AI projects fail for predictable reasons. The good news is that most of those reasons are identifiable before a single line of code is written — if you ask the right questions at the start. Here is the checklist we use with every prospective client before we agree to take on a project.
1. Do you have a specific, measurable problem to solve?
"We want to use AI" is not a project brief. "We want to reduce our contract review time from 8 hours to under 1 hour" is. Every successful AI project starts with a specific, measurable business problem. If you cannot articulate what success looks like in concrete terms — time saved, error rates reduced, conversion rates improved — you are not ready to start.
2. Do you have the data to train or ground the AI?
Most AI systems require either training data (for custom models) or retrieval data (for RAG systems that answer questions based on your knowledge base). Before committing to a project, inventory your data: where it lives, how much of it there is, how clean it is, and whether you have the right to use it for AI training.
3. Is the data good enough to use?
Answering "yes" to question 2 does not mean the data is ready. Common data quality issues that derail AI projects: inconsistent formats, missing values in key fields, outdated records, and data that was collected for a different purpose than what you are trying to use it for. Budget for a data quality assessment before the AI build begins.
4. Who owns the AI project inside your organization?
AI projects that lack an internal champion fail at a higher rate than those with one. The champion is not just a sponsor — they are someone who understands the problem, can make decisions about requirements, has the authority to clear roadblocks, and is accountable for the outcome. If no one fits this description, you have a governance problem to solve before an AI problem.
5. What does the impacted team think?
The people whose workflow the AI will change need to be involved early, not informed late. AI systems that are designed without input from frontline users consistently underperform because they miss edge cases, misunderstand the actual workflow, and face adoption resistance at launch. Involve the team in requirements definition from the start.
6. What does failure look like, and how bad is it?
Risk calibration is essential. In a customer service chatbot that handles routine inquiries, occasional errors are tolerable because a human agent can catch and correct them. In a financial compliance system that makes automated decisions, errors may have regulatory consequences. Understanding your error tolerance shapes every architecture decision.
7. How will you measure success?
Define your KPIs before you start, not after. Time saved per task, error rate reduction, throughput improvement, cost per transaction — whatever metrics matter for your use case, agree on them upfront, establish baselines, and build measurement into the system from day one. You cannot evaluate success without baselines.
8. What happens to the people whose work changes?
AI implementations that eliminate or significantly reduce work need a plan for the people doing that work. This does not necessarily mean job cuts — in most cases, the freed capacity gets redirected to higher-value work. But the plan needs to exist before deployment, not be improvised afterward when people are anxious and resistant.
9. What are the compliance and privacy requirements?
Depending on your industry and jurisdiction, AI deployment may require privacy impact assessments, regulatory approval, or specific data handling practices. Healthcare AI in BC must navigate PHIPA. Financial services AI must consider FINTRAC and OSFI guidance. Legal and professional services have confidentiality obligations. Map these requirements before you design the system.
10. What is your plan for the system after launch?
AI systems are not set-and-forget. Models drift as the world changes. Business requirements evolve. Performance degrades. You need a plan for ongoing monitoring, maintenance, and optimization — including who is responsible, what the SLA is, and what the escalation path looks like when something goes wrong. If the answer is "the vendor handles it," make sure that is explicitly defined in your contract.
What To Do With Your Answers
If you can answer all ten questions clearly, you are ready to move forward with confidence. If several answers are unclear or missing, that is useful information — it tells you what to resolve before engaging an AI vendor.
At SysBuddies, we go through this checklist with every prospective client in our initial strategy call. We have declined projects where the answers suggested that a client was not ready — because those projects would have failed, wasting both the client's investment and our time. We would rather help you get ready than take on a project set up for disappointment.