Every week, a new AI tool launches. Every week, someone tells you it's the thing that will finally change how you work.
Most of them are not.
The problem isn't that AI tools don't work. The problem is that most of them aren't worth paying for. They solve a problem that doesn't exist, or solve it worse than the tool you already use, or solve it for a version of your workflow that isn't your actual workflow.
Before you add another subscription, ask three questions. If a tool doesn't answer all three well, don't buy it.
Question 1 — What specific task does this replace?
Not "what does it help with." Not "what does it do for my business." What specific task does it replace?
Concrete answers sound like:
- "Drafting responses to the 30 customer emails I handle daily"
- "Turning 20 product photos into 200 variations for social media"
- "Reading contracts and flagging the 3 clauses I need to worry about"
Vague answers sound like:
- "Improving productivity"
- "Saving time on repetitive tasks"
- "Helping me work faster"
Vague answers mean the tool doesn't actually know what problem it solves. That usually means it doesn't solve a real one.
The test: if you removed the tool tomorrow and had to do things the old way, what would you actually miss? If the answer is specific, it's a real tool. If it's "I don't know, but it feels useful," it's probably not.
Question 2 — What does it cost per task?
Not "what's the monthly subscription." What does each task cost you?
This is where most AI tools reveal themselves.
Example:
- Tool A: $50/month, handles 500 tasks/month = $0.10/task
- Tool B: $20/month, handles 50 tasks/month = $0.40/task
Tool A looks more expensive. It's actually cheaper for the workflow.
The test: calculate the per-task cost against your actual volume.
If you process 20 items/month, and the tool costs $50/month, you're paying $2.50 per item. Is that worth it? If you process 500 items/month, and the tool costs $50/month, you're paying $0.10 per item. Completely different story.
Most AI tools market to high-volume users because that's where the economics work. If you're not a high-volume user, the economics may not work for you.
Related question: what's the cost when things go wrong? If a tool makes a mistake you have to fix manually, that's a hidden cost. Some AI tools require so much correction that they're slower than doing it yourself.
Question 3 — What do I lose if I stop using it?
This is the question most people skip and then regret.
Every tool creates some amount of lock-in. The question is how much.
Low lock-in: you could stop using it tomorrow and not lose much. A ChatGPT subscription, for example — you could switch to Claude or Gemini.
High lock-in: stopping would be disruptive. If a tool stores all your customer data in its own system and doesn't let you export it, stopping means losing your data.
Before committing to any AI tool, ask:
- Can I export my data?
- Does my workflow depend on this tool's specific format or interface?
- If this tool shut down tomorrow, what would I have to rebuild?
Tools with high lock-in deserve much more scrutiny than tools with low lock-in.
The test: imagine the tool's company announced it was shutting down in 6 months. What would you need to do? If the answer is "nothing, I'd just use something else," it's low lock-in. If the answer is "I'd have to rebuild everything," think very carefully before starting.
The framework, applied
Here's how to use these three questions in practice.
Tool: an AI tool that generates social media posts.
Question 1: What specific task does it replace? → "Generating 30 social posts per month that would normally take me 10 hours." Specific. ✓
Question 2: What does it cost per task? → $20/month ÷ 30 posts = $0.67/post. If it saves 20 minutes per post, that's $0.67 for 20 minutes of time. Reasonable. ✓
Question 3: What do I lose if I stop using it? → All the posts live on the platform, but I can copy them out. Low lock-in. ✓
Verdict: Worth trying.
Tool: an AI-powered CRM that handles all your customer interactions.
Question 1: What specific task does it replace? → "Managing all my customer relationships and follow-ups." Vague. ✗
Question 2: What does it cost per task? → $99/month, and it handles "customers" not "tasks." Per-customer cost depends heavily on volume. Might be reasonable, might not. ✗
Question 3: What do I lose if I stop using it? → All customer history, conversations, and relationship data lives in the platform. If I stop, I lose it. High lock-in. ✗
Verdict: Not a clear yes. Needs deeper evaluation before committing.
Tool: an AI assistant that drafts emails using your writing style.
Question 1: What specific task does it replace? → "Drafting the 20 client emails I write daily." Specific. ✓
Question 2: What does it cost per task? → $30/month ÷ ~500 emails/month = $0.06/email. Saves 5 minutes per email. Cheaper than your time. ✓
Question 3: What do I lose if I stop using it? → Nothing. Your emails are yours. Low lock-in. ✓
Verdict: Definitely worth it.
What to do when a tool fails the framework
If Question 1 fails (vague task): don't buy it. The tool is looking for a problem, not solving one.
If Question 2 fails (poor unit economics): don't buy it. Or wait until your volume justifies it.
If Question 3 fails (high lock-in): be extremely careful. Consider a trial period before committing. Read the terms of service carefully. Export your data regularly.
If two or three questions fail together, don't buy it. The tool is probably being sold to you rather than solving something real.
A fourth question, honestly
There's a fourth question that most AI tools won't want you to ask, but it's worth considering:
Would this tool still be worth it if the AI part didn't exist?
Meaning: strip away the "AI" label from the tool. Would the underlying function still be useful?
Most AI tools are wrapping a fairly ordinary function in an AI label. Sometimes that matters. Often it doesn't.
If a scheduling tool charges $50/month because it "uses AI to suggest meeting times," but a $10/month scheduling tool does the same scheduling without AI, the $50 one is charging for the label, not the function.
The test: if the marketing didn't say "AI," would you still want it?
If yes → it's a real tool. The AI is an enhancement. If no → the AI label is the product. Skip it.
The uncomfortable truth about AI tools in 2026
The market is noisy. Every SaaS company has added an AI feature, every startup is pitching AI, every marketing email mentions AI.
Most of these tools will not exist in 3 years. Not because AI failed, but because the tools solved problems that weren't problems.
The AI tools worth using share two characteristics:
- They solve a real, specific workflow.
- They'll still be useful when the AI label stops being a selling point.
The tools that survive the next few years will be the ones that pass the framework above: real task, real economics, low lock-in, real function.
Everything else is noise.
One last thing
You don't need to be an AI expert to evaluate AI tools. You need to ask basic questions that apply to any software purchase.
What does it do? What does it cost? What happens if I stop using it?
Those three questions are enough to save you from most bad decisions. And they'll keep you from buying the tool of the week — which is a habit that, over a year, costs thousands of dollars for tools you don't actually use.
Buy less. Use more. Replace what doesn't work.
That's the whole strategy.
Want to know what to actually automate first? Read what small businesses should actually automate first. Or if you're evaluating specific tools, start with our honest guide to AI tools for small business owners.
