Multi-Agent Orchestration in Barcelona, Spain

Tourism / Hospitality

A Barcelona hotel group (14 properties, 2,800 rooms total, ranging from 3-star to luxury 5-star, €95M annual revenue) managed revenue across 18 distribution channels (Booking.com, Expedia, Airbnb, direct website, 8 tour operators, 5 corporate booking platforms, and 2 cruise line partnerships). The problems: (1) Pricing decisions were reactive and manual. Revenue managers updated room rates across channels using spreadsheets and manual channel manager inputs. Rate adjustments happened 2-3 times weekly — but competitor pricing, demand signals, and market conditions changed hourly during peak seasons (June-September, Mobile World Congress week, La Mercè festival). The group estimated it captured only 72-78% of achievable revenue due to pricing lag. (2) Channel management was fragmented. Inventory allocation across 18 channels was managed manually — overbookings occurred 3-4 times monthly (requiring expensive relocation — an average of €450 per walked guest in Barcelona), while rooms went unsold on slow channels. (3) Forecasting was inaccurate. Demand forecasts relied on historical occupancy data and revenue manager intuition. Forecast accuracy: ±22% for 30-day forecasts, ±35% for 90-day forecasts. The inaccuracy led to either overpricing (losing bookings to competitors) or underpricing (leaving revenue on the table). (4) Regulatory compliance required attention. The Catalan tourist tax (IEET — Impost sobre les estades en establiments turístics), VAT management across booking channels (different VAT treatment for OTA vs. direct bookings), and Generalitat licensing requirements consumed significant administrative time.

We built a 5-agent revenue management orchestration system over 10 weeks. (1) Demand Forecasting Agent: predicting occupancy and demand by property, room type, and guest segment using: historical booking data (3 years of booking curves by property and room type), forward-looking signals (flight search volume to Barcelona — BCN airport — from key source markets, event calendar — MWC, Sónar, La Mercè, Champions League matches — with estimated visitor impact, cruise ship schedule — the Port of Barcelona handles 3M+ cruise passengers annually with predictable demand patterns), competitor intelligence (rate scraping from 25 competitor properties across Booking.com and direct channels — refreshed every 4 hours during peak season), and macroeconomic indicators (EUR exchange rates affecting key source markets — GBP for UK tourists, USD for US tourists). Forecast accuracy: ±8% for 30-day forecasts (vs. ±22% previously), ±15% for 90-day forecasts (vs. ±35%). (2) Dynamic Pricing Agent: setting optimal room rates by property, room type, date, and channel using: demand forecast (from the Forecasting Agent), competitor pricing (real-time from rate shopping data), price elasticity models (trained on the group's historical booking data — different elasticity by segment, season, and advance booking period), group booking displacement analysis (when a group inquiry arrives, calculating whether accepting the group at a discounted rate or holding rooms for higher-rated individual bookings maximises total revenue), and minimum rate floors (ensuring profitability after commissions, taxes, and variable costs for each channel). Rate updates: every 4 hours during peak season, daily during shoulder season, twice weekly during low season — vs. 2-3 times weekly manual updates previously. (3) Channel Distribution Agent: allocating inventory across 18 channels to maximise revenue while preventing overbookings. The agent: maintaining real-time availability across all channels (two-way sync with each channel's API), implementing channel-specific strategies (higher rates on high-commission channels, best-rate guarantees on direct channels, exclusive packages for tour operators), managing closeout rules (closing high-commission channels first as occupancy approaches capacity), and overbooking management (calculated overbooking based on historical no-show and cancellation rates by channel and segment — eliminating accidental overbookings while maintaining optimised occupancy). (4) Compliance Agent: automating regulatory requirements. The agent: calculating and documenting Catalan tourist tax (IEET) per stay — different rates by star category and zone, generating tax declarations for quarterly IEET filing with the Agència Tributària de Catalunya, managing VAT treatment by channel (direct bookings vs. OTA commissions vs. tour operator rates — each with different VAT implications under Spanish tax law), and monitoring licensing compliance per Generalitat de Catalunya tourism regulations. (5) Analytics and Reporting Agent: synthesising performance data across all properties and channels. The agent: generating daily revenue reports (RevPAR, ADR, occupancy, and channel mix by property), producing weekly competitive position analysis (the group's pricing and occupancy vs. competitive set), creating monthly board reports with trend analysis and forecast accuracy assessment, and identifying revenue optimisation opportunities (underperforming room types, channels with excessive commission costs, pricing gaps in specific date ranges). The orchestrator coordinating agents with human gates for: significant pricing strategy changes (revenue director approval for rate changes exceeding ±15% from the AI recommendation), group booking decisions (revenue manager approval for group quotes above threshold), and tax filing submission (finance team review of AI-generated tax declarations).

Revenue per available room (RevPAR): increased by 14.2% in the first full year of operation — EUR 12.8M in additional revenue across the portfolio. The improvement driven by: better pricing (capturing peak demand with higher rates, and stimulating demand during slow periods with targeted pricing), channel optimisation (shifting bookings toward lower-commission channels — direct bookings increased from 18% to 31% of total revenue), and reduced revenue leakage (overbookings eliminated — from 3-4 per month to zero; no-shows managed through calculated overbooking). Forecast accuracy: ±8% at 30 days (from ±22%), enabling confident advance pricing decisions. Channel management: overbooking relocations eliminated entirely — saving €450 per incident × 42 annual incidents = €18,900 plus immeasurable guest satisfaction impact. Direct booking share: increased from 18% to 31% — reducing commission costs by €1.9M annually. Regulatory compliance: IEET tax filings automated — saving 120 hours of finance team time annually. Zero tax filing errors (vs. 3 corrections required in the previous year). Annual value: €12.8M (incremental revenue) + €1.9M (commission savings) + €180K (operational efficiency) = €14.88M annual value. Development cost: €195,000. Payback: under 6 days of operation.