Hotel AI Needs a Decide, Recommend, Stop Framework and an Override Log
Sirma's Darko Bosancic sorts hotel AI tasks into decide, recommend or stop, while the Mews Hotelier Survey 2026 finds 41% of hoteliers have no formal AI policy.
98% of hoteliers used AI in the past six months, yet 41% have no formal AI policy, according to the Mews Hotelier Survey 2026 as cited by Sirma SVP Darko Bosancic in a Hospitality Net opinion piece. The h2c AI Opportunity Study 2026 points the same way: 91% of participating chains use AI, but only 28% have a company-wide strategy led by senior leadership.
Bosancic, whose firm supplies hotel software, argues that “human in the loop” often becomes a human in every loop: friction without meaningful oversight. He offers three questions for each task. How reversible is a mistake? What does it cost in money and trust? Does clean, unified data exist across the PMS, CRM and POS? The answers sort tasks into decide, recommend or stop. Night-audit reconciliation and invoice matching are decide tasks. Rate moves and suite upsells are recommend tasks. A complaint about an unready room needs a human. He also flags the opposite error: automating guest-facing tasks while back-office invoices are still keyed by hand.
Autonomy should be earned. Every task starts in recommend, overrides are logged, and a task is promoted once override rates stay low. Hotel revenue management shows why. RoomPriceGenie’s 2026 State of Revenue Management Report, a vendor-run survey of 450 senior revenue decision-makers, found that 77% trust AI pricing to some degree but only 10% would let it make most pricing decisions, and most check recommendations manually. CEO Chas Scarantino said hoteliers need to understand why a price is recommended.
For a GM or ownership group, the operating takeaway is a log. Record every override by task and let the data decide what graduates. A night-audit task that earns decide status frees labor hours; a rate move that stays in recommend keeps the revenue manager accountable for flow-through. Bosancic notes his own firm has not yet chosen a pilot, so treat the framework as a starting design to test on one task, not a proven playbook.
Source: Hospitality Net — The Human-in-the-Loop Hotel: Where AI Should Decide, Recommend and Stop Auto-generated brief — verified before publishing.