AI readiness: data, processes, people
Readiness for AI is not about technology. It is whether your data is reachable, your processes are written down, and your people know the rules.
The short answer
Whether an AI project succeeds is decided before anyone picks a model, by three things: is the data the project needs reachable and reasonably clean, is the process it touches written down and agreed, and do the people involved know what is expected of them and of the system. Technology is rarely the constraint. Those three are, and they can be assessed in an afternoon.
Readiness is per project. A company can be entirely ready for email triage and entirely unready for sales forecasting on the same day.
The three dimensions
| Dimension | Question | Ready looks like | Not ready looks like |
|---|---|---|---|
| Data | Can the system get what it needs, and can it trust it? | Data in known systems, accessible through exports or APIs, with obvious errors fixed | Data in personal spreadsheets and inboxes, duplicates, fields used inconsistently |
| Process | Is the process written down and agreed? | Steps, exceptions and decision points on a page, the same in three people’s telling | ”It depends”, three versions, exceptions nobody can list |
| People | Do the people know the rules and their role? | Clear rules on tools and data, involvement in the design, a person who decides | Tools used on personal accounts, fear of replacement, nobody accountable |
The assessment, in an afternoon
- List the candidate projects. Three to five, from wherever the repetitive work is.
- For each, name the data it needs. Where does it live, who can access it, how clean is it? Score high, medium or low.
- For each, check the process. Can it be written on one page, with exceptions? Do the people who do it agree with the page? Score it.
- For each, check the people. Do they know the rules? Were they involved? Is there an owner? Score it.
- Read the table. The project with three highs is your first. Projects with a low on data need a data step first; a low on process needs a mapping session; a low on people needs a conversation, not a tool.
What usually comes out
- Data scores lower than expected. Businesses discover that their most important information lives in inboxes and spreadsheets. That discovery is itself valuable.
- Process scores vary wildly. Some processes are crisp; others turn out to be three people’s habits. Mapping the second kind is worth doing whether or not AI follows.
- People score higher than feared. Staff who do repetitive work are usually keen to lose the repetitive part, provided the decisions stay theirs and the rules are clear.
What this means for you
Run the assessment on your candidate projects. Pick the one with three highs and build it. Use the lows on the others as a to-do list: which data to bring into a known system, which process to write down, which conversation to have. Readiness is not a state you reach; it is a table you keep updating as projects clear the way for the next ones.
Frequently asked questions
Our data is a mess. Does that rule out AI?
It rules out projects that depend on the messy data, for now. It does not rule out projects that work on text arriving fresh, such as email triage or document extraction, which need almost no historical data. Start there, and let the clean data those projects produce become the foundation for the next ones.
Do we need a data team or a data warehouse first?
For a small business, no. You need to know where each kind of data lives, who can access it, and how clean it is. A spreadsheet answering those three questions for your ten most important data sources is the readiness work. Warehouses come later, if ever.
How do we prepare our people?
Involve the people who do the work in mapping the process and choosing the first project. Give them the one-page rules on what may and may not go into AI tools. Show them the system prepares and they decide. Preparation is participation and clarity, not a training course.