# AI readiness checklist

> A business is ready for an AI project when five things are true: the process is written down with a clear definition of done, the data the AI needs is reachable and consistent enough to trust, the systems it must touch have a way in, someone owns the outcome and will review the work, and there are rules for what the AI may do alone. Missing any one of these is fixable, but it has to be fixed first. The best first project is the one where all five are already close to true.

Source: https://risendigital.io/guides/ai-readiness-checklist  
Publisher: Risen Digital  
Updated: 2026-09-21

## At a glance

- **Five conditions:** A documented process, reachable and consistent data, accessible systems, an accountable owner, and rules for autonomy.
- **Most common gap:** The process exists only in the heads of the people who run it, so nobody can say what a correct result looks like.
- **Cheapest fix:** Write the process down as it really happens, exceptions included, in an afternoon with the person who does it.
- **Best first project:** High volume, low stakes, well documented, with data in one place and a person who wants it to work.

Most AI projects that fail were never going to work, and the reason was visible before anyone wrote a prompt. The process was in someone's head, the data was in six places, nobody could grant the system access, or nobody owned the outcome. This checklist puts those conditions in front of you now, shows how to fix the gaps, and helps you choose a first project that can succeed.

![A team working through a plan at a whiteboard](https://risendigital.io/images/experience/whiteboard-office-1600.jpg)

## What has to be true before an AI project works?

The AI has to know what a good result looks like, be able to reach the information, be able to act in your systems, have someone checking it, and know its limits. Those are process, data, systems, people and governance.

A language model is good at reasoning over what it is given. It is not good at guessing what you meant, finding data you did not connect, or deciding what it is allowed to do. Readiness is about removing those guesses before the project starts. Each condition is ordinary operational hygiene, and each one is also the thing that lets an AI system produce a result you can trust.

Process readiness means the work can be described: inputs, steps, decisions, exceptions and what done looks like. Data readiness means the records the AI needs exist, are reachable through something better than a screenshot, and agree with each other often enough to rely on. Systems readiness means the tools it must read from and write to have an interface, and someone can grant access to it. People readiness means a named owner, reviewers who will actually look at the output, and a team that has been told what is changing. Governance readiness means written rules about what the AI may do alone, what needs approval, and what it must never touch.

> **Readiness is per project, not per company** A business can be ready for an AI assistant in support and nowhere near ready for one in finance. Run the checklist for the specific process you have in mind.

## How ready is the process you have in mind?

Work through the five groups for one specific process. Every unchecked item is a task to do before the project, not a reason to abandon it.

_Interactive on the web page: Check the five conditions for your process. The checklist covers five groups. Process: the work is documented, done is defined, exceptions are listed and volume is known. Data: the records exist in a known place, can be reached through an interface, are consistent enough to trust, and have an owner. Systems: the tools involved have an API or export, access can be granted, and a test environment exists. People: a named owner, reviewers with time, and a team that knows what is changing. Governance: written rules for what the AI may do alone, an approval path, a way to stop it, and a record of what it did. A process with most items ticked is a good first project; one with gaps in every group needs preparation before any AI work starts._

## How do you fix the gaps the checklist finds?

Most gaps close with a short piece of ordinary work: a documented process, a data clean-up in one system, an API key, a named owner, a one-page rule. None of them needs AI to fix.

| Gap | What it looks like | The fix |
| --- | --- | --- |
| Undocumented process | Only one person really knows how it works | Sit with that person for an afternoon and write it down as it happens, exceptions included |
| Scattered data | The answer is in an inbox, a spreadsheet and a CRM, and they disagree | Pick one system of record for this process and move the rest into it before connecting anything |
| No way in | The tool has no API and the vendor will not add one | Use a scheduled export, or start with the read side only, or reconsider the tool |
| No owner | Everyone is interested and nobody is accountable | Name one person whose job includes the outcome and give them the decisions |
| No rules | Nobody has said what the AI may do on its own | Write a one-page rule: alone, with approval, never. Start narrow and widen on evidence |

The order matters. Fix the process first, because it tells you which data and systems actually matter. Then fix data and access for that process only, not for the whole company. Name the owner before the build starts, and write the rules before the AI touches a live system. Doing this for one process makes the second project far easier, because the habits and the plumbing carry over.

## How do you pick a first AI project?

Choose the process where the checklist is already mostly true, the volume is high enough to matter, a mistake is recoverable, and the person who runs it wants help.

1. **List the candidates** Ask each team for the repetitive work they would gladly give up. Inbox triage, data entry between systems, first-draft replies, report preparation and scheduling usually appear.
2. **Run the checklist on each** Score them on the five groups. The one with the fewest gaps goes to the top, even if it is not the most exciting.
3. **Check the stakes** A mistake should be catchable and reversible: a draft rejected, a record corrected, a message held. Anything irreversible or customer-facing waits for the second project.
4. **Check the volume** Enough cases per week that saving minutes on each adds up, and enough that the AI's track record builds within weeks rather than months.
5. **Check the appetite** The person who does the work today should want this. If they are worried, involve them in defining done and reviewing output; if they are opposed, pick another process.
6. **Start read-only and draft-only** The AI looks things up and proposes; people approve. Widen its permissions as the accepted-unchanged rate earns it, and keep the audit trail from day one.

## Frequently asked questions

**Q: Do we need clean data before starting an AI project?**  
A: You need data that is clean enough for the one process you are starting with, which is a much smaller job than cleaning everything. Pick the system of record for that process, fix the fields the AI will rely on, and leave the rest. The AI will surface further problems as it runs, which is useful.

**Q: What if our process is not written down anywhere?**  
A: Write it down before anything else, with the person who does the work, as it really happens. Include the exceptions and what they do about each. This usually takes an afternoon and is the single most valuable step in the whole project, because it becomes the AI's instructions and the reviewer's definition of done.

**Q: Do we need a data team or an AI specialist on staff?**  
A: Not for a first project. You need a named owner who understands the process and can make decisions, someone who can grant system access, and reviewers with time. Technical work can come from a partner, but ownership of the outcome has to sit inside the business.

**Q: How do we know if a process is too risky for AI right now?**  
A: Ask whether a wrong action can be caught and undone before it causes harm. If the answer is no, such as an irreversible payment or a legal commitment, keep a person in the loop for the action and let the AI only prepare and recommend. Risk is managed by where you put the approval, not by avoiding AI entirely.

**Q: Should the first project be customer-facing or internal?**  
A: Internal, almost always. Internal users forgive early mistakes, give direct feedback and let you build the review habits before customers see anything. A customer-facing assistant is a fine second project once the plumbing and the rules exist.

## Related pages

- [What is an AI agent?](https://risendigital.io/guides/what-is-an-ai-agent)
- [90-day AI implementation roadmap](https://risendigital.io/guides/90-day-ai-implementation-roadmap)
- [AI data privacy and security questions](https://risendigital.io/guides/ai-data-privacy-and-security-questions)
- [How to integrate AI into existing software](https://risendigital.io/guides/how-to-integrate-ai-into-existing-software)