Emergence of Universal Integrations: how SaaS build customer-facing integrations in 2023
AI is changing how engineering teams build product integrations. If you are building a SaaS product, you should know what is happening.
Better integrations help you convert more prospects into customers and keep them longer. Research consistently show that integrations are becoming #1 criteria when choosing or switching business apps. Best SaaS products have hundreds of integrations, which becomes a major component of their competitive moat.
Large language models are about to disrupt this moat, and democratize the integrations landscape.
Before: Point-to-Point Integrations
Since popularization of APIs 15-20 years ago, all native integrations were built in the same way:
- Prioritize apps you need to integrate with.
- Pick the next one on the list.
- Research application, its API, and design the integration logic and UI.
- Code and test integration logic and UI.
- Go to step 2.
Depending on the complexity of the integration, it could take from days to months of engineering time to build each one.
Let’s say your app integrates with project management apps. Your integration development process would look something like this:
While researching and designing an integration is somewhat creative, the rest of the steps are pretty repetitive and tedious: you need to learn how to get access to a given API, how map its data structures to yours, and how to perform operations you need to perform.
As a result, building integrations is often delegated to junior engineers or third-party agencies, which further increases time to production and decreases quality.
But here come Large Language Models to completely revolutionize the process.
LLMs and integrations
There is an ongoing debate on how exactly AI and LLMs in particular will change the world, but one thing that LLMs are clearly good at is finding and using patterns.
Implementing integrations is 95% finding and applying patterns.
How do I get a list of tasks from a JIRA or Asana API? How do I map one set of fields to another? LLMs are great at answering this type of questions. All you need to do now is ask the right questions, which is the fun part of building integrations.
If used right, AI speeds up each integration development cycle by 90% of more. Now your process looks like this:
This would be great on its own, but with a small change in mindset you can take it to the next level. Enter Universal Integrations.
Universal integrations
What if instead of planning each integration separately, you planned each type of integration once, and implemented it for every application it applies to?
If you want your application to create a task in external application, instead of thinking “how should it create a task in JIRA?” and “how should it create a task in Asana?”, think “how should it create a task in any application that has tasks?”.
AI lets you scale this thinking to all the relevant apps automatically by generating application-specific code and data mappings, and the right integration framework lets you verify that whatever AI came up with actually works.
Thinking this way is harder to figure out because you need to keep in mind the range of possibilities (how tasks can be organized, which fields may be required, etc), but it is worth it. After switching to universal integrations mindset, you can build dozens or even hundreds of integrations in one cycle.
Most SaaS applications have a relatively small number of integration scenarios applied to a large number of external applications. They have the same integrations with all CRMs, all HR systems, all Project Management apps, etc.
The ability to build an integration for the whole category of applications at once radically changes how you think about integrations. Instead of thinking “what is the next app I want to integrate with?” you think “what is the next integration feature I want to add for every app we integrate with?”. It lets you iterate on your whole integration portfolio 10-100x faster than before.
How to start building universal integrations
If you have good enough engineering team, you can try to build the next integration this way: select not one but a dozen apps and ask them to build the feature that works with all of them at once, using AI to generate application-specific code. OpenAI APIs are cheap and simple to start.
Alternatively, try out integration.app - an AI-powered integration stack I’m working on. It is built around the concept of universal integrations and dozens of our customers successfully built integrations that work across whole categories of apps. We are still in invite-only mode, but if you have an interesting use case, we’d love to get you on board and level up the way you build integrations for your app.
