ARARAT Group

Launching PMax cold on 10 venues with no historical data would have produced months of poor performance while the algorithm guessed at audience signals. Running Search first created real conversion data — which venue queries converted, which audiences, which times. That data became the training signal for PMax, shortening the learning phase and improving the quality of the first optimization cycle.

A single PMax for 'ARARAT Group' would have produced a blended CPA with no way to know which venue was efficient and which was wasting budget. Cloud9 at €1.01 CPA and Breakfast & Lunch at €0.59 CPA require different bidding strategies, different audience signals, and different creative directions. Separate campaigns make each decision traceable.

When Search campaigns were paused and budget moved entirely into PMax, CPA spiked to €2.88. This was expected: PMax needed to rebuild its audience model without the Search data it had been learning from. The question was whether to abort and revert — or hold the new structure through the learning phase. Holding was correct: by March the algorithm had found its footing, and by May CPA hit a record low of €0.66.

Cyprus has a significant and growing Russian-speaking tourist and expat community. Adding Russian-language asset groups to Macallan, Amorevino, Little Italy, and Cloud9 opened a segment that English-only campaigns cannot reach — these users search in Russian and respond to Russian copy. The incremental cost is low; the incremental reach is material.