AI readiness audit: how to choose your first AI workflow
A suitable first AI workflow combines a recurring problem, usable data, a team willing to adopt it and manageable risk. This guide shows owners and operations leads at SMEs how an AI readiness audit finds that workflow, and how to run a first pass themselves.
Companies often start AI adoption with a tool trial. A few staff members test a chatbot, someone drafts a policy, and a department asks for automation. That energy is useful, but it does not tell the business where to invest first. A readiness audit turns the scattered ideas into a ranked implementation plan.
The audit identifies where AI can reduce repetitive work, delays or rework without creating avoidable risk. It ends with one workflow chosen as the first implementation and a list of what has to be ready before anyone builds it.
What an AI readiness audit examines
An audit looks at work, one workflow at a time. In Block Consult's AI readiness and workflow audit we start from recent cases, such as an enquiry, an invoice, a support request or a reporting cycle. The people involved show us the steps, source documents, decisions and handoffs, and we record where work waits, where information is copied and where exceptions come back for another check.
For each candidate workflow we then examine six areas: the data, the system connections, the consistency of the process, the people, the governance and the economics. The results are a workflow map, a prioritised roadmap, a data and risk checklist and a brief for the first build. Every estimate is recorded with its assumptions and with the evidence that is still missing.
The sections below describe a first pass you can run yourself. It follows the same logic on a smaller scale.
Start with workflows
Build the list of candidates from tasks, before anyone names a product. Look for places where people copy information between systems, rewrite similar documents, search for internal knowledge, classify incoming requests or wait for someone to produce a summary. Ask each team for examples from last week. Recent cases make better candidates than a brainstorm about what AI might do.
For each candidate, note five things.
- Frequency: how often the task happens.
- Time loss: how much effort goes into repetition that needs little judgement.
- Data availability: whether useful inputs already exist in a usable form.
- Review path: who can check outputs before they affect customers or money.
- Risk level: the privacy, compliance, brand and operational consequences if the tool is wrong.
If frequency and time loss are guesses, measure them first. A workflow audit follows single cases from start to finish and puts numbers on work, waiting and rework. Those numbers are the baseline for the scoring.
Score the use cases
A simple scorecard is enough for the first pass. Rate each candidate by business value, implementation complexity, data readiness, change effort and risk. A scale of low, medium and high is fine. Write one line of reasoning beside each rating, so that a colleague who was not in the room can challenge it.
Then place the candidates on a grid with value on one axis and effort on the other. Complexity, data gaps and change effort all count as effort. Use risk as a filter: a workflow involving sensitive personal data or financial approvals can stay on the list, but it is a poor first project. The first project should score high on value and data readiness while staying moderate on complexity and risk.
Value needs a figure behind it, even a rough one. Hours per month and the cost of an error are enough to compare candidates, and our guide to calculating the return on an AI workflow sets out the method.
A sentence that passes the test names the trigger, the output and the reviewer. "When a quote request arrives, draft a reply from the current price list for the sales lead to approve" passes. "Use AI in sales" does not.
Separate configuration from custom development
Some opportunities need only configuration: a prompt library, a document workflow or an AI feature inside an existing tool. Others require a custom internal app because the workflow depends on specific data, permissions or review logic. Mixing those categories leads to weak estimates and slow decisions.
Our audit sorts its recommendations into three groups: process changes, better use of software you already have, and work that needs an integration or a custom tool. Label every shortlisted candidate with its group before estimating it, because each group has a different budget and a different owner. For the third group, the articles on connecting AI to the software your team uses and on buying or building AI tools cover the next decisions.
Make adoption part of readiness
A workflow is ready for AI when the team has a clear reason to use it, managers know what good output looks like and someone can review the results. Check all three before choosing.
Ask the people who do the work whether this is a task they want help with. Ask their manager for a few examples of good output; these later become the acceptance criteria. Name the reviewer and confirm they have time for the check, because a review that nobody has time for gets skipped. Our article on designing human review covers that step.
The usual failure is a candidate that scores well on paper and has no owner. If nobody can say who will answer questions about the tool after launch, move it down the list.
Choose the first workflow and write the brief
Pick one workflow and give it an owner, an identifiable result and a way to test whether the change helps. Write the decision down as a brief: the recommendation, a budget range, the dependencies and the acceptance criteria.
List what has to be resolved before a build, such as a source document nobody maintains, a missing interface or an approval rule that differs between teams. Then set priorities for the next 30, 60 and 90 days. The brief should be complete enough for your own team, Block Consult or another partner to build from, since implementation is a separate decision.
Frequently asked questions
- How long does an AI readiness audit take?
- For most companies, a focused audit takes one to three weeks, depending on the number of departments and systems involved.
- What is the best first AI use case for an SME?
- Usually a repetitive workflow with clear inputs, available data, low regulatory risk and an obvious human review step.
- Do we need clean data before starting?
- You need enough reliable data for the specific workflow. A perfect company-wide data platform is not required. The audit should identify what is ready now and what needs cleanup first.
- What do we get at the end of an AI readiness audit?
- With Block Consult, a workflow map, a prioritised roadmap, a data and risk checklist and a brief for the first build. Implementation is a separate decision: we can deliver it, your team can take it on, or you can use the brief with another partner.
Related reading
Let's build something durable.
Tell us about your project. We reply within two business days.

