Free tool
AI feasibility checker
Many founders want “AI” in their product but don't know what's realistic. Describe your AI idea and find out exactly how to build it — or whether you should.
- Describe your AI idea in plain English — no ML knowledge needed
- Get a recommended approach: API, RAG, automation, fine-tuning, or custom model
- Understand data requirements and accuracy risks before you build
- Full implementation plan and architecture notes available free with email
API
Easiest approach
covers 70% of AI feature needs
RAG
For your own data
documents, knowledge bases, FAQs
Fine tune
When style matters
specialist tone, domain vocabulary
Custom
Rarely needed in v1
only when off-the-shelf fails
Is your AI idea actually buildable?
Describe what you want AI to do in plain English. Be specific about what it reads, what it produces, and who uses it. The more context, the more accurate the analysis.
Try an example:
Results are AI generated based on your inputs and current AI capabilities. They are a starting point, not a formal technical assessment.
Frequently asked questions
What does the feasibility score mean?
The score from 0 to 100 reflects how well-suited your described use case is to current AI capabilities. A score above 85 means the idea is straightforward to build with existing APIs like GPT-4o or Claude. Scores between 65 and 84 indicate achievable but non-trivial builds. Scores below 40 suggest that either the use case requires training data you likely do not have, or a rule-based system would be more reliable. The score is directional — not a guarantee of success or failure.
What is the difference between RAG and a direct API call?
A direct API call sends your prompt to an LLM and gets a response using only the model's training knowledge. Retrieval augmented generation (RAG) first searches your own documents or database for relevant content, then injects that content into the prompt so the model can answer questions grounded in your specific information. Use RAG when the AI needs to reason over content that was not in its training data — your product documentation, support tickets, or internal knowledge base.
When does fine-tuning make sense over prompt engineering?
Fine-tuning makes sense when you need consistent output style or format that prompt engineering alone cannot reliably produce, and when you have at least a few hundred high-quality labeled examples. For most early-stage products, start with prompt engineering — it is faster, cheaper, and easier to iterate on. Fine-tune only after you have validated the use case in production and identified specific gaps that prompting cannot close.
How much does running an AI feature typically cost per month?
At low volume, most AI features cost between $20 and $100 per month using GPT-4o or Claude Sonnet. At scale, costs depend heavily on token count and request frequency. A support email triage tool handling 500 emails per day might cost $30 to $80 per month. A document analysis tool processing large files could cost $200 or more. The detailed implementation plan includes cost estimates at different usage levels and suggestions for reducing cost through caching and model selection.
Do you really need AI in your product?
The most common mistake founders make with AI is adding it because it sounds impressive, not because it solves a real problem better than a simpler solution. Before building an AI feature, ask: could a well-designed search, filter, or rule-based system solve 80% of this problem at 10% of the cost?
When AI makes sense:the task requires understanding context or language, outputs vary based on nuanced inputs, the feature needs to generalise across many situations it hasn't seen before, or a human expert is currently doing this job manually.
When to avoid AI in v1:you don't have a clear definition of “correct” output, mistakes would cause user harm or legal liability, you don't yet have user data to validate the feature, or the same result could be achieved with a deterministic rule.
The four main implementation approaches:prompt engineering + OpenAI API (fastest, cheapest, works for most language tasks), RAG pipeline (best when you have your own documents or knowledge base), fine-tuning (best when you need a specific style or domain vocabulary and have labelled examples), and custom model training (rarely needed — only when existing models fundamentally can't solve the problem).