An AI consultant’s most useful question may be a practical one: where does work slow down? A model can draft an answer, but it cannot decide on its own which business problem deserves attention or who should approve an action.
Start with evidence. Imagine a property office that receives enquiries but takes too long to reply. Before adding more advertising, record how many enquiries arrive, how quickly staff respond and how many lead to appointments. If those figures do not exist, agree on a short baseline measurement. This is an illustrative scenario, not a reported result.
Rules and practical implications
Turn a broad ambition into a goal that can be observed, such as reducing first-response time. Break the work into receiving an enquiry, assigning it and preparing a reply. A notification or an existing software setting may solve part of the problem without generative AI. Select the simplest suitable approach and define what a successful pilot would look like.
Set boundaries before connecting accounts. Specify which data can be used, who owns the accounts, what the system may do and which actions require a person’s approval. Review provider terms, access rights and applicable data protection requirements. A draft contract or financial recommendation needs qualified human review; an AI-generated answer should not become a final decision by default.
Test ordinary, incomplete and difficult cases against criteria agreed in advance. Track accuracy, unsuitable responses and interventions, including individual serious failures rather than only an average score. Anthropic describes combining automated, model-based and human assessment. NIST’s voluntary AI risk framework treats mapping, measurement and management as continuing activities.
Evidence and responsible use
After a limited trial, compare results with the baseline and ask the team what helped or failed. Document handover, support, costs, access and a way to stop or restore the workflow. A change in the metric alone does not prove that AI caused it; seasonality and other changes may matter too.
Mert Durmazer’s Digital Academy video outlines a consulting workflow from identifying constraints to team adoption. His accounts of clients and income were not independently verified and are not a promise of earnings or customer acquisition. Ambiguous automatic-caption terminology has not been silently treated as a verified product name.
