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.