What to Ask Before Starting an AI Project

Who this is for

Business owners and IT leaders evaluating an AI initiative and wanting to ask the right questions before committing budget.

Question this answers

What should we assess and what questions should we ask before starting an AI project?

What you'll leave with

Business case questions that reveal whether AI is the right solution
Data readiness questions that prevent expensive surprises
Technical questions to understand scope and feasibility
Red flags in vendor proposals that signal trouble

Why these questions matter

Most AI projects that fail don't fail because of bad technology. They fail because the business problem wasn't well-defined, the data wasn't ready, or the expectations didn't match reality.

Asking the right questions upfront costs nothing and can prevent expensive mistakes. Here are the questions we ask every client before recommending an AI approach.

Business questions

Name the problem before you name the tool

What specific business problem are we solving?
If the answer is "we want to use AI" rather than a specific problem, step back.
How do we solve this problem today?
Understanding the current process reveals what AI needs to improve.
What does success look like?
Time saved? Cost reduced? Quality improved? Define it before you start.
How will we measure success?
If you can't measure it, you can't prove it worked.
What happens when the AI gets it wrong?
Every AI system makes mistakes. Plan for error handling and human review.
Who will use this and how often?
Usage patterns determine whether the investment makes sense.

Data questions

Is your data ready for AI?

Where does the relevant data live?
Multiple systems? Spreadsheets? Email? The more scattered, the more prep work.
How clean is the data?
Inconsistent formats, duplicates, and gaps all need fixing before AI can use it.
How much data do we have?
Not all AI needs big data, but RAG systems need sufficient content to be useful.
How often does the data change?
This determines whether you need a one-time pipeline or ongoing sync.
Are there privacy or compliance constraints?
Healthcare, financial, or personal data adds requirements.

Technical questions

Does this need AI or would rules-based automation work? If the logic is fully definable, automation is cheaper and more reliable than AI.
Where will the AI run? Cloud, on-premises, or hybrid? This matters for latency, cost, and data privacy.
What systems does it need to integrate with? APIs, databases, document stores, email systems, each integration adds complexity.
What's our tolerance for error? An AI that's right 90% of the time is amazing for some uses and unacceptable for others.
Do we need explainability? Some use cases (legal, medical, compliance) require the AI to show its reasoning.

Questions for AI vendors

Questions to put to any AI vendor

Can you show a similar project you've delivered?
What data preparation is needed before we start?
How long will a proof of concept take?
What are the ongoing costs after launch?
How do you handle model updates and drift?
What happens if we want to switch vendors later?

Red flags in vendor responses

Warning signs in an AI pitch

Promising specific accuracy before seeing your data
Skipping the discovery phase and going straight to development
Unable to explain how the AI will work in plain language
No mention of data preparation, testing, or error handling
Quoting on a fixed scope without understanding your data
Using buzzwords without concrete deliverables

Key takeaways

Start with the business problem, not the technology. "What are we trying to improve?" comes before "Should we use AI?"
Data quality is the number one risk factor in AI projects. Assess it before you commit budget
If you can't describe how you'd measure success, the project isn't ready
Be wary of vendors who promise specific accuracy numbers before seeing your data
A $5K-$15K discovery phase can save you from a $100K mistake

AI StrategyProject PlanningRAG