Search conversion journeys and the missed opportunity of associated keywords

David Rothschild ^1* Coen Needell ² Joe Veverka ³ Elad Yom-Tov ⁴

• ¹ Microsoft Research, New York, United States
• ² Reality Defender, New York, United States
• ³ Microsoft (United States), Redmond, Washington, United States
• ⁴ Bar-Ilan University, Ramat Gan, Tel Aviv District, Israel

We map the conversion journey (i.e., the path a costumer takes to purchase a product or service) for phones, laptops, and vehicles (i.e., cars and trucks) by analyzing a massive corpus of online searches on Bing by people in the U.S. during 2021. We show that, contrary to the idealized version of the model, the observed path of customers is often circuitous and haphazard. Nevertheless, major advertisers heavily concentrate their advertising late in the customer's journey, on people who are not just already likely to buy in the product category, but likely to be committed to a specific brand. Using a natural experiment we show that, overall, ads correlate with increased conversions for vehicles by 7%, 22% for laptops, and have no discernible increase for phones. We introduce ``associated keywords'', which are keywords related to the conversion journey, but are not the actual products, brands, or product categories, as cost effective keywords for targeting people who are committed to buying in the product category but not yet committed to a specific brand. Finally, we show that, if approximately 5 days of search history of each user are taken into consideration, the accuracy in estimating if a user is intending to convert increases 8 to 15 percentage points (depending on the product category) compared with the common current practice of just focusing on the current query for targeting. These findings show how advertising can be optimized and tailored to customers throughout the conversion journey. Our results provide a critical baseline and precursor to both wider targeting scope and increased contextual data to target, becoming available in the large language model-based search world.