# AI readiness: the five dimensions to review before funding a project

> Review data, technology, strategy, people and culture, and governance together. The output is a picture of gaps, risks and accelerants, not a score.

URL: https://preview2.snowfox-ai.com/blog/ai-readiness-five-dimensions
Last updated: 2026-11-10

Published: 2026-11-10

By the Snowfox AI team

## Key takeaways

- AI readiness is reviewed across five dimensions together: data, technology, strategy, people and culture, and governance and processes.
- A single maturity score hides the one gap that blocks a project, so the output should be a picture of gaps, risks and accelerants.
- Review readiness per candidate use case, not for the whole company in the abstract.
- The review sequences the work: start where the outcome is clear and the blocking gaps are few, and let named use cases pay for foundation work.

Before funding an AI project, review five dimensions together: data, technology, strategy, people and culture, and governance and processes. The useful output is not a maturity score. It is a picture of the gaps that will block the work, the risks that need an owner, and the accelerants you already have, and that picture tells you which use cases to do first.

## What is AI readiness, and why review it before funding?

AI readiness is an organization's practical ability to take a use case from an idea to a system that runs in production and changes a business result. A use case is one specific application, such as forecasting demand for a product line or extracting the parties from a contract.

Projects that stall rarely stall on the model alone. The data could not be joined, the output had nowhere to land in the existing systems, nobody on the business side owned the result, or the approval to go live was never defined. All of these can be found before the money is committed, when fixing them costs less.

AI also does not only mean generative AI. The answer may be a machine learning model, workflow automation, a system integration, analytics or an AI agent, and often several together. Start with the problem, not the technology.

## What are the five dimensions?

| Dimension | What it reviews | A typical gap |
| --- | --- | --- |
| Data | What exists, where it lives, whether it can be trusted | Key fields sit in spreadsheets or in a system nobody can export from |
| Technology | Systems in use, how they connect, what they cannot do | The model output has no route back into the system where people work |
| Strategy | Which business outcomes the work is meant to move | The project is defined by a technology, not by a metric |
| People and culture | Skills, ownership, and appetite for changing how work is done | No named owner on the business side |
| Governance and processes | Decision rights, risk, and how changes get approved | Nobody knows who can approve a model going live |

## What should you look at in the data?

Three things: what exists, where it lives, and whether it can be trusted.

For each candidate use case, list the sources it needs. Check how much history there is and at what level of detail. If the use case needs supervised learning, meaning a model that learns from past examples with known answers, check that those answers were recorded. Then check whether two systems give the same number for the same question.

Typical gaps: customer or product identifiers that do not match across systems, history that is overwritten instead of kept, and definitions that differ between departments.

What not to do: audit all the data in the company. Review the data that the candidate use cases depend on.

## What should you check in the technology?

Look at the systems in use, how they connect, and what they cannot do.

Map where the output has to land: the ERP, the CRM, a pricing tool, a service desk screen. Check whether those systems expose an API (an application programming interface, the way one system reads from or writes to another) or only allow manual exports. Check where a model would run and who would operate it. MLOps is the set of practices for putting machine learning models into production, monitoring them and improving them over time, and it needs an owner before the first model ships.

What not to do: buy a platform before you know what the first two use cases need from it.

## How do you test the strategy?

Ask which business outcome the work is meant to move. Each candidate use case needs a named metric, a baseline, and a person who answers for that metric.

A simple test: if this works perfectly, which decision changes, and who makes it differently? If nobody can answer, it is a demo, not a project.

Decide how the result will be measured before building. In Snowfox's retail price optimization work, the published case reports a 6% margin increase compared with matched comparison stores. The comparison group is what makes the number meaningful.

## What does readiness mean for people and culture?

It covers skills, ownership, and appetite for changing how work is done.

Identify who will build the system, who will operate it, and who will use it. A model that recommends prices has no effect if category managers do not trust it or are measured on something it does not improve.

A skills gap is not a blocker in itself. It decides how you staff the work: build in-house, run models with a partner under an MLOps arrangement, or use pre-built models as a service (MLaaS).

## What belongs under governance and processes?

Decision rights, risk, and how changes get approved.

Find out who approves a model going live, what happens when it is wrong, who can turn it off, and how a new version is tested. Check which rules apply to the data involved: security, retention and privacy.

There are two typical gaps. Either no process exists, so the launch is stopped late by legal or security. Or the process was designed for conventional software releases and cannot handle a model that is retrained on new data.

What not to do: write an AI policy in the abstract. Define the rules for the first use case, then generalize from there.

## Why should the output be a picture and not a score?

A score averages things that do not average. A strong result in technology and a weak one in data does not make you half ready. It makes you blocked on data.

The output should list three things per use case:

- **Gaps**: what is missing for this use case to run in production.
- **Risks**: what could go wrong, and who owns it.
- **Accelerants**: what you already have that shortens the path, such as a clean transaction history, a team that already decides from dashboards, or a warehouse that is already in place.

The dimensions also interact. Two departments disagreeing on a number looks like a data problem, but the cause is often governance: nobody owns the definition. That is why the five are reviewed together.

## How does the review sequence use cases?

1. List the candidate use cases with the business outcome each is meant to move.
2. For each one, mark which gaps block it and which only slow it down.
3. Start with the use case that has a clear outcome and the fewest blocking gaps, even if it is not the largest prize.
4. Schedule the foundation work that unblocks the next ones, such as a pipeline, an integration or an approval rule, and attach each item to a named use case.
5. Set aside use cases with no owner or no measurable outcome until they have both.

Sometimes the result is that the first project is a data project, not an AI project. That is a valid outcome. Data and AI are two halves of the same job.

## Where to start

- The [AI Readiness Assessment](/ai-readiness) reviews the five dimensions together and returns a picture of gaps, risks and accelerants.
- [AI consulting](/services/ai-consulting) covers strategy, use case prioritization and implementation.
- If the main gaps are in data, see [data strategy](/services/data-strategy).
- Or [tell us the problem](/contact) and we will work out whether data or AI is the right approach.

## Common questions

### What are the five dimensions of AI readiness?

Data, technology, strategy, people and culture, and governance and processes. They are reviewed together because a gap in one usually shows up as a symptom in another.

### Should an AI readiness assessment produce a score?

No. A score averages dimensions that do not average, so a strong technology rating can hide a data gap that blocks the project. A list of gaps, risks and accelerants per use case is more useful for a funding decision.

### What if the review shows we are not ready for AI?

That is a useful result. It usually means the first project is a data or integration project tied to a specific use case, and the review tells you which one.
