AI readiness assessment: a checklist for any industry

Why assess readiness before starting?

Most stalled AI efforts do not fail because the model was weak. They fail because the problem was vague, the data was scattered, nobody owned the risk, or the people who would use the result were not involved. A short readiness assessment surfaces those issues while they are cheap to fix, and helps you choose a first use case that can succeed.

What does a readiness assessment cover?

Six areas matter in almost every industry. Each one can be assessed with a handful of direct questions.

1. Business case and strategy

Is there a specific problem with a named owner and a measurable outcome? Can you describe how work changes if the AI works? Is leadership willing to fund a pilot and act on what it learns?

2. Data

Do you know where the relevant data lives, who owns it and how good it is? Can the team access it without months of approvals? Is there a data catalog, master data for customers and products, and a way to check quality before models consume it? Is there enough history, and is it representative of the situation you want to predict or support?

3. Governance, security and privacy

Do you have rules for acceptable AI use, a way to classify risk, and a process to review models before and after launch? Is personal or confidential information handled under clear policies, in line with the privacy laws that apply to you, such as the Personal Information Protection and Electronic Documents Act (PIPEDA) in Canada, the General Data Protection Regulation (GDPR) in Europe or U.S. state laws? Is there human oversight for decisions that affect people?

4. Technology

Is there a platform where data can be prepared, models built and results delivered into the tools people use? Examples include a data warehouse or lakehouse, orchestrated pipelines, a feature store and model registry for machine learning, and retrieval-augmented generation (RAG) with vector search for generative AI. Can you monitor, secure and update what you deploy, using MLOps practices, infrastructure as code and continuous integration and delivery (CI/CD)? Does it fit your existing cloud, whether AWS, Microsoft Azure, Google Cloud or another?

5. Skills and operating model

Who will build, who will run and who will use the solution? Do you have, or can you borrow, data engineering, analytics, security and product skills? Is there a clear route from a pilot to a supported service?

6. Change and adoption

Have the people affected been involved from the start? Is there training, and a plan for how roles and processes change? Staff who distrust or avoid a tool will quietly defeat the best model.

How do you score yourself?

Rate each area on a simple four-level scale. The scores are for conversation and prioritization, not for external reporting.

Area	Early	Developing	Ready to pilot
Business case	Ideas without owners	Owner and goal, no baseline	Owner, goal, baseline and funding
Data	Scattered, unknown quality	Located and partly profiled	Owned, accessible and fit for the use case
Governance	No AI-specific rules	Draft policy, informal review	Risk tiers, review and monitoring in place
Technology	Manual, spreadsheet-based	Some shared tooling	Platform able to build, secure and run models
Skills	Dependence on a few individuals	Team forming, gaps known	Roles covered for build and run
Adoption	Users not yet consulted	Sponsors engaged	Users involved, training planned

What does the checklist look like in practice?

• Choose one use case with a named business owner and a measurable goal.
• Capture a baseline of how the process performs today.
• Identify data sources, with owners and a quality check.
• Complete a privacy and security review of the data and the intended use.
• Set a risk tier and decide where a person stays in the loop.
• Reserve budget and time for a pilot, with agreed success and stop criteria.
• Name a team to build, to run and to use the result.
• Plan how to monitor quality and handle errors after launch.

What might this look like for a hypothetical organization?

This hypothetical example is for illustration only. A regional distributor wants AI to forecast demand. The assessment shows a clear owner and goal, but sales history lives in three systems with different product codes, nobody owns data quality, and there is no policy covering AI use. The sensible first step is not a model. It is to consolidate one product family's history, name a data owner, and write a short AI use policy. With those in place, a small forecasting pilot can run with a fair chance of success.

What should you do after the assessment?

Choose the one or two use cases where value is clear and the gaps are small. Fix the gaps that block those use cases first, rather than trying to raise every score at once. Run a time-boxed pilot with agreed measures, review what you learned, and decide whether to scale, adjust or stop. Repeat the assessment when your portfolio of use cases grows, because readiness is not a one-time state.

Where ARCHR fits

ARCHR can help organizations run an AI readiness assessment, translate the findings into a prioritized roadmap, and deliver a pilot with governance, security and human oversight designed in. We work across industries on AWS, Microsoft Azure and Google Cloud, and shape the approach around your existing stack and your risk profile.

This article is for general information only and is not legal, financial or investment advice.