Journal / Kirill Naumov
Choose the right scope before choosing a provider: diagnostic, implementation or team training. Questions to ask about evidence, ownership, measurement and handover.
By Kirill Naumov ·
Choose an AI consultant by the decision you need to make, not by the number of tools in a presentation. A retail business deciding where to begin needs a different engagement from a business integrating a tested workflow into its systems. Neither is the same as training store colleagues.
A diagnostic helps define a problem, compare possible interventions and choose an initial test. It should leave you with a clear decision, the main unknowns and a next step. It is not a production system or a promise of a return on investment.
Implementation requires named responsibility for integrations, approved information, testing, access controls, monitoring and support. Ask who will do each part and whether that work is included or needs another specialist. An adviser’s retail experience does not by itself establish engineering capacity.
Team training should address a real workflow, not just demonstrate prompting. Ask how colleagues practise exceptions, check answers and hand a problem to a person. Attendance is easy to count; consistent use and decision quality are more useful to observe.
For example, describe a customer who wants to know whether a product will work with something they already own. Ask the consultant what must be known before a system can answer. Listen for product data ownership, ambiguity, the cost of being wrong and the point where a colleague takes over. This is an evaluation scenario, not a claim about a completed project.
Then ask for the non-AI alternative. Better information, a clearer process or a small change to staff responsibilities may solve the problem. A recommendation is more credible when it explains why AI is warranted—and when it is not.
Distinguish a concept, a prototype, a pilot and a live deployment. They answer different questions. A concept shows thinking; a prototype shows feasibility; a pilot tests a bounded hypothesis; a live deployment adds operating evidence. Ask what the provider personally did and what the figures actually measure. Respect client confidentiality rather than requesting private customer data.
For an outcome claim, ask for the baseline, observation window, denominator and comparison. A faster response may conceal more corrections. Higher sales during a promotion may have nothing to do with the tool. A credible answer includes limitations rather than only the best number.
The scope should identify the business owner, deliverables, exclusions, milestones, access needed, cost limits, acceptance checks and the handover. Ask what happens if the information is incomplete, the team does not adopt the workflow or the pilot misses its target. Decide who maintains the system after the consultant leaves.
Do not send confidential customer or employee information just to obtain a proposal. Use an anonymised process description until appropriate data handling and access arrangements are agreed.
Kirill works with retail, luxury, hospitality and service leaders on AI adoption, customer experience and team development, in English and Russian. His background includes 11 years at Apple. The starting diagnostic is $500 for 45 minutes: one business problem and three practical priorities. Implementation, workshops and a longer roadmap are scoped separately.
The IKEA, Dyson, LEGO and Starbucks materials on this site are independent concepts, not commissioned implementations. Use them to assess the approach, not as evidence of measured client results. For a first conversation, describe the process, the current difficulty and the decision you need help making.