Stop Building AI Features Nobody Asked For
I was sitting in a cramped demo room at a tech conference, watching a founder show off his SaaS product. He was a sharp guy, clearly passionate, and he had just raised a decent seed round. His product was a project management tool for remote teams, and he was about to show us the crown jewel of his latest release.
“We’re calling it the AI Insight Engine,” he said, beaming. The screen filled with a dashboard that analyzed every project, task, and deadline. Then, with a flourish, he clicked a button. A panel slid in from the right, containing a block of text: “Your team’s velocity is trending downward. Consider reducing scope for the next sprint to improve completion rates.”
There was a polite murmur from the audience. Someone nodded. The founder looked proud. I later caught him at the bar and asked, “What problem does that actually solve?” He paused. His smile faded. “Honestly? We needed to show AI. The board was asking about our AI strategy.”
That moment stuck with me. Not because he was dishonest—he was refreshingly honest—but because it perfectly captured the disease sweeping through the SaaS world. We are building AI features nobody asked for, and we’re doing it with the confidence of someone handing out free advice nobody wanted.
The Great AI Feature Gold Rush
Since ChatGPT went mainstream, every SaaS startup has felt the pressure. The playbook used to be: find a pain point, build a solution, charge money. Now it’s: find a pain point, build a solution, then bolt on an “AI copilot” that whispers generic suggestions into the user's ear.
I’ve seen it happen to a dozen startups. They add a chatbot that answers questions with hallucinated confidence. They add an auto-summarizer that condenses a 500-word email into three bullet points that miss the entire point. They add a “smart” recommendation engine that suggests features the user has already used or, worse, features they explicitly hid.
The result? Users click the AI button once, laugh nervously, and never touch it again. But the feature stays, consuming server costs, UI space, and engineering talent. It becomes digital furniture—clutter that everyone sees but no one sits on.
I call this the “AI Button” trap. It’s the belief that if you add “AI” to your feature name, it automatically adds value. It doesn’t. It adds a button. And buttons without jobs are just design noise.
A Tale of Two Features
Let me give you a concrete example. A few years ago, I worked with a small SaaS company that built a time-tracking tool for freelancers. Their users were designers, writers, and developers who needed to log hours, send invoices, and track project profitability. The product was solid, but the founder wanted to go bigger.
“We’re adding an AI assistant,” she announced at a team meeting. “It will analyze how users spend their time and suggest ways to be more productive.”
The team was skeptical. Users had never asked for this. They asked for better integrations with payment tools. They asked for a mobile app that didn’t crash. They asked for a simpler way to categorize expenses.
But the founder was adamant. She’d read a trending LinkedIn post about AI-powered productivity, and she didn’t want to be left behind. So the team spent three months building the assistant. They trained it on time-entry data. They created natural language prompts. They even gave it a cute name: “Pip.”
When Pip launched, the reaction was underwhelming. Users tried it, got generic advice like “You spend 30% of your time on emails—try batching them,” and then went back to logging their hours. The feature had a 2% weekly adoption rate. Worse, the engineering time spent on Pip meant the payment integration they actually needed was delayed by six months.
Meanwhile, a competitor launched a simple feature that let users send invoices directly from their email client. It wasn’t AI. It wasn’t flashy. But it solved a real problem, and it drove more signups than Pip ever did.
The lesson is painful but clear: if the user didn’t ask for it, it’s not a feature—it’s a distraction.
Why Do We Keep Doing This?
If the evidence is so obvious, why do we keep building AI features nobody asked for? Because we’re human, and humans are terrified of missing out.
There’s a phenomenon I call the “Copilot Cascade.” When a giant like Microsoft or Google releases an AI copilot, every startup with a similar product category feels a primal urge to respond. “If they have AI, we need AI.” Never mind that your startup has 50 users, not 50 million. Never mind that your users are niche and their workflows are idiosyncratic. The fear of being perceived as “behind” is stronger than the fear of wasting months on a useless feature.
Investors don’t help. I’ve sat in pitch meetings where VCs asked, “What’s your AI moat?” Founders, eager to please, invent answers. They promise AI-powered insights, AI-driven automation, AI-everything. Then they go back to their teams and say, “We need to build an AI feature before the next board meeting.”
The tragedy is that most of these features are not based on any user research. They’re based on competitive pressure and boardroom anxiety. It’s the software equivalent of buying a treadmill because your neighbor has one, then using it as a clothes rack.
The Real Cost of Unwanted AI
Let’s talk about cost, because “we’re just experimenting” is a dangerous mindset. Every AI feature has a bill—and it’s not just the API calls.
- Engineering time: Your best developers are spending weeks on model tuning instead of fixing the bug that crashes when users upload a large CSV.
- Maintenance burden: AI models drift. Data changes. You need to monitor, retrain, and update. That’s a permanent tax on your team.
- User trust: When an AI feature gives a wrong answer, users don’t just shrug. They lose trust in your entire product. One bad recommendation can undo months of goodwill.
- Onboarding complexity: Every new button, every new panel, every “Here’s what AI thinks” pop-up adds cognitive load. Your users are busy. They don’t want to learn a new mental model for a feature they never requested.
