Most AI projects fail before any model is trained, because the organization wasn't ready for what the project actually required — not because AI was the wrong idea. A readiness assessment is how you find that out before spending a quarter's budget on it.
Here's the framework we use, condensed into something you can run yourself with your own team in a couple of working sessions.
1. Audit your data before you audit your use cases
It's tempting to start by brainstorming AI use cases. Start with data instead — the use cases you can actually pursue are constrained by what data you have, not by what sounds valuable. For each candidate use case, ask three questions: is the data available at all, is it accessible without a six-month integration project, and is it clean enough to trust an automated decision on. A use case that fails any of these isn't dead, but it needs a data project before it needs a model.
2. Score opportunities on feasibility and impact, not excitement
Every organization has a list of AI ideas generated in a leadership offsite. Very few of those lists are scored on anything other than how compelling they sounded in the room. Score each candidate on two axes: feasibility (do we have the data, the integration points, and the technical path to build this in a reasonable timeframe) and impact (does solving this actually move a metric leadership cares about). The highest-value quadrant is high impact, high feasibility — and it's usually smaller and less glamorous than the initial wishlist.
3. Assess your infrastructure honestly
- Do you have a way to deploy and version models, or will the first production model be the first thing your team has ever deployed this way?
- Do you have monitoring and logging in the systems the AI would integrate with, so you can tell when something breaks?
- Do you have a data pipeline that can deliver fresh, correctly-labeled data on an ongoing basis, or was the training set a one-time export?
- Is there a security and compliance review process that AI projects can actually pass, or will this stall in legal review for reasons nobody anticipated?
None of these need to be fully solved before you start. But if the answer to most of them is no, your first project should build this infrastructure — not chase the most ambitious use case on the list.
4. Check organizational readiness, not just technical readiness
Who owns the outcome once it's live? Who has the authority to change the process the AI is now part of? Is there a team that will actually use the output, and have they been part of defining what "good" looks like? Projects that skip this step tend to produce technically correct systems that nobody adopts, because the people whose workflow changed were never consulted.
5. Build a roadmap, not a single bet
A readiness assessment should end with a phased roadmap, not a single go/no-go decision. Sequence projects so early wins build the infrastructure, trust, and internal expertise that later, more ambitious projects will need. A 12–18 month roadmap with clear milestones gives leadership something concrete to fund incrementally, rather than one large bet that has to be right on the first try.
The goal of a readiness assessment isn't to find the most impressive AI use case. It's to find the one your organization can actually ship, learn from, and build on.
If you want a second, outside opinion on where your organization actually stands, this is the exact engagement we run for clients before any engineering begins — data audit, opportunity scoring, infrastructure review, and a prioritized roadmap you can act on.