An AI readiness assessment for one business process
A small business does not need a company-wide AI score to choose its first project. It needs one process with accessible inputs, a clear owner and a result that somebody can check.
Assess a task with a beginning and an end
“Use AI in the business” is too broad to assess. “Draft the weekly report from these three exports” has an input, an output and a person who knows whether the result is useful. Start at that scale.
Observe the current process before proposing a tool. Write down who does it, how often it happens, how long the work and checking take, and what happens when an input is missing. Include the awkward cases that get handled in email or in someone's memory.
My own starting point was internal agency work: connecting live systems and keeping account context available to the team. The AI connector write-up describes that work. This assessment is a planning worksheet, not a certification or a tested prediction of project success.
Get the process assessment worksheetCSV · fill in one copy per process · no email required
Seven questions to answer before a pilot
Mark each area Ready, Needs work or Unknown. Add a short piece of evidence and name the person who can resolve a gap. An unanswered access question should stay visible; averaging it into a numerical score would hide the blocker.
| Area | Question | Useful evidence |
|---|---|---|
| Purpose | What finished output is needed, and who uses it? | A recent example that its recipient considers good. |
| Inputs | Are the required records accessible and understandable? | A representative sample, including missing or messy entries. |
| Access | Can the chosen tool use these records for this purpose? | Confirmed permissions, data restrictions and supplier terms. |
| Judgement | Does the task require language interpretation, or can a rule handle it? | Examples of decisions and the instructions behind them. |
| Checking | Can a person recognise a wrong answer before it causes harm? | A review checklist and a named reviewer. |
| Ownership | Who deals with failures and changes after launch? | A process owner, a fallback and a maintenance route. |
| Value | Will the result justify building, checking and maintaining it? | A current time baseline and a rough total running cost. |
Data access is more than a working login. Check which records are appropriate to use, what the supplier retains and whether the proposed use fits the business's obligations. If personal data is involved, the ICO's AI and data protection guidance is a starting point for that review.
A readiness assessment can end without AI
Use an existing feature
A scheduled export, CRM rule or spreadsheet formula may already complete the task. Test that route before commissioning a connection between systems.
Build a plain automation
If the input and decision rules are consistent, a conventional integration may be easier to inspect and maintain. Moving an invoice reference between two known fields does not need a language model.
Run a narrow AI pilot
Language-heavy work can be a reasonable candidate when the inputs are permitted, mistakes can be detected and someone owns the result. Keep the first version as a draft for review.
Resolve the gaps first
Pause where data permissions are unknown, nobody can check the output or a failure would create an unacceptable consequence. Assign the missing work to an owner and revisit the decision when there is evidence.
The longer process automation guide explains how existing software and model-based steps can fit together.
Turning a weekly report into a pilot brief
This is an illustrative scenario, not a client result. A team assembles a report from three exports, writes a commentary and sends it to a manager. The figures can be joined and calculated with ordinary code. Drafting a short explanation of the changes is the part worth testing with a model.
- Input: agreed exports for the same reporting period, with definitions for each metric.
- Output: a draft commentary linked to the figures used to produce it.
- Check: the reviewer verifies every number and removes explanations that the data cannot support.
- Failure: missing inputs stop the draft and produce a visible message for the owner.
- Boundary: the pilot cannot send the report externally or change source records.
Measure the current process and the pilot with the same method. Count preparation, corrections and review time as well as generation. A fast draft that takes longer to check has not saved the team time.
Monthly time released = time spent on the current process − time spent operating and checking the new process.
Compare that with implementation, subscriptions, usage charges and ongoing maintenance. Keep estimates separate from measured results.
Write the exit conditions before building
The pilot brief needs an owner, a defined set of inputs, an output example and a list of actions the system is allowed to take. Use examples that represent ordinary work and the cases that are likely to go wrong.
- Agree what counts as an acceptable result and how it will be checked.
- Keep test cases separate from the examples used to build the workflow.
- Record failures, correction time and any unsupported statements.
- Decide who can approve wider use, and what would stop the pilot.
- Keep the current manual process available until the replacement is accepted.
If missing context is the main problem, start with the shared working-memory system. A new model cannot recover decisions that were never written down.
My AI consulting work in Manchester covers selecting a process and building the integration. More technical explanations sit in the tracking and automation library.
Questions about assessing readiness
Is this assessment intended for the whole business?
No. It assesses a single workflow. One team may have a suitable reporting task while another lacks reliable records or an owner. Repeat the worksheet for each candidate process and compare the evidence.
Does a small business need a large dataset before trying AI?
Not necessarily. The requirement depends on the task. A document-based workflow may need a small, well-maintained set of records. It still needs representative examples for testing and a person who can judge the output.
What should happen when the assessment finds a blocker?
Record the gap, its owner and the evidence needed to resolve it. Fixing access, documenting a process or choosing a reviewer can be the whole next step. The assessment does not commit the business to buying an AI tool.