20 questions across 5 dimensions. Answer as your organisation is today - not as you hope it will be.
◎
For each question: Yes = clearly true today. Unsure = partially true or uncertain. No = not yet in place. Hover the ⓘ icon for guidance on what each question is really asking.
Data notice: When you submit, your aggregated scores (totals per dimension) are sent anonymously to the author for research purposes. No personal data, IP address, or identifying information is collected or stored. You do not need to create an account or provide an email address to use this checklist.
PROGRESS - 0 / 20 answered
1
Strategy & Scope
There is a specific, named problem this AI initiative is meant to solve.
i
Can someone on your team write it in one sentence right now? If not, the problem is not yet defined - it is a direction. That is not enough to start.
This problem occurs regularly - not once a quarter or in edge cases only.
i
AI and automation deliver value through repetition. A problem that appears rarely is not worth the investment of a full initiative.
We know what a successful outcome looks like in concrete terms.
i
Not "better" or "faster" - something measurable. Time saved per week, error rate reduced, response time cut by X. If you cannot measure it, you cannot manage it.
The first version of this initiative could realistically be live within six weeks.
i
If the plan requires months before anything goes live, the scope is too large. A working MVP in six weeks is a discipline test, not a technical one.
2
Data Readiness
The data this initiative needs is stored in one place and accessible to the project team.
i
If the data lives across five departments, three tools, and two email inboxes - it is not accessible. Centralisation is a prerequisite, not an afterthought.
The data is structured - in tables, fields, or consistent formats - not free text or scanned documents.
i
Unstructured data (PDFs, emails, handwritten notes) can be used, but requires significant preparation work before it is useful to any AI system.
It is clear who owns this data and who has permission to use it for this project.
i
Data ownership disputes mid-project are a common cause of delays. If nobody knows who to ask for access, assume it is not ready.
There is enough data volume and quality to produce reliable outputs from the start.
i
Ten records are not enough. If the dataset is thin or full of inconsistencies, outputs will be unreliable - and adoption will fail faster than the technology.
3
Leadership Alignment
This initiative has been communicated to the teams it will affect - not just approved at the top.
i
A leadership decision that never reaches the people affected is not a decision - it is a plan. Communication down the chain is where most AI initiatives silently die.
One person inside the organisation is accountable for this initiative delivering results.
i
Shared ownership is no ownership. If everyone is responsible, nobody is. Name the person - if you cannot, that is your answer.
No department head has the authority and motivation to block or stall this initiative.
i
One resistant manager with enough influence can outlast any consultant or project timeline. This is not a technology problem - it is a political one.
Leadership expectations for this initiative are grounded in your company's actual size and maturity.
i
A 30-person company does not need a Silicon Valley AI platform. Misaligned expectations are the fastest way to declare a working initiative a failure.
4
Use Case Fit
The task you want AI to handle is repetitive and pattern-driven - not highly creative or context-dependent.
i
AI performs well on tasks it has seen many times. Writing personalised outreach emails, summarising documents, classifying support tickets - these work. Unique strategic decisions do not.
You have considered whether simple automation - without AI - would solve this problem faster.
i
Most operational problems in SMEs are automation problems, not AI problems. If every click in a process is predictable, you do not need a language model - you need a workflow.
A human remains responsible for any decision that carries legal, financial, or reputational risk.
i
AI can assist, draft, and surface options. It should not be the final decision-maker on anything with real consequences. Human-in-the-loop is not a limitation - it is the design.
You are starting with one concrete use case - not a platform, department rollout, or company-wide transformation.
i
The most successful AI adoptions start with something embarrassingly small. One use case, one team, one workflow. Prove it works, then scale.
5
Organisational Readiness
The people who will use this tool know it is coming and were involved in defining what it should do.
i
Surprise deployments fail. People adopt tools they helped shape. Even a single conversation with end users before build starts changes the outcome.
At least one person at team level is genuinely enthusiastic about this initiative and will support colleagues.
i
An internal advocate - someone who is not management but believes in it - is the single strongest predictor of adoption success. Without one, uptake stalls the moment the consultant leaves.
There is a plan for onboarding and training - not just access and a link to documentation.
i
A tool without training is a tool that will not be used. Even a 30-minute walkthrough with a follow-up session doubles adoption rates compared to self-service rollouts.
The organisation accepts that version one will be imperfect and has a feedback process planned.
i
Organisations that expect perfection from launch kill their own initiatives. A structured feedback loop from week one is not optional - it is how the system improves.