There's a funny thing about building software. You spend hours writing documentation, explaining how features work, and putting together a nice FAQ page. Then someone signs up and asks a question that's literally answered in the first paragraph of your docs.
And you can't really blame them.
I've done the same thing with other products. If I get stuck, my first instinct isn't to dig through five documentation pages. I just want someone to tell me what to do.
This got me thinking while we were working on FeedLog.
Documentation doesn't always mean answers
A lot of SaaS products already have documentation. Some have really good documentation. But users still get confused, and founders still spend time answering the same questions.
The problem isn't necessarily that the information is missing. Sometimes it's just too much effort to find.
Imagine someone trying to integrate a widget into their website. They don't want to read the entire integration guide. They want to know which script to install, where to put it, and why it isn't working.
You might have written all of that down already, but from the user's perspective, the answer is still hard to find.
And if they're stuck long enough, they might just leave.
That's why we started working on AI customer support
When we were building FeedLog, we originally focused on product feedback. We had feedback boards, roadmaps, and changelogs, but after struggling to find users, we started looking more closely at the problems small SaaS teams deal with every day.
Customer support kept coming up in my thinking.
For a small team, answering questions isn't always a full-time job, but it interrupts everything else. You might be fixing a bug, working on a feature, or trying to find your next customer when another support question comes in.
So we added an AI customer support assistant and a built-in knowledge base.
The idea is pretty simple: instead of expecting every customer to search through documentation, let them ask a question directly. The AI can use the knowledge base to find relevant information and help them get an answer.
Of course, that doesn't mean AI will magically understand everything about your product. Someone still has to create useful documentation, keep it updated, and make sure the answers are accurate.
But it changes how customers interact with that information.
The knowledge base is only as useful as the content inside it
This is something I've been thinking about a lot.
It's easy to get excited about AI answering customer questions. But if the underlying documentation is outdated, incomplete, or confusing, the AI doesn't have much to work with.
You can't fix bad documentation just by putting a chatbot in front of it.
For example, if customers keep asking how to connect your product to their website, maybe the integration guide needs improvement. If people repeatedly ask about pricing, perhaps the pricing page isn't clear enough.
In that sense, customer questions are useful signals.
They show you where your documentation is failing, where your onboarding might be confusing, and sometimes where the product itself needs to change.
That's a much more interesting problem to me than simply building a chatbot that answers questions.
Not every question should be answered by AI
There's another thing I don't think we should ignore.
Sometimes customers don't need another AI-generated answer. They need a human.
Maybe they're reporting a bug that hasn't been documented. Maybe their account has a specific problem. Or maybe they're asking for a feature that doesn't exist yet.
That's why we're also building support workflows around the AI assistant.
The goal isn't to replace every human conversation. It's to handle the repetitive questions where possible and make it easier for teams to follow up when something needs real attention.
And when a conversation contains a feature request or product suggestion, it shouldn't disappear after the question is answered.
It should have a way to become product feedback.
That's where our original feedback boards, roadmaps, and changelogs still fit into the picture.
I'm still figuring out whether this is the right approach
One thing I've learned from trying to build a SaaS is that a product idea can sound completely reasonable in your head and still be something nobody wants to use.
We've been working on combining AI support, a knowledge base, and product feedback into one platform. The project is open source on GitHub, partly because we want developers to be able to explore how it works and run it themselves.
But I'm not going to pretend we've figured everything out.
I'm still curious how other developers think about documentation and support.
Do your users actually read your docs before asking for help? Have you tried using AI to answer questions from your knowledge base? And when the AI gets something wrong, does it save time or just create another problem?
I'd genuinely love to hear what has worked for you.
Top comments (0)