Skip to content
ASSESSMENT AND IMPLEMENTATION

Decide where AI is useful before committing to implementation.

Assess your processes, data and team capacity to decide what to test, simplify or defer.

Let’s discuss your process

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.

Read the details

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.

Read the details
  • 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.

Read the details

Scroll the table to compare all columns.

Illustrative assessment to repeat with your own evidence
IdeaEvidence neededPossible decision
Route support requestsStable categories and previously classified requestsTest a sample and inspect assignment errors
Forecast demand for a new productRelevant history and a simple forecast baselineDefer until a verifiable comparison exists
Copy a table between toolsStable schema and available export or APIConsider rules-based automation without a generative model

Specify the outputs you expect to receive.

Read the details
  1. An essential map of observed processes

    Show inputs, outputs, people, tools and exceptions. Distinguish observed facts from estimates and claims that still need checking.

  2. 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.

  3. 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.

  4. 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.

Read the details

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.

FROM IDEAS TO A BRIEF

A template to work from.

Business process inventory

A process map with boundaries, an owner and a baseline.

Download the Markdown template

Automation opportunity scorecard

A comparison of missing evidence, constraints and the next experiment.

Download the Markdown template

AI data readiness worksheet

Sources, defects, remediation actions and an approved sample.

Download the Markdown template

Practical 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

NIST — AI Risk Management Framework 1.0: Core

Voluntary framework referenced to organise context, measures and responsibilities; not a certification of this service.

Which process should improve first?

Let’s discuss your process