The AI Readiness Checklist: Is Your Business Ready for AI?

You can spot an unready company by what it tries to fix first. The teams that ask "which model should we use?" before "is any of our data even queryable?" are the same teams that, twelve months later, have a vendor invoice, a slide deck, and nothing in production. AI readiness for a business is not about the technology stack. It is about whether the operating substrate underneath — data, processes, people, governance — can absorb an automated decision without breaking in a way customers see.

The AI readiness checklist that follows is not a maturity model. Maturity models are for analyst firms and for IT leaders who want to look like they are advancing without committing to a number. This is a binary checklist: for each item, your business is ready or it is not, and the items you are not ready on are the project plan for the next ninety days. According to McKinsey's State of AI research, roughly 78% of organizations now report using AI in at least one business function — and the same body of research keeps finding that the gap between leaders and laggards is widening, not closing. The companies pulling away are the ones that figured out readiness before they shopped vendors.

The four pillars of AI readiness every business should pass before buying

Real AI readiness for a business sits on four pillars. A weakness in any one of them caps the value of everything you build downstream, no matter how good the model is. A serious readiness conversation walks each pillar in order — never as a single "are we ready?" question, which is unanswerable.

• Data readiness. Is the data you would feed an AI system actually accessible, cleaned, structured, and current? Most businesses overestimate this by an order of magnitude. The CRM has gaps. The contracts live in three folders, two SharePoints, and one ex-employee's laptop. The customer record in the support tool does not match the one in the billing system.
• Process readiness. Are the workflows the AI will touch documented well enough that you could hand them to a new hire? If a process only exists in the head of one tenured employee, an AI agent cannot learn it — and even if it could, the legal review of automating an undocumented process will stall the project for a quarter.
• People readiness. Does the team know what the project is for, what it is not for, and what their role becomes after it ships? AI readiness is rarely a technology problem. It is almost always a change-management problem in disguise, which is why most AI projects fail in their first year on the people side rather than the model side.
• Governance readiness. Do you have a clear answer for who is accountable when the AI makes a decision the customer or the regulator does not like? "We will figure it out" is not a governance posture — it is a future incident report.

If even one of data, process, people, or governance fails an honest readiness check, the AI project will hit a ceiling that no model upgrade can break through.

Red flags: signs your business is not AI-ready, no matter what the vendor demo shows

The AI readiness conversation often happens after a great vendor demo, which is exactly the wrong order. A demo proves the model can do the task on clean inputs in a controlled environment. It proves nothing about your environment. The red flags below come up in roughly every other readiness audit, and any one of them is enough to slow the project to a crawl.

Your business-critical data lives in spreadsheets that get emailed around. If the customer ledger, the price list, or the inventory record exists primarily as pricing_v47_FINAL_USE_THIS.xlsx, the AI does not have a system of record to work against. It has a snapshot, possibly out of date, possibly different from what your colleague is using right now.

You cannot answer "what is the source of truth for X" without a meeting. If three people have to confer to decide whether the CRM or the billing system or the spreadsheet has the right customer status, you have a master-data problem, and master data is the single biggest blocker to enterprise AI readiness. The model will eagerly pick a side; you will not like which one.

Your last process documentation update was over two years ago. Either the processes have not changed in two years (improbable) or the documentation is wrong. Both fail the readiness check, because the AI will encode whatever process you point it at, and pointing it at fiction produces fictional outputs at scale.

Nobody on the leadership team can name three specific places AI would generate measurable value this quarter. If the answer to "where should we apply AI?" is "everywhere" or "I do not know yet, that is what the consultant is for," your business is not ready. The companies that succeed have already identified the constrained, high-cost workflow they want to compress — they just do not know how yet. That is solvable. "Everywhere" is not.

If any of these red flags is present, fix it before signing the vendor contract — not after, when you have a sunk cost and a project review at the end of the quarter.

The AI readiness checklist: 12 questions to answer before you green-light a project

This is the actual checklist. Walk through it with the executive sponsor, the data lead, and the process owner — in one ninety-minute meeting, not three. Each item is binary. Score honestly. If you cannot give a confident "yes" to at least nine of twelve, the project does not start; the readiness work does.

