Free resource
The AI project feasibility checklist.
The 20 questions we run through before taking on any AI engagement. If you can answer these, you are ahead of most projects we see — and if you cannot, better to find out now than six figures later.
01
Problem fit
Most failed AI projects were never AI problems. Check these first.
- Can you describe the outcome in one sentence without the word "AI"? (e.g. "cut invoice processing from 3 days to 1 hour")
- Does a human currently do this task successfully? If no human can do it, a model probably cannot either.
- Would a 90%-accurate system still be useful? If you need 100%, you need software, not ML.
- Is there a measurable number this project moves — hours saved, revenue, error rate?
- Have you checked whether a plain workflow automation (no AI) solves 80% of it cheaper?
02
Data readiness
AI projects die in data access more than in modeling.
- Does the data this system needs actually exist today, in a system you control?
- Can you get access to it within two weeks — or is it locked behind another team, a vendor, or compliance?
- Is there enough of it? (For document AI: hundreds of real examples, not five.)
- Do you know what "correct" looks like — labeled examples, historical outcomes, or a human who can judge output?
- Is the data legally usable for this purpose (PII, PHI, customer contracts)?
03
Budget & timeline reality
Honest numbers from real engagements, not vendor decks.
- A production-grade AI pilot realistically starts around $10K+ and 6–8 weeks. Anything cheaper is a demo.
- Do you have budget for month 2+? Models degrade, prompts drift, APIs change — production AI needs ownership.
- Is one named person on your side able to spend 2–3 hours a week with the engineering team?
- Do you need this in under 4 weeks? Then descope to a pilot, or do not start.
- Have you accounted for inference costs at real volume, not demo volume?
04
Risk & production readiness
The questions that separate demos from systems.
- What happens when the model is wrong — is there a human fallback path?
- Does the output need audit trails or compliance sign-off (HIPAA, SOC 2, legal review)?
- Who owns the system after launch — your team, or are you expecting the vendor to babysit it forever?
- Can the system be tested against a golden set before it touches real users?
- If the AI vendor/API you depend on doubles prices or deprecates a model, what breaks?
Get the checklist + practical AI delivery notes
Occasional, genuinely useful notes on shipping AI systems. No spam, unsubscribe anytime.
Scored well on the checklist? The next step is a 30-minute feasibility call — we will tell you honestly whether the project is worth building, and whether we are the right team for it.
See if we’re a fit