Discovery should narrow scope, not expand it
A serious GenAI engagement begins with interviews, document sampling, and a ranked list of user journeys, not a generic "AI strategy deck." You should leave discovery with one MVP hypothesis, success metrics, and explicit non-goals.
Expect deliverables: data inventory, risk notes, architecture sketch, and a week-by-week plan. If a vendor cannot name the first corpus and the first hundred evaluation questions, they are not ready to build.
Build phase rhythms
Weekly demos on staging with real documents, not lorem ipsum. Fortnightly security checkpoints if you handle regulated data. A shared eval spreadsheet where product, legal, and ops mark answers pass/fail.
Indian clients often underestimate internal time: someone must approve uploads, triage bad answers, and sign UAT. Consulting accelerates engineering; it does not replace your subject-matter experts.
Handoff and long-term ownership
The project ends with deployed infrastructure, admin guides, monitoring dashboards, and optional train-the-trainer sessions. You should be able to add a PDF without filing a ticket, or know exactly when you need Sabrixa for phase two.
Transparent pricing beats black-box "AI transformation." Fixed-scope MVPs, clear hosting cost estimates, and optional retainers for model upgrades keep GenAI consulting aligned with business outcomes instead of billable mystery.