, more revenue, more users
Labeling data is a painful, yet small, part of the journey to data-driven decision making. It builds momentum because data has gravity. And its easy to see how it naturally extends into the much bigger opportunity of helping these companies with their full data science and machine learning workflows. By wedging into dataflow, you can craft and integrate the upstream and downstream interfaces.
Another company that does this well is dbt Labs, which started as a tool to do data transformations on data warehouses. Since it sits on top of the data warehouse, it has become the natural extension point for downstream tasks like testing, documentation, and metrics definition.
You can also create dataflow wedges in more creative ways. Ramp and Brex are spend management companies that found their wedge with issuing corporate cards that didn't require a personal guarantee. Lenny's first criteria for a good wedge,
Is narrow and focused—solves a very specific problem, for a specific group, extremely well
These companies aren't valued on their effectiveness in providing credit cards to companies, but rather their new roles as systems of record for company transactions. They can expand easily into different adjacencies and provide solutions because they already have the context that many spend platforms don't. Today, it's difficult for startups to have this level of control over a dataflow – many times products have to wedge within an existing dataflow – e.g. dbt on Snowflake.
There's other types of technical wedges out there that are more nuanced – wedges at the network layer, wedges that can turn into platforms, and wedges that can encapsulate or unbundle layers. For another post.