Walk into any multi-property hotel group's paid media account in the GCC and you'll find a pattern that nobody on the brand side intended and almost everyone on the agency side tolerates: campaigns from different properties bidding against each other on the same audiences, in the same auctions, for the same conversions.
It's not a creative problem. It's not a budget problem. It's an architecture problem — and it's costing most multi-property operators 15% to 25% of their media spend.
This is what we found inside the Hilton, Marriott, and a third (unnamed for confidentiality) GCC hospitality engagement. The numbers reconciled across all three.
How the cannibalisation happens
The standard pattern is straightforward and depressingly common:
- Each property has its own marketing budget, its own KPIs, and often its own agency or in-house team.
- Each property runs paid media targeting "people interested in luxury hotels in Dubai" or some near-equivalent.
- The targeting overlaps. The audience is the same population. The keywords overlap. The dynamic-search-ad coverage overlaps.
- The platforms are happy to show ads from multiple properties to the same user, in the same auction, often within hours of each other.
- Cost-per-click rises. Cost-per-booking rises. Each property reports the same conversions through their own attribution. Sum-of-properties exceeds true revenue.
The platforms have no incentive to surface this. From their perspective, more advertisers bidding against each other is the entire business model. The agency has no incentive to surface it either, since fixing it usually means consolidating accounts and reducing the visible scope of work.
The brand pays for the friction. Twice.
The three-layer architecture that fixes it
What we deployed for Hilton, and refined for the others, is a three-layer campaign architecture that respects each property's identity while preventing the auction collisions:
Layer 1: Brand layer (portfolio-wide)
A single set of campaigns targeting users in the upper-funnel "considering a hotel in [city]" segment. Creative speaks to the brand portfolio rather than any individual property. Budget is allocated centrally and the audience is excluded from layers 2 and 3 until they show property-specific intent. This layer captures the cheapest impressions and warms users for property-specific targeting downstream.
Layer 2: Demand layer (property-specific, gated)
Per-property campaigns targeting users who have shown property-specific signals — visited the property's page, queried by property name, engaged with property-specific content. These campaigns can use property-distinct creative and bid against each other only in the narrow case where a user has shown signals for multiple properties (which is rare and worth competing for).
Layer 3: Intent layer (booking-stage)
Conversion-focused campaigns targeting users with high commercial intent (travel-date searches, room-type comparisons, abandoned booking flows). Property-specific. Tightly geo-bounded. Optimised against confirmed bookings rather than upstream conversions. This is where the highest CPCs are tolerable because the conversion rate is highest.
The architecture works because each layer has a clear signal threshold for entry and clear exclusions for users who've moved into a higher-intent layer. The auction collisions stop because users are systematically routed to the campaign that is most likely to convert them, not the campaign that is most willing to bid for them.
The data architecture that makes it possible
The architecture above only functions if the audience signals can be cleanly defined and shared across the layers. That requires a unified data layer that most hotel groups don't have. What we set up:
- A property-agnostic event schema in the analytics layer. Page views, search queries, and conversion events are tagged with property, but the schema is unified.
- A central audience definition layer (typically in a CDP) where the gating logic lives. New properties inherit the existing audiences rather than creating fresh ones.
- Cross-platform exclusion lists synchronised across Google, Meta, TikTok, and DV360. Users in the intent layer are excluded from the brand layer in every channel, and the exclusions are refreshed on a daily cadence.
- A reporting layer that rolls up portfolio-wide and decomposes by property, so the central team and the property-level teams see the same data with appropriate slicing.
This setup takes four to six weeks to implement properly and is the precondition for the campaign architecture working. Most engagements skip it and try to fix the campaigns first; that's why most engagements don't deliver the ROAS lift the architecture should produce.
The political problem
The technical architecture is the easy part. The hard part is political. A few patterns we've worked through:
Property GMs are evaluated on property-level revenue. Centralising the brand layer means a portion of "their" demand-generation budget moves to a layer that doesn't sit under their P&L. This needs to be solved upfront with a credit/allocation model, or it will be re-fought every quarter.
Multi-agency setups multiply the friction. If three different agencies run three different properties' paid media, the architecture cannot be unified without consolidation. We've seen consolidation done badly (a single agency taking over all properties with no architectural change — same problem, fewer agencies). It has to be paired with the architecture work.
The first month looks worse before it looks better. Disabling overlapping campaigns reduces total impressions and total clicks before the audience routing finishes calibrating. Property-level dashboards will show a dip. Without buy-in from the property GMs, this triggers a rollback before the architecture has had a chance to compound.
What the numbers looked like
For Hilton's UAE portfolio (four properties, six-month engagement) the architectural rebuild produced:
- −22% on overlapping audience spend. The cleanest single-number proof that the cannibalisation was real.
- +38% on direct-booking ROAS blended across properties, measured against pre-rebuild baseline with controlled holdouts.
- 3.2× ROAS at the portfolio level, sustained for the full engagement period.
- Lower CPCs at the brand layer, higher CPCs at the intent layer — the right shape, since the intent-layer users are the ones actually converting.
The Marriott engagement, structured similarly, found a slightly different mix because the property portfolio shape was different — more international demand, shorter booking windows — but the directional result was the same. Architecture rebuilds find structural inefficiency that creative and bid optimisation alone cannot reach.
The takeaway for multi-property operators
If your group is running independent paid media programmes per property, the question is not whether you have cannibalisation — you do, the only question is how much. The honest test is to run a single weekend's campaigns through an audit tool that flags overlapping auctions across your campaigns. Most groups are surprised by the answer.
Fixing it is not a creative refresh, not a bid optimisation, not an agency review. It's an architecture project. It usually pays for itself inside one quarter. And it's the prerequisite for any meaningful AI-native scaling — agents can only optimise within the architecture they're given. A bad architecture rate-limits the entire engagement.