# AI readiness checklist

> Most failed AI projects fail before a model is chosen. The process was unclear, the data was scattered or nobody owned the result. Score these before you spend on a build.

- Page: https://liftup.tech/checklists/ai-readiness-checklist/
- Type: Checklist
- Updated: 2026-11-18

## In short

A business is ready for an AI project when one process is clearly defined, its data can be found and read by software, a named person owns the outcome, and the cost of a wrong answer is known. Score the 30 questions below from 0 to 2 each. Fifty or more means start. Under twenty means fix the basics.

## How the score works

Each question gets 0 (no), 1 (partly) or 2 (yes). There are six groups of five questions, so the maximum is 60. Score honestly. A low score is useful information, not a failure.

| Total | Reading | What we would do next |
|---|---|---|
| 50 to 60 | Ready | Pick the first use case and run a four to six week pilot. |
| 35 to 49 | Nearly ready | Fix the groups scoring under 7 of 10, then pilot. |
| 20 to 34 | Early | Document the process and clean the data first. A consulting call may be enough. |
| Under 20 | Not yet | Buy off-the-shelf tools for now and revisit in six months. |

Download the checklist as a file and let two or three people score it separately. Differences in their scores show where the real disagreements are.

## Group 1: the process

- Can you write the process as numbered steps on one page?
- Does it repeat at least weekly?
- Can you say what a correct result looks like?
- Do you know how long it takes a person today?
- Is there a clear start trigger, such as an email arriving or a form sent?

> If you scored under 6 here, stop. AI automates a process you can describe. It does not discover one.

## Group 2: the data

- Is the information the process needs stored in digital form?
- Is it in a few systems, not forty spreadsheets?
- Is it mostly current, with a known owner for updates?
- Can software read it through an export or an API?
- Is personal or confidential data tagged so you know what must not leave your control?

A retrieval chatbot over messy documents answers messily. We often spend the first week of a project on this group alone. See [RAG versus fine-tuning](https://liftup.tech/compare/rag-vs-fine-tuning/) for how the data shapes the technical choice.

## Groups 3 to 6 in short

| Group | Five questions to ask yourself |
|---|---|
| People | Is there one named owner? Will the people doing the work help test it? Is the team told what changes for them? Is there time in their week for review? Has anyone senior said yes to the pilot? |
| Risk | What does a wrong answer cost? Is there a human check on high-cost steps? Do contracts or regulation limit where data can go? Do you know your retention rules? Can you explain the output to a customer if asked? |
| Systems | Do your tools have APIs or webhooks? Is there a test environment? Can you add a service account? Is single sign-on in place? Is there somewhere to run the workflow, such as a cloud account? |
| Measures | Do you have a baseline number for time, cost or errors? Can you measure it after launch? Is there a target? Is there a review date? Do you know what would make you stop? |

## A worked scoring example

A 12-person accounting firm wants to automate client document intake. The process is clear (10 of 10), the documents arrive by email in PDF form (8 of 10 for data, since filenames are inconsistent), one partner owns it (9 of 10 for people), the risk is moderate because client financial data is involved (6 of 10), systems have an API (8 of 10) and they know how long intake takes (7 of 10). The total is 48. The reading is nearly ready. The group to fix is risk: a data processing agreement with any AI vendor and a rule that a person checks every extracted figure. After that, a pilot makes sense. The firm may also find the [AI document processing](https://liftup.tech/services/ai-document-processing/) page useful.

This is an illustration, not a client result.

## Questions, answered

### What if we score low but want to start anyway?

Start small and use tools that already exist. A shared inbox rule or a template can fix a process problem cheaper than a model can.

### Do we need a data scientist before a pilot?

Usually not. Most business pilots use hosted models through an API, so the work is process design and integration, not model training.

### How long does an assessment take?

Scoring takes about an hour with the right people. Our consulting engagements add a review of your systems and a ranked list of use cases.

### Does readiness include legal review?

The risk group flags the questions. Ask your lawyer about data transfers, retention and any sector rules before a pilot touches real customer data.

### Can the score change?

Yes. Re-score every quarter. Teams often move from 30 to 45 just by documenting the process and naming an owner.

## Related

- [AI consulting services: decide what to build, what to buy and what to skip](https://liftup.tech/services/ai-consulting/)
- [AI use policy template](https://liftup.tech/checklists/ai-use-policy-template/)
- [AI automation ROI calculator](https://liftup.tech/tools/ai-automation-roi-calculator/)

---

Get a fixed quote: https://liftup.tech/contact/ · email rohit@liftup.tech · WhatsApp +91 77689 10234 (https://wa.me/917768910234) · reply within one working day.
Source page: https://liftup.tech/checklists/ai-readiness-checklist/ (Liftup Technologies)
