Finding and Optimizing Purchase Behavior Using Custom Attribution Models
We recently had help from the Attribution feature in Acquisio to try to outsmart the market in a specific and very competitive industry: New Condos Marketplace. The campaign was targeted to Montreal and was promoting a new condo project.
The challenge we had was to build a new strategic way to buy media on a performance basis given a medium to long-term buying cycle (Condos purchase can take from six months to a year). As most of the players in the industry are simply investing their entire budget in Adwords Search and Adwords Remarketing creating a massive increase in CPC, we decided to reverse engineer the market trend and find a path to invest more wisely across search and display with the data we had on hand.
First finding: Potential Correlation between our Keyword List and our Site List.
We started our campaign using a large keyword list, which was providing a CTR below 1% and a poor conversion rate. Our display advertising initiatives (sites list) was generating a CTR of 0.06% and a poor conversion rate as well. As soon as we refined the keywords to only keep the most relevant ones (CTR and Conv. Rate), we saw a major increase in the display advertising CTR (up to 0.3%), and our conversion rate increased by 15% in both channels. The 15% increased in Conv. Rate was foreseeable in search but surprising in display.
In a competitive environment like the Condo Industry, we found a potential complementary correlation with two alternating ways of prospecting new clientelle. We also found using first click attribution that search engine and real estate portals were acting in the industry as introducer or influencer (whether the user initiated a search and then visited a portal or vice-versa). In both cases, we were getting an impression, a click, and a conversion.
This was the starting point to create our attribution model. We pursued our experiment by creating two separate remarketing campaigns: a site wide remarketing strategy, which was linked to the display (site list) advertising campaign, and a search retargeting strategy, which was directly linked to the search campaign. We were fortunate to have amazing creatives to help us get conversions.
With two complementary entrance funnels, which have great conversion rates, the remarketing initiative simply increased the campaign’s overall performance but also enabled us to finally build what we wanted to do from the beginning.
We were able to monitor different types of buyers in the market and associate a specific cost-per-conversion for each. Imagine how easy it is to pitch this story to a contractor who only talks about sales and profit. You have buyer “A” which costs this amount of dollars and buyer “B” which has this cost of acquisition.
Second finding: Creation of two types of buyers based on the data. The Ready-to-Buy
The ready-to-buy customers will search for a specific keyword and click or see our ads in the search results. Based on our experiment, he will also most likely visit a local portal that presents different new condo projects. If our banner ad is displayed there, we get the click and the conversion. It works both ways, if we first display a banner ad, we may get an impression or a click. Then, when the user pursue his shopping process and performs a search, we get the click and the conversion. It is similar to remarketing – however, the first brand interaction point might be an impression (whether in search or display), which is impossible to track using a pixel. Using a precise and relevant site list has helped us to uncover this problem. In this case, we are able to bid less for keywords and placements by finding the break-even point for which our model does not work anymore. We keep an eye on metrics like average position & CTR for search as well as win rate & CTR for display. The conversion rate becomes highly correlated with those 4 metrics.
The long shot was the second buying model that we found. The initial process is the same as the Ready-To-Buy, but instead, this one needs remarketing to convert. In order to invest wisely in remarketing, we experimented two models. If the click was coming from search, the buyer would be sent to the Search Retargeting Audience seeing banners designed based on its search query. If the click had come from a banner, we would send the user directly to the general site remarketing. With creative optimization and frequency adjustment, we were able to increase conversion rate from this model.
Obviously, the cost per conversion was more expensive in the second model. However, since the added touch point was remarketing, we are talking about a fairly small increase.
Overall, this experiment has enabled us to be more strategic in the buying process, attribute a cost of acquisition based on behavior, and built a smarter way to invest. Not all clients or industries are similar to this case. However, with more media channels, you can simply find data relationships between channels and build experiments to measure the possibility of scaling those relationships.