The AI Project Scoping Checklist

Most AI project estimates are wrong before the first call ends — because the questions that actually drive cost were never asked. This checklist is the set we run through when scoping AI work. Answer what you can before talking to any vendor (including us), and you'll get a real number instead of a shrug.

How to use this

Work through the five sections below — rough answers are fine, unknowns are useful too.
Anything you can't answer is a discovery item, not a blocker. Knowing what's unknown is half the value.
Send your answers with your enquiry and the first conversation starts weeks ahead.

1. The problem

One sentence: what changes for the business if this works? ("Support agents resolve tickets 30% faster" beats "we want an AI chatbot.")
Who uses it, how often, and inside which workflow or tool?
What does a wrong answer cost? A mild annoyance, a lost customer, or a regulatory incident? This single question sets the engineering bar.
How will you measure success? Hours saved, tickets deflected, revenue influenced — pick the number now, not after launch.

2. Data readiness

Where does the data live? List the systems — CRM, warehouse, wikis, spreadsheets, PDFs.
Can it be accessed programmatically, or does someone export it by hand today?
How clean is it? Duplicates, gaps, stale records, inconsistent formats — an honest guess is enough.
How fresh does it need to be? Real-time, daily, or "last quarter is fine" — each is a very different pipeline.

3. Integration surface

Which systems must it touch to be useful — and are you reading from them, writing to them, or both?
Do those systems have APIs, or are they legacy tools with no integration surface?
Who owns each system internally? Access approvals are a schedule risk, not a technical one.

4. Risk & compliance

Is the domain regulated (finance, healthcare, public sector)? Which frameworks apply?
Will the system see personal or sensitive data? PII handling changes the architecture, not just the paperwork.
Do you need an audit trail of what the AI said or did, and who acted on it?

5. Run & ownership

Who operates this after launch — your team, the vendor, or a mix?
Is there budget for run costs (hosting, monitoring, model usage, tuning) every year — not just the build?
What skills exist in-house today, and what would your team need to own the system fully?

A project scoped with these answers gets a smaller, sharper estimate — because nobody is padding for the unknowns you've already resolved.

What to do with your answers

Paste them into an email or the contact form, gaps and all. We'll come back within one business day with a clear read on feasibility, approach, and cost — free, no obligation. For the deeper reasoning behind these questions, read What Enterprise AI Really Costs.