AI readiness assessment checklist: score your data, process, systems, people and risk

An AI readiness assessment scores one workflow on five things: the data it runs on, how well the process is defined, whether your systems can be connected, whether people will own and use it, and what happens if it is wrong. Score each from 1 to 5, add them up, and watch for any single dimension that blocks the rest. The checklist and scoring table below take about an hour with the workflow owner. They tell you whether to build, prepare first or wait.

#Assess a workflow, not the whole company

A company-wide readiness score is too vague to act on. "Our data is average" tells you nothing. "Supplier invoices arrive as PDFs from 300 vendors and our ERP has an API" is something you can build against.

Pick one workflow with steady volume and a clear owner, such as supplier invoice entry, support ticket triage or customer onboarding. Score each workflow separately, then compare to decide where to start.

#The five dimensions

#1. Data

The system can only work with inputs it can read, and it needs examples to be tested on.

• We can export at least 200 recent real items (invoices, tickets, emails, forms).
• The inputs are digital, or scans are legible.
• We know how many formats, languages or layouts exist.
• The fields we need are present in most items.
• A person can say which items were handled correctly, so we have right answers to test against.
• We are allowed to process this data with an AI provider, or in our own environment.

#2. Process

Automation encodes a process. If the process lives in people's heads, it needs writing down first.

• Someone can describe each step and decision in plain language.
• Volume is steady enough to justify building, for example hundreds of items a month and not a handful.
• We know the handling time per item, measured on a sample.
• Rules for common cases are clear. Exceptions are known and have an owner.
• The process has not changed much in the last year, and is not about to.
• We know what an error costs.

#3. Systems

Automations read from and write to your software.

• We can name every system the workflow touches.
• Each has an API, a file import or a database we can use, and not only a screen.
• We can get a sandbox or test environment.
• Credentials can be issued with least-privilege access, and IT can approve them in reasonable time.
• Master data (vendors, customers, items) is clean enough to match against.

#4. People

Systems that nobody owns degrade.

• A named business owner will review results weekly during rollout.
• The people who do the work today are involved and know what changes for them.
• Someone will handle the exception queue.
• We know who will maintain the system after launch, inside the company or through a partner.
• Leadership has agreed what the saved time will be used for.

#5. Risk and controls

The more damage a mistake can do, the more checks the system needs.

• We know which actions are sensitive: payments, deletions, external messages, regulated records.
• We can place human approval in front of those actions.
• An audit trail of inputs, decisions and actions is required, and we know the retention period.
• Legal or compliance has seen the plan where regulations apply.
• We have a fallback if the system is switched off.

#How to run the session

Book 60 to 90 minutes with the workflow owner and one person who does the work every day. Bring a screen share of the actual process and a folder of 20 to 30 real items. Work through the checklist one dimension at a time, and write the evidence beside each answer, such as "ERP exposes a REST API, sandbox requested on 12 October". A score with no evidence is a guess.

Two habits keep the score honest. Score before you discuss, then compare. The person who owns the process and the person who does the work often differ by two points. And when something is unknown, record it as an action with a name and a date instead of assuming the answer is good.

#The scoring table

Score each dimension from 1 to 5, using the descriptions. Be strict. If you hesitate between two numbers, take the lower one.

#What your total means

Add the five scores for a total out of 25. These bands are our rule of thumb for sorting workflows. They aren't a published benchmark.

Override the total with one rule: any dimension scored 1 is a blocker, whatever the sum. A workflow with perfect data and no way to add approval before payments should not be automated yet.

#An example (illustrative)

A finance team scores supplier invoice entry. Data: 4 (clean PDFs, history in the ERP). Process: 3 (documented, exceptions known, handling time not yet measured). Systems: 4 (ERP API and a sandbox). People: 3 (an owner, but no one to maintain it). Risk: 3 (approval before posting identified, audit trail not yet specified). Total 17. The reading: ready for a pilot after measuring handling time, naming a maintainer and defining the audit trail. None of those needs a build. All three fit in a few weeks.

#What to do with the result

1. Write down the lowest dimension for each workflow. That is your preparation work.
2. Rank workflows by score and by value. A high score on a low-value task makes a poor first project. The method in how to calculate the ROI of AI automation will help you estimate value.
3. Measure the baseline by timing a sample of 30 to 50 items.
4. Run an audit or pilot on the best candidate.

An audit turns this checklist into evidence. It times a real sample, tests a prototype on historical items and ranks the opportunities. Our sample audit shows what that output looks like. To budget the next step, read AI automation cost, and for how a project runs, see how we work.

Key point: Readiness belongs to a workflow, and the usual blocker is rarely the model. It is missing sample data, an undocumented process or nobody to own the system after launch.

#Common mistakes when assessing readiness

• Scoring the company. Pick a workflow.
• Rating on hope. If you haven't looked at the data, the data score is 1 until you do.
• Rating people last. Technical teams often do, and it is the dimension that sinks projects first.
• Leaving risk until the end. Decide where approvals go before you design the build.
• Reading a low score as a no. A low score is a to-do list.

If you want a second pair of eyes on your scores, book a strategy call. We will tell you whether a workflow is worth an audit.

#Frequently asked questions

#What is an AI readiness assessment?

It is a structured review of whether a specific workflow, team and set of systems can support an automation or AI project. It covers data, process, systems, people and risk, and ends with a decision: build, prepare first or wait.

#How long does an AI readiness assessment take?

The checklist above takes about an hour per workflow with the workflow owner. A deeper audit that times real samples and tests a prototype takes longer, often one to two weeks.

#Do we need clean data to start?

You need representative data, and it does not have to be perfect. You need enough real examples to test against and a way to know which were handled correctly. Messy inputs raise the review rate, and they do not by themselves rule a project out.

#Who should take part?

The person who owns the workflow, someone who does the work daily, an IT contact for system access, and someone from finance or compliance if money or regulated data is involved.

#What if our score is low?

Use the lowest dimension as your preparation list. Document the process, gather samples, request system access or name an owner, then score again. Most of those fixes need no engineering.