Start with the task
Choose a recurring task with clear boundaries: classifying enquiries, preparing a draft or searching approved documents. Record who performs it, which information they need and which decisions they make.
Define success
Measure the current process first. Time, errors, exceptions and review effort provide a useful baseline. Agree what the pilot should improve and which prerequisites must be met before wider use.
Test the difficult cases too
One successful example does not prove the entire process works. Test missing information, ambiguous requests and conflicting sources. Define when the system should stop and ask for human help.
Expand what delivers value
After the pilot, compare results with the baseline. Include operating costs, maintenance and human review. Expansion makes sense when value remains positive in everyday use.
A practical example of choosing a starting point
Suppose a sales team spends too much time organising new contact enquiries. Define a focused task: prepare an internal summary and flag missing information. The scope excludes automatic quotations, pricing and promises to the customer. Those limits make the test easier to evaluate. The team can check whether the summary is accurate and the follow-up questions are useful.
Build a small, representative sample
Collect varied examples of the same task: a complete enquiry, one with missing information and one outside the usual scope. Use appropriately anonymised data. For each case, describe an acceptable result and the errors that would make it unsuitable. The sample becomes a shared reference for design, testing and deciding whether to expand.
Compare the whole process
Drafting time is only part of the picture. Measure data preparation, editing, corrections and exceptions. Add tool and support costs. If a system quickly produces something that needs extensive rewriting, its initial speed may not deliver the expected value. Evaluation should focus on the final usable result.
When to stop or change direction
A pilot may reveal that the data catalogue or internal process needs improvement first. It may also show that simple automation is enough. These are useful findings. Record what the team learned, what is needed to continue and which alternatives deserve consideration. Expansion should depend on actual usefulness and the quality achieved.
Frequently asked questions
What is AI consulting?
It connects AI capabilities with a specific business's needs. It includes understanding processes, assessing data, choosing applications and planning implementation. The goal is to understand what is worth doing, why and under which conditions.
Do I need to know AI tools already?
No. We start with your work and the difficulties you face. Tool selection follows an understanding of the task. If you already use AI, we assess what works, where it fails and what could be better organised.
What will I receive from an AI Audit?
The agreed deliverable may include process mapping, data-readiness assessment, priorities and a proposal for a focused pilot. We define the exact scope before starting so you know which teams, processes and systems will be examined.
How do we choose the first project?
We compare expected usefulness with complexity, cost and the ability to review the result. We prefer a task with clear inputs, outputs and an accountable evaluator. An impressive demonstration alone is not enough.