An assessment should answer a decision.
Which AI tool should we buy is often a premature question. More useful starting points are which delay prevents better service, where corrections accumulate and which tasks depend on information scattered across people and tools. Observe a sample of work and compare the manager’s account with the operator’s experience.
The output should be a short, reasoned set of priorities with prerequisites, a non-AI alternative and a suggested first experiment. An opportunity can be attractive but unready: the required data may have no owner, or nobody may be able to judge whether the output is correct.
Prepare a small set of useful evidence.
A polished process map is not a prerequisite. Missing evidence should remain visible, with a practical way to collect it. A precise number built on untested assumptions is less useful than a supported range with an open question. Include differences between locations or language groups when they change how the process actually works.
- Three to five troublesome processes with a starting event and an end result.
- Ordinary cases, difficult examples and requests that remain unresolved.
- Separate estimates of volume, active effort, waiting time and correction frequency.
- Tools, owners, available access and changes already planned.
- Someone who can judge whether a trial result is both correct and useful.
Worked comparison: three ideas, three possible decisions.
The highest priority is not necessarily the most impressive demonstration. A modest, frequent task with a clear owner and verifiable output can be a stronger first experiment than an ambitious programme with many dependencies. These decisions illustrate an assessment method; they do not describe your organisation or a previous client.
Scroll the table to compare all columns.
| Idea | Evidence needed | Possible decision |
|---|---|---|
| Route support requests | Stable categories and previously classified requests | Test a sample and inspect assignment errors |
| Forecast demand for a new product | Relevant history and a simple forecast baseline | Defer until a verifiable comparison exists |
| Copy a table between tools | Stable schema and available export or API | Consider rules-based automation without a generative model |
Specify the outputs you expect to receive.
An essential map of observed processes
Show inputs, outputs, people, tools and exceptions. Distinguish observed facts from estimates and claims that still need checking.
A reasoned priority list
Compare expected usefulness, data feasibility, review capacity and dependencies. Volume alone is not a measure of value, especially when an error has substantial consequences.
A first-experiment brief
Describe examples, expected output, measures, internal effort, cost categories and go/no-go conditions. Access that has not been verified remains an explicit dependency.
A preparation backlog
Assign owners to data cleanup, procedure decisions or access requests. The brief should remain understandable and usable by your internal team.
Evaluate assumptions rather than certifying a promise.
Context, measurement and responsibility are consistent with the voluntary NIST AI Risk Management Framework. This reference helps organise the assessment questions. It is neither a service certification nor evidence that a proposed project will succeed.
Separate assessment from implementation in the proposal. State which processes will be examined, who needs to participate and which trials are possible with available access. Where evidence is insufficient to recommend AI, a useful result explains the gap and the next step to resolve it. Avoid a tool shortlist that leaves the operating decision unanswered.
A template to work from.
Business process inventory
A verifiable process map with a beginning, an end, an accountable owner and a baseline. Attach it to your project brief.
Download the Markdown templateAutomation opportunity scorecard
A reasoned comparison, including missing evidence, blocking conditions and a decision about the next experiment.
Download the Markdown templateAI data readiness worksheet
A source inventory with observed defects, remediation actions and an approved sample for the trial.
Download the Markdown templatePractical questions
Does the assessment require a later implementation purchase?
The assessment does not automatically start implementation or recurring charges. Its proposed output is a readable assessment with priorities and prerequisites. Any later implementation and ongoing-care project requires a separate proposal to be agreed and accepted.
Can we start without an estimate of wasted hours?
Yes, if we can observe a sample. Separate active effort from waiting time and make uncertainty visible. The assessment may include an initial evidence collection step before estimating a benefit.
References and method
Voluntary framework referenced to organise context, measures and responsibilities; not a certification of this service.