The 2016 Programmatic Handbook features John Goulding and Niek Sonneveld
The Programmatic Handbook by the IAB brings together the smartest digital minds to contribute to this handbook.
The latest handbook features our very own John Goulding, Global Product Director, and Niek Sonneveld, Inventory Manager.
John shared his thoughts on ‘Data Management: Audience Analytics and Attribution Modelling’ and here’s what he had to say-
Audience analytics is vital for any advertiser looking to understand who their customers are, what inﬂuences their buying decisions and how and where they can ﬁnd more people like them. Data Management Platforms (DMPs) can provide part of the answer – they provide the ability to on-board data such as on-site behaviours and CRM data, to layer this with 3rd party audiences, segment users based on historic behaviours they’ve displayed, and then to output this to a range of audience buying platforms such as DSPs. Solutions that allow advertisers to go beyond segmentation, and which on-board more diverse datasets relating to consumer behaviour, can lead to better new customer acquisition, deeper insight on consumers and answers to complex questions.
- What are typical paths my customers take to make a purchase? Integration of device graph data combined with high-scale browsing data can begin to join online journeys made across separate devices.
- How can I acquire new customers? Lookalike modelling using classiﬁcation techniques to identify common trends across a group of customers is a powerful prospecting tool.
- What are the key moments that inﬂuence my customers? It’s not always about ‘who?’ it’s also about ‘when?’ and ‘why?’ Real-time data, on factors such as weather, social trends or economic factors can unlock answers here.
- What kind of places do my customers visit? Finally, there’s also that question of ‘where?’ GPS data gathered from mobile devices can link online proﬁles to physical location visits. For this, partners with access to relevant data, and possessing data science capabilities, are necessary in order to unlock true value from audience analytics.
Attribution Modelling example
A marketer is running a campaign for a CPG brand wherein the mobile ads are giving good conversions. Using standard last click-view attribution model, they would conclude that search and mobile ads are driving conversions and the brand should invest more in them. But, the reality could be that users are watching an ad on TV, they get reminded of the product while going home through billboards or seeing a display ad while browsing online, and then they click on ads on their mobiles to buy the product. How does the marketer ﬁnd out to what extent each marketing touch point is inﬂuencing their target audience? Data-driven attribution modelling could help map the value of each channel and create an effective strategy based on the whole picture. Such models help attribute due credit to each touch point involved in a consumer journey and help a marketer understand how effective each channel is. There are no standard data-driven models and each brand will require a bespoke model according to their necessities, channels and the data available. Better the data management, better would be custom attribution model.
A couple of examples to showcase the data-driven approach could be-
- A simple time and position based hybrid model where we can look at the time lag of the ads to conversion and position of the ads before the user conversion.
- A more complex approach can be a Hidden Markov Model(HMM) which ﬂags a user’s state based on their engagement with the brand. Let’s say, x number of ads seen via multiple channels can revive a user’s interest towards the brand from the dormant state ultimately making him purchase the product, in this case, we cannot give credit of the inﬂuence to just one channel. Every channel contributed in some way to make sure the user is engaged and active to make the purchase of the product.
Niek, on the other hand, provided information on ‘Three Steps Towards Minimising Fraud’.
The subject of ad fraud has received a lot of attention over the past few months, with global fraud rates quoted as high as 60% in some research papers. Everyone in the industry is affected by fraud, whether that be buyers, publishers or exchanges. To eliminate ad fraud completely is unrealistic, similar to how eliminating all crime would be unrealistic, however, it is possible to minimise fraud from your buying strategy to low single digits by combining analytics, technology, and common sense.
Programmatic advertising has the great advantage of being able to track every single bid request and impression, allowing you to perform advanced analytics on the data you are collecting. With this data you are able to determine the average number of bid requests per user, per IP address, per location, etc., and are quickly able to spot outliers. For example, a domain that sends thousands of bid requests for the same user within a matter of seconds, suggests that either this domain or user is suspicious. You can then dive deeper into the data to ﬁnd the exact issue and where this is coming from. Another way to spot fraud would be to analyse mouse movement (number of movements, the time between the ad loading and clicking on it, where in the ad the user clicks) to detect anomalies. These are just examples, and with the amount of data that is being collected, a data science team can have a ﬁeld day applying machine learning or time series analysis techniques.
There are many technology vendors that provide both buy and sell side with data on brand safety and fraud, as well as provide pre- and post-bid solutions to block fraud. By choosing the right vendor you not only ensure that you minimise your exposure to ad fraud, but you also enrich the data you collect on each impression, by adding valuable data on an IP, user or domain level with regards to the type of fraud, viewability and brand safety. This can then further fuel your own analysis to ﬁnd the source(s) of ad fraud and you can take steps accordingly. Each technology provider has its own advantages and disadvantages, so make sure to spend enough research on determining what is the right ﬁt for your business. Some vendors focus more on analytics, whereas others are better at blocking fraud in real time, so a combination of vendors will prevent you from bidding on fraudulent impressions, inform you of fraud in real time, as well as block your ads from loading on unwanted sites. Vendors in the UK are currently undergoing the ﬁrst round of fraud certiﬁcation with a view to being certiﬁed by Q1 2017 under the JICWEBS Anti-Fraud Good Practice Principles.
Finally, when buying media, common sense goes a long way. Premium inventory costs money, so don’t expect to be able to buy a premium domain for a CPM that is unrealistically low. At the same time, if premium inventory is suddenly sold by a completely unknown entity, make sure to check that you are actually buying what you think you are buying. Combining the above three should result in a solid framework for tackling ad fraud, and will make sure you get the most out of programmatic.
View the complete IAB Programmatic Handbook here.