I remember a support tool that added an AI sentiment analyzer. It would scan customer emails and flag “negative tone” with a red warning. The company proudly announced it at a user conference. The users were horrified. They thought the AI was suggesting they were rude to customers. Within a week, the feature was disabled. But the damage was done—a Twitter thread about the “judgmental chatbot” went viral, and the startup had to apologize.
Unwanted AI features are not neutral. They actively harm your product by adding complexity and eroding trust. You’re not just wasting time; you’re making your product worse.
What Do Users Actually Want?
Let’s step back and ask a fundamental question: what do users want from your SaaS product? They want to complete a job with less friction. They want to feel competent. They want to get back to their actual work.
A few years ago, I worked with a legal document management startup. Their users were paralegals who spent hours tagging contracts with metadata. It was tedious, error-prone work. The startup’s founder noticed that users were manually typing the same tags over and over. So she built a simple auto-tagging feature using a basic machine learning model. It wasn’t flashy. There was no chatbot, no “insight engine.” It just learned from the user’s past tags and suggested the next one.
Adoption was nearly 90%. Users loved it because it saved them time. They didn’t care about the AI. They cared about not having to type “non-disclosure agreement” for the hundredth time.
That’s the secret. The best AI features are invisible. They don’t announce themselves. They don’t have a cute name. They just quietly make the user’s life better.
Consider Google Maps. It doesn’t say “AI-powered route optimization” every time you drive. It just tells you the fastest way. Consider Grammarly. It doesn’t ask you to “unlock the power of natural language processing.” It just underlines a passive sentence. The best AI is a utility, not a spectacle.
How to Stop Building Features Nobody Asked For
So how do you break the cycle? How do you ensure your next AI investment is actually something your users want? It’s not complicated, but it requires discipline.
1. Listen to the complaints.
Your support tickets are a goldmine. Every time a user says “I wish I could…” or “It’s annoying when…”, that’s a potential feature. But don’t jump to AI first. Ask yourself: is AI the right tool for this job? Sometimes a simple filter or a keyboard shortcut is enough.
2. Conduct “job interviews,” not “feature brainstorming.”
Talk to your users about their daily workflow. Ask them to show you how they use your product. Notice where they pause, where they curse, where they switch to a spreadsheet. Those moments of friction are your opportunity. If an AI feature can remove that friction, great. If it can’t, don’t build it.
3. Build a “test of no.”
Before you commit to a feature, ask your team: “If we built this, would users be upset if we removed it?” If the answer is “they probably wouldn’t notice,” then it’s not a feature. It’s a toy. You can apply this test to existing AI features too. If you removed the AI button tomorrow, would anyone care? If not, you might be wasting money.
4. Ship a tiny version first.
Don’t build a full “AI insight engine” with a dashboard and natural language prompts. Build a single, narrow use case. For example, instead of “AI-powered search,” build “semantic search for tagged documents.” Test it with five users. Measure adoption. If they don’t ask for more, stop.
5. Be boring.
This is the hardest one. In a world where every startup is screaming about AI, being boring feels like failure. But boring products are profitable. Boring features are reliable. Boring is what gets you to a million dollars in ARR. I’ve seen a SaaS company that helps landlords screen tenants. They added AI to detect forged pay stubs. It’s not a feature they market heavily, but it saves their users hours of manual review. That’s the kind of AI that builds a business.
The AI Feature That Actually Worked
Let me end with a success story. I consulted for a small CRM startup that sold to real estate agents. These agents were drowning in follow-up emails. They had to send personalized messages to every lead, and most of them were terrible at it. They didn’t need a “copilot” that wrote entire emails. They needed help with the first sentence—the icebreaker.
So the startup built a tiny AI feature that suggested a first line based on the lead’s profile. For example, if the lead was a first-time homebuyer, the AI suggested: “Congrats on starting your home search! What’s your top priority for a first home?” That was it. No full email generation. No “smart follow-up cadence.” Just a single sentence.
The feature was a hit. Agents used it because it made them feel more confident, not because they cared about the AI. The startup later expanded it to suggest follow-up questions, but only after users explicitly asked for that.
The founder told me, “We didn’t start with AI. We started with a problem: agents don’t know how to write engaging emails. Then we asked, ‘Can AI help?’ It could. So we built the smallest possible version.”
That’s the mindset we need. Start with the user, not with the technology.
A Final Plea
If you’re a founder or a product manager, I’m asking you to pause before you build your next AI feature. Ask yourself: Did anyone ask for this? Not your investors. Not your competitors. Not your own ego. Your actual users.
If the answer is “no,” you have two choices. You can build it anyway and hope it becomes a magical surprise. Or you can spend that time talking to your users, finding out what they actually need, and building the boring, useful thing that makes them love you.
I know which one pays the bills.
The AI hype cycle is going to keep spinning. There will be new models, new frameworks, new buzzwords. But the fundamentals of product design haven’t changed. Solve a real problem. Make it simple. Let the AI be invisible.
If you want to read more about pragmatic product thinking and how to avoid the traps of shiny technology, I’ve written about this extensively on my blog. Check out https://www.harishapc.com for more ideas on building SaaS products that people actually use.
And next time someone shows you their new AI feature, ask them one question: “What problem does that solve?” If they can’t answer, you know they’re building it for the wrong reason.
Let’s stop building AI features nobody asked for. Let’s start building products people love. The AI will come along for the ride—silently, invisibly, and only when it’s actually useful.
Top comments (0)