VIVI president Lance Thompson joins Angelina Yang on TwoSetAI for a 65-minute conversation about building and selling AI agents into hospitality.

Integrations carry the work

  • Hotels combine aging phone systems, outsourced reservations, property-management software, room-service ordering and maintenance dispatch.
  • A reservation agent must handle connecting rooms, bed preferences, flexible dates, availability and a payment handoff.
  • Breakfast orders expose missing domain knowledge: eggs, toast and accompaniments need follow-up questions before an order is complete.
  • Thompson describes separate engineering and customer-success tracks: one builds connections; the other turns operating procedures into agent behavior.

Knowledge, feedback and proof

  • An agent can draw from a hotel’s website, daily SharePoint updates and the ski mountain’s operating information.
  • His description of “training” centers on retrieved knowledge, prompts, testing and transcript review; he does not describe a custom model-training procedure.
  • A swimsuit question initially failed because the knowledge base omitted where guests could buy one. The team added the missing information.
  • His end-to-end test: make a reservation by phone and watch it appear in the property’s actual system.

Sell a relationship that permits experiments

  • Thompson says the first client’s roughly 30 proposed use cases led to a monthly advisory retainer with no separate build fees under that arrangement.
  • Clients can test and abandon ideas without defending a separate sunk build cost for each agent.
  • He estimates that 80–85% of clients use the advisory arrangement; other engagements can carry build, monthly and consumption fees.
  • Frontline staff help test real scenarios, contributing expertise while reducing anxiety about the rollout.

Price around usage and customer constraints

  • Historical call volume helps estimate costs; some clients pay per active minute, while others negotiate predictable monthly charges.
  • Thompson says agents typically start around $500 per month, with consumption charges depending on the arrangement. He also describes lower-cost exceptions.
  • Small overruns may be absorbed; larger changes trigger a discussion. The episode gives no universal per-minute tariff.
  • Per-room pricing frustrated him as an operator: a complex 90-room luxury property can require more work than a simple 300-room hotel.
  • Seasonal properties, owner-operators and management companies need different commercial terms.

Keep the human service visible

  • Thompson’s examples include answering a hairdryer question before a possible 30-minute reservations queue and relieving an overloaded front desk.
  • He wants saved capacity directed toward guests, housekeeping and kitchens; that is his deployment goal, not measured evidence of industry-wide employment effects.
  • His suggestions for builders include hotel procurement and making properties discoverable to personal AI agents. He explicitly says guaranteed AI-search placement remains unsolved.
  • Caveat: customer outcomes, adoption figures and reliability claims are Thompson’s account. The interview supplies no independent evaluation or published pricing schedule.

“You don’t want to be a mile wide and an inch deep.”