AI Readiness Assessment: What to Prepare Before Your First AI Project

Updated October 2026

In short

For a first project, AI readiness comes down to one process in the product you already run. You do not need a score for the whole company. You need to know whether that process has data you can send, a person who will accept the result, and a place in your product where the answer can land.

Most of it takes days of work on your side. The slowest item is usually access. Below are the seven checks, who does each one, how long it takes, and how to read the result. At the end is a comparison of free questionnaires, paid consultants, and doing it in-house.

Defining the scope: what an AI readiness assessment actually checks

A readiness assessment from a large vendor scores your organization on five or six pillars. For a first project you need a narrower check: whether one process has the data, the owner, and the place in your product for AI to do its job.

The public tools are built for the first kind. Microsoft’s readiness wizard asks ten questions across five areas: business strategy, technology and data, AI experience, culture, and governance. Cisco’s readiness index uses six pillars and returns a score from 0 to 100 with four levels.

Both are useful for a board slide. Neither tells you whether the support queue in your product can be handed to a model next month. That question has a yes or no answer. That is the AI readiness worth measuring.

Why it matters: few companies are fully prepared, and that is normal

Only about one large company in eight makes the top readiness group, and the gap shows up in plain things, such as agreed use cases and data kept in one place.

The 2025 Cisco readiness index surveyed 8,000 executives at companies with 500 or more employees. The top group has stayed at about 13% for three years. 64% of respondents said they struggle to centralize data. 18% had finalized their use cases, against 77% in the top group, and 35% had a change-management plan, against 91%. Readiness shows up in results: the most ready companies were four times more likely to move pilots into production.

Data leaders are cautious too. In an IBM study of 1,700 chief data officers, 26% were confident their data could support new revenue from AI.

Gartner adds a forecast. It predicted in February 2025 that through 2026 organizations would abandon 60% of AI projects that lack data prepared for them. A gap in AI readiness surfaces when the pilot needs data nobody can send yet. The checks below find that gap first.

Running it on your own product: seven checks, each with an owner and a time frame

Seven checks, most of them a few days each. In our projects, the slowest has been getting access to the data.

You can run this AI readiness self-assessment without outside help. Each check names who does the work, so you know whom to ask. The example throughout is a support queue.

1. One process and today’s number

Pick one process and write down how it performs today. For a support queue, that is minutes per ticket, error rate, or tickets per day. The process owner can usually do this in a day or two. Without a “before” number there is nothing to compare the pilot against, and the ROI conversation gets stuck. IBM found that only 25% of AI initiatives delivered the expected return, and you cannot answer that question without a baseline.

2. An export you are allowed to send

You need several months of the process as a file, plus permission to share it. The owner is whoever runs the system the data lives in. We cover the details in our guide to AI agent development. Plan for the gap: in our projects, an export promised in a week has taken six. Ownership is a common blind spot: dbt Labs found that 41% of organizations have no defined owner for their data.

3. A test set of a hundred answered cases

Take a hundred real cases and write down the correct answer for each. For tickets, that is the right category or the right reply. Someone who works the queue daily needs two or three days for it. This set becomes your definition of “done correctly,” the one your team and any vendor will check against later.

4. The place in your product where the answer lands

Integration blocks pilots as often as data does: 43% each in Informatica’s 2025 survey of chief data officers. So decide where the AI output goes: an API response, an event in a queue, a field in a form, or a draft reply an agent approves. Your engineer owns this, and it usually takes one design session. Decide also what happens when the model is wrong. With a draft, a person fixes the answer before any customer sees it.

5. A sandbox and a security review

A vendor will need a test environment and an account with limited rights. Your security lead owns this check. If personal data from the EU is involved, the GDPR requires a contract with any processor that handles it, so the data processing agreement comes before the data moves. Decide early whether the vendor works on an anonymized copy or on live data, because that choice shapes the review. Security reviews have their own queue, so start the request in the first week.

6. One person who accepts the result

You need one person who looks at an answer and says it is correct. It may not be the budget holder. Naming this person takes one conversation, and if two people both claim the role, settle it before the pilot starts. Our guide to implementing AI explains why projects without this person get stuck.

7. A running budget next to the build budget

AI costs money every month after launch, including model calls and someone watching quality. The budget holder owns this, and the number should be agreed before the build starts. Without it, each fix becomes a separate purchase: IBM’s CEO study found that 50% of CEOs ended up with disconnected, piecemeal technology after investing fast.

Reading the result: go, fix first, or not yet

If the process, the test set, and the owner are in place, start a pilot. If only access or integration is missing, start anyway and plan around the wait. If nobody owns the result, do not start yet.

For a first project, an AI adoption readiness assessment can end in one of these three outcomes instead of a score. Data problems are common. In Informatica’s 2026 survey, 57% of data leaders named data reliability a top barrier. Harvard Business Review Analytic Services and Cloudera found that 7% of enterprises consider their data fully ready for AI, while 73% find preparing it difficult.

Few companies have data that is fully ready. A missing owner is the one gap a pilot cannot work around.

A missing owner is different from missing data. If the full export is stuck in a queue, the pilot can start on a smaller sample and switch over later. A result nobody accepts never leaves the pilot.

Doing it yourself, with a free tool, or with a consultant

Free questionnaires are short and show where the company stands. They make a good opening for a talk with leadership. Consultants charge for a full assessment: Advisor Labs, for example, quotes $18,000–35,000. For one process, the seven checks above, done in-house, cover most of what you need.

A consultant is the right tool when you have dozens of processes and need a map for the board. For a first project inside a working product, you may need less than a full report.

Our approach is narrower. A short free review call decides whether this process needs AI at all, and then we check readiness during the pilot: four to six weeks on one process and on your data. Projects start at $30,000. If the process does not need AI, we say so.

Wrapping up: what to bring to the first conversation

Bring one process, today’s number for it, a sample export, and the name of the person who will accept the result. Everything else can be sorted out during the pilot.

• The process. One, described the way it runs today.
• The number. How long it takes or how often it goes wrong now.
• A sample. A slice of the export you are allowed to send.
• The owner. A person with a name. A department cannot sign off.

Start with the export request this week. It is the item most likely to wait.

What to read next

If one process passed the checks, the next step is a pilot. If it did not, the guides below cover what to fix first.

• AI automation services: how we run a pilot on one process.
• How to implement AI in business: where it pays off and when it is too early.
• Why AI pilots fail to reach production.
• AI agent development: stages and timelines.
• AI ROI: how to measure it.