Journal / Kirill Naumov
How to choose your first retail AI pilot
Choose a bounded use case, establish a baseline and test whether AI actually helps employees and customers.
By Kirill Naumov ·
“Which AI should we buy?” is the wrong starting point. First identify the work you want to make faster or better and who will notice. This is an advisory framework, not a report of deployment outcomes.
Start with recurring work
A sales associate needs product compatibility information. Measure lookup time, answer accuracy and escalations to experienced colleagues. Check for an up-to-date source. If none exists, the pilot begins with organising information, not buying a chatbot.
Compare candidate use cases
Record task frequency, consequences of error, data readiness, human review requirements and current effort. The priority is not the most impressive demo: it is a useful hypothesis that can be tested safely and affordably. Compare AI with conventional search or a short reference guide.
Limit the scope
Start with one category, a small staff group and approved sources. Assign a process owner and someone responsible for updates. Specify what the system must not answer and when a specialist takes over. Customer data must not enter unapproved tools.
Agree on stop conditions
Critical errors, excessive checking or a lack of voluntary reuse are reasons to reconsider. Set criteria before launch. Compare commercial outcomes across comparable periods or groups, accounting for footfall, stock and promotions. Higher sales alone do not establish an AI effect.
Prepare for an initial discussion
Bring one workflow, available baseline measures, data constraints and a named process owner. That is enough for a useful conversation about pilot boundaries. There is no need to buy a new tool before a diagnostic.