Value for the business
How often does the problem occur? How does it affect revenue, cost, capacity or timing?
Where to start with AI?
Start with the work that needs improving.
A lost sale, a request sent back, a slow reply. Understand what limits the outcome and choose the next step to investigate.
Describe the task.
Say what needs to change.
Who does the work? What gets in the way? What improvement needs to be seen? A concrete answer helps you choose what to observe.
An example of the reasoning
The workCheck requests
Check requests
and their attachments.
Investigate why requests go back to the people who submitted them.
The desired improvementFewer returns.
Fewer returns.
Reliable checks.
Reduce returns without missing important issues.
What do we need to find out?
Do requests get sent back because of missing fields, unclear criteria or difficulty interpreting attachments? Examine examples of each situation with the people who check them.
A rule can prevent a field from being left blank. AI can support document reading. Clear criteria and human review help assess whether an issue was correctly identified. The choice depends on the cause observed.
Which opportunity
deserves attention first?
Compare the impact, effort and conditions for investigating. Separate what the records show from what is still an expectation.
Available evidence
Are there examples, records and people who know the task? What still needs confirming?
Conditions for moving forward
Can the data be used? Who will participate? What effort, dependencies and risks need to inform the decision?
When is it worth improving the process or using rules?
If the difficulty lies in ambiguous guidance, clarify the criterion. If the condition is predictable, consider a rule. If it requires reading and interpretation, investigate AI assistance. Also compare a more structured manual check, considering quality and total effort.
The next step depends
on what you already know.
Four independent services. You can start with the one that addresses your current decision, without committing to a series of services. A specific question may only need a focused investigation.
A priority still needs to be chosen
Assessment
Understand the process, compare opportunities and recommend where to focus investment.
Explore the assessmentThe idea is there. Evidence is missing.
Validation
Compare the solution with daily work and assess quality, errors and review effort.
Explore validationApproach evaluated. Still to be put into use.
Implementation
Build and integrate the solution, define responsibilities and prepare for operation.
Explore implementationThe solution is already part of daily work
Operations and improvement
Monitor quality, use and costs, address incidents and prioritize improvements.
Explore operations and improvementFor the next conversation
Bring a one-page outline.
You don't need everything resolved.
- The task and the people who know how it is done.
- The outcome that needs improving.
- Available examples and open questions.
- An option with AI and one without it, where appropriate.
- Who will assess the next piece of evidence.
Reference for further reading
The NIST AI RMF 1.0 · Manage 2.1 considers viable non-AI alternatives when managing risks.
Talk to Gaulett
What work needs
improving first?
Bring the challenge and what your team already knows. Let's discuss the desired outcome and the next step.