Start with the work, not the model
A good use case has a clear user, repeatable input, a desired output and a clearly defined decision or action. Only then should you choose AI, rules, integration or conventional automation.
What we map
- Trigger and starting point
- Users, roles and responsibilities
- Inputs, sources and data quality
- Decisions, rules and exceptions
- Outputs, recipients and subsequent actions
- Failure modes, escalation and approvals
Success criteria before a pilot
We define the expected form of value rather than inventing an ROI figure. Examples may include shorter preparation time, fewer missing details, better traceability or more consistent documentation. The baseline and measurement method are agreed before the pilot.
Technology choice
Fixed rules
Conventional workflow
When steps and exceptions can be described precisely.
Interpretation
AI agent
When the task involves searching, classifying or evaluating unstructured information.
Workspace
Custom app
When the user needs status, sources, forms and approvals in one place.
Physical process
Industrial automation
When sensors, PLC/SCADA, machines and operator workflows are involved.
What must be clear before you automate
A solution can only be as good as the understanding of the process. Before anything is built, six things must be describable: inputs and outputs, roles and responsibilities, rules and permitted variations, exceptions, escalation paths, and what currently causes delays or errors.
Automating an unclear process does not make its problems disappear. The flawed process simply runs faster and more consistently, and errors become harder to spot because a system now performs them with confidence.
Baseline, success criteria and the simplest solution
A baseline makes value measurable: how long does the task take today, how often does it fail, and what is the impact of those failures? Without a baseline, no one can subsequently say whether the solution worked.
The mapping also determines the choice of tools. If every step can be described with fixed rules, conventional automation is often sufficient, less expensive and easier to test. AI is relevant where inputs require interpretation: free text, documents, images or judgement. Many effective solutions combine the two.
Common questions
Is process optimisation required before a pilot?+
At a minimum, the process must be scoped clearly enough for a pilot. The level of detail depends on the complexity and risk involved.
Can you automate a process that is not standardised?+
Yes, but variations, exceptions and responsibilities must first be made visible so the solution can handle or escalate them.
How is value measured?+
With a baseline and a small number of specific indicators suited to the task, such as time, omissions, errors or response time.