AI readiness assessment: find out what stands between you and AI in production

Most AI programmes don't fail on the model. They fail on the things around it: data that isn't fit for purpose, no route to deploy and monitor, unclear accountability for risk, and teams without the skills to run what was built. A readiness assessment surfaces those conditions before you commit budget, and it gives leadership an honest view of what "AI-ready" would take.

• Strategy and use cases: a prioritised portfolio of use cases tied to measurable outcomes, with sponsors and success criteria.
• Data readiness: quality, coverage, labelling, access and lineage of the data each use case depends on. AI amplifies data problems; it doesn't fix them.
• Platform and MLOps: the ability to train, deploy, version, monitor and retire models and LLM applications, including evaluation and cost control.
• Governance and risk: policies for acceptable use, privacy, security, model risk and human oversight, mapped to the regulations you operate under, such as GDPR and the EU AI Act.
• People and skills: machine learning, AI and LLM engineering, data engineering and product skills, plus the literacy of the business teams who'll use the outputs.
• Adoption and change: whether workflows, incentives and leadership behaviour will actually absorb AI.

Rate each pillar against clear descriptors, from exploring to embedded. Then do two things simple averages miss: treat critical blockers (no governance owner, no deployment path) as gates rather than averaging them away, and score each priority use case separately, since a team can be ready for document search and nowhere near ready for automated credit decisions.

1. Fix the gating blockers first: governance ownership, data access and a deployment path.
2. Pick one or two use cases with high value and high feasibility, and ship them end to end.
3. Build the reusable pieces as you go: evaluation, monitoring, prompt and model versioning.
4. Close skills gaps deliberately, by hiring or upskilling, rather than hoping pilots teach them.
5. Reassess each quarter.

AI readiness sits on top of data maturity. If the core assessment shows weak foundations or governance, fix those before scaling AI. See our data maturity assessment guide.

A financial services team has a successful chatbot prototype built on internal policy documents. The readiness assessment shows strong sponsorship and a good use case, but three gates: no evaluation process to catch wrong answers, no agreed owner for model risk, and document access rules that the retrieval layer ignores. Scaling the prototype as it stands would create real exposure. The plan becomes: put an evaluation harness and access-aware retrieval in place, name a model-risk owner, then roll out to one department and measure.

Which use cases are we funding, and why? Who signs off model risk? Can we deploy, monitor and roll back a model today? Do we have the skills in-house, or do we need to hire? If any answer is unclear, that's where readiness work starts.

The carbon.iq Pro AI Readiness assessment scores your organisation across these pillars. Results feed the AI Advisor, which answers questions grounded in your own scores, and the Intelligence Roadmap, which turns gaps into a prioritised plan. carbon.iq Pro also includes every recommendation for your gaps, the Team, Technology Stack and extended capabilities assessments and native Google Workspace / Microsoft 365 export. Pro is $19 a month (charged as R300). See carbon.iq Pro →