• 1. Is the target workflow named and bounded? One specific process, with a start, an end, a volume, and a cost.
• 2. Does the workflow have a documented current state? A diagram, a SOP, or a Loom video that someone outside the team could follow without a tutor.
• 3. Is the relevant data accessible via API or a managed export? Not "we can pull a CSV with a week of notice" — programmatically retrievable, on demand.
• 4. Is the data cleaned and consistently structured? Same field names, same units, same identifiers across systems.
• 5. Is there a single agreed-on source of truth for each entity? One canonical customer record, one canonical product record.
• 6. Is there an executive sponsor with authority over the affected team? Not a champion. A sponsor — someone who can reassign headcount.
• 7. Is the success metric quantified, with a baseline? "Reduce average handle time from 6.8 minutes to under 4." If you cannot write it that way, you are not ready to measure it.
• 8. Is the ROI math defensible at the project level? Costs, expected savings, payback period, all written down — and stress-tested for the case where the model performs at 80% of the pitched accuracy.
• 9. Is the human review and escalation path designed before the system ships? Who reviews what, on what cadence, and when does the system get pulled?
• 10. Are the governance, security, and compliance reviewers in the room? Not informed late — involved in the design.
• 11. Has the affected team been told what is being built and why? The biggest single predictor of failure is the front-line team finding out about the project from a Loom video forwarded by a peer.
• 12. Is there a kill criterion? A clear "we shut this down if X" — without one, the project cannot fail formally, which is worse than failing.

Twelve out of twelve is rare. Ten or eleven is workable — the missing items become explicit pre-work with an owner and a date. Below eight, the conversation is about readiness, not about implementation, and confusing the two is how the AI budget gets burned in the first quarter.

The AI is the easy part. The readiness work that comes before the AI is what decides whether the project ships, scales, or quietly dies in committee.

If the checklist score is below nine, the next phase is readiness work — not a vendor RFP, not a model selection, not a pilot.

Where to start when the answer is not yet

"Not yet" is the most common honest answer to AI readiness for a business, and it is also the answer most leaders refuse to give because it sounds like falling behind. It is not. According to Gartner's research on AI strategy, organizations that rush the readiness phase consistently overspend in year one and under-deliver against the same KPIs by year two. There is a sequence that works, and it starts well below the model.

Start with one workflow, not with AI. Pick the single highest-friction process you would automate, and document it. Map the inputs, the decisions, the outputs, the exception cases, the human handoffs. Do this on a whiteboard, then in a document. You will find inconsistencies, undocumented edge cases, and disagreements about how the process actually runs today. Those discoveries are the most valuable output of week one — they would have torpedoed the AI project on contact a quarter later.

Fix the data backbone for that one workflow. You do not need an enterprise data platform to start. You need the data the chosen workflow touches to live in one place, with one schema, accessible by API. Do this for one workflow and you have a template; trying to do it for the whole company first is how readiness projects become twenty-four-month consulting engagements with no software shipped.

Run one constrained pilot with a real success metric. Pick a target, write down the baseline, set a kill criterion, ship. If it works, the readiness team that built it is now the team that runs the next ten — and crucially, you have evidence the rest of the business cannot argue with, which is the political fuel for everything that follows. The same logic underpins our automations playbook: small, bounded, instrumented bets compound; sprawling all-at-once programs collapse under their own weight.

For teams that already have a clear use case in mind — support deflection, document review, knowledge retention — the readiness work is narrower and faster than people fear. We have seen this most clearly in knowledge-management programs that target tenure-driven attrition: companies that thought they were a year out from AI turned out to be six weeks out, once the readiness gaps were named explicitly instead of waved at.

"Not yet" is the right answer for most businesses today — and the readiness work it triggers is what makes the AI investment that comes next actually compound.

The readiness work that pays for itself

The under-appreciated truth about an AI readiness program is that the readiness work itself has positive ROI before any AI ships. Cleaning the customer data improves the existing reporting. Documenting the workflow lets you onboard faster. Establishing the governance posture closes audit findings the company has been carrying for years. Naming the single source of truth eliminates 3+ recurring meetings a week. These are not consolation prizes for a project that has not launched yet — they are the project, and they fund the project that comes next.

Treat the readiness work as a deliverable, not as overhead — it pays for itself before the first model goes live, and it pre-qualifies every project that follows.

For a leader looking at an AI budget, an internal evangelist, and a row of vendors lined up to demo, the most useful thirty minutes you can spend this quarter is running this checklist with your sponsor and your data lead. If you score above ten, start scoping the pilot. If you score below nine, that is good information — it tells you exactly where the readiness work needs to start, and it saves you from buying a tool that you cannot yet feed. For an operations-heavy business or any organization where the workflow surface area is larger than the data surface area, the readiness work is the project. The AI is just what you turn on when the substrate is ready to support it.