Until recently, digital marketers and ad agency professionals who were looking to make programmatic ad buys confronted a black box data scenario. Most aren’t engineers or data scientists, and they had to take a leap of faith and trust the data provider’s claims about data quality, reviewing results after implementation — and after the purchase was already made.
But what should marketers and agencies look for in a data provider? How can they determine which provider offers the most accurate, transparent solution? Here are some questions to ask:
How is the data gathered?
Is it through direct observation of every user, or is it inferred data, where behavioral patterns are detected in a small group of users and then extrapolated out for larger groups? If the data is inferred, accuracy is highly dependent on the size of the measured group — so it’s important to check group size when assessing providers. But keep in mind that whatever the size, inferred data always involves a decline in accuracy when extrapolated out. And don’t forget that when data is modelled into segments, predictions will be based on predictions rather than real information. This dynamic exponentially increases the risk that the data won’t perform.
It’s a good idea to ask common-sense questions that allow you to assess the strength of data across the funnel, looking beyond simple demographics to factor in transactions, metadata tracking and other signals that more accurately predict purchasing intent. Skimlinks captures 15 billion shopping intent signals from a network of 1.5 million publisher domains and 20,000 merchants every day. By applying machine learning and enriching analysis in their product intelligence layer, Skimlinks understand the taxonomy and metadata of 100 million product references and links. They use this information to build high-converting audience segments based on the products and brands users are likely to purchase, enabling more effective display, social, and video campaigns.
What type of data is collected?
Next on the list is to find out what kind of data is gathered. Categories can include clicks, links, metadata, page content, search terms, brands and products, pricing information, transaction occurrence, date and time. The more kinds of data are gathered, the more raw material predictive models will have to work with, which can significantly improve accuracy. If only a few types of data are collected — for example, just impressions or clicks — there will be limited information that can be used to cross-check predictions or enhance user profiles. In this scenario, the risk is that overly simplistic and inaccurate user profiles will be generated.
Skimlinks collects and analyzes data and detects patterns across multiple publishers and merchants to accurately predict purchase behaviors. For instance, the combination of one user visiting 10 pages across five different websites might be identified as a pattern that indicates an interest in making a purchase in the next week. No single publisher could produce the data Skimlinks accesses through its network of 1.5 million domains, but publisher information is just one part of the signal data. Skimlinks also analyzes the data sourced from the 20,000 merchants in its network, including pricing information, order value, and purchase history.
In doing so, Skimlinks combines signals from the entire retail ecosystem.
How is the data validated?
Another critical capability to look for when evaluating data providers is the ability to validate predictions in practice. For example, any provider who claims their segments will drive conversions should capture transaction data to confirm that the purchase takes place. Without transaction data, it’s not possible to validate the value proposition.
Skimlinks has a programmatic audience targeting service that helps advertisers target users according to where they are in the buying cycle. Predictions are made using contextual, product and pricing data, and they are validated using transaction information. Users are tracked to check if they made the expected purchase, and the machine learning system that creates segments is continuously trained based on this information. That helps buyers avoid a scenario in which they target consumers who may have researched a product they can’t afford or have no real intent to purchase. The result is better segment performance.
Digital marketers and agencies that engage in programmatic advertising must choose the right data provider to optimize their cost per thousand impression (CPM) or cost per action (CPA) rates. The rate of growth in the programmatic advertising and data-driven marketing sectors can make it difficult to know how to choose the right data provider. But by applying these three common-sense questions when assessing a data provider’s value proposition, digital marketers and agencies can open up the black box and find the right data mix.