Recent forecasts for programmatic media spend estimate that it is set to grow hugely over the next four years, according to the recently published 'Guide to Machine Learning' from Intelligent Optimisations (IO).
According to a report from Magna Global, global programmatic advertising hit 42% of total display related media spend in 2014, up from 33% in 2013. Magna further forecasts global programmatic spend to reach 48% by the end of 2015. And longer-term predictions are strong too with Magna Global's forecast for the next four years being for market growth to reach US$53bn by 2018.
And the IAB in the UK predicts the volume of UK digital ad spend traded programmatically could grow to 60-75% market share of total digital display advertising by 2017.
The growth in programmatic buying has been fuelled by advertisers need to reach new customers more cost effectively and to deliver the right message to those users in real-time. As more and more companies seek to take advantage of the benefits offered by programmatic, this guide highlights some of the challenges inherent in current techniques used for buying media programmatically and looks ahead to explain why innovative new technologies, drawing on the concepts of machine learning and data science, are helping to overcome those challenges.
What is Programmatic Buying?
Programmatic media buying has been defined as the use of software to purchase digital advertising. This is an evolution from the traditional process which involved RFPs, human negotiations and manual insertion orders. Basically, it's using software to buy ads.
To extend that definition further, programmatic buying offers the potential for advertisers to automate media buying throughout a customer's entire path-to-purchase engagement with the advertisers brand and across all addressable media channels - social, display, mobile and video.
The key benefits for advertisers from buying digital advertising programmatically are; more efficient media purchasing and improved response to the advert, leading to higher return on investment (ROI) and a more relevant advertising experience for the consumer.
Overall, buying advertising programmatically produces greater digital marketing performance by enabling the real-time optimisation of every customer interaction with an ad campaign. Advertisers can drive their marketing campaigns through the purchase funnel and across all channels and devices.
What is Real-Time Bidding (RTB)?
RTB is a function of programmatic buying - specifically, it is the buying of individual ad impressions through real-time auctions that occur in the time it takes a webpage to load. RTB's key benefit is providing advertisers with access to real-time inventory at scale, and most importantly, at a price those advertisers want to pay.
By using programmatic RTB and leveraging data and analytics to inform targeting and media optimisation, programmatic advertising enables brands to better marshal and manage user engagement at every point in the customer buying journey. Ultimately, real-time bidding allows advertisers to move from generic segment-based ad buying to a fine-grained approach of, nearly, one-to-one marketing, but at scale.
Scalability issues in Real-Time Bidding
One of the most widely used techniques when buying media programmatically is retargeting. Retargeting has proven to be an extremely successful targeting technique, highly efficient in converting users who are already 'in-market' (tangibly demonstrating some level of interest) for an advertisers products or services. For this reason, retargeting often forms the core targeting capability provided by many DSPs. However, there are challenges to be mindful of when using retargeting, primarily in attribution and scalability.
Retargeting is effective because it is applied to users who have already demonstrated a level of interest in the advertisers products (i.e. are at the bottom end of the sales funnel). In order to work, retargeting solutions rely (almost) totally on traffic generated by an advertisers own marketing activity.
The conundrum for an advertiser is how many visitors to the advertiser's website would have converted anyway, without the retargeting activity? That is, how many sales are attributed to retargeting campaigns even though the user would have returned to the site and purchased without the additional ad messaging?
Intelligent Optimisations believes that it is relatively straightforward to chase conversions that are likely to happen on their own and that this often occurs at the expense of generating genuine new business.
As retargeting works only on users who have already visited the advertiser's website, there are inherent scale limitations. Place any frequency and recency filters on consumers visiting your website and, very quickly, you will be only reaching a small number of users. By not placing recency or frequency filters on your retargeting campaign you risk reaching an audience which is no longer engaged with your product. It's a real conundrum.
Audience Segmentation
One option to address the problem of scale is through the injection of the advertiser's first party data. Effective use of first party (including offline) data comes down to whether the marketer: (1) has all their data in a single accessible format (many advertisers still have their data siloed, so it can be difficult to make the data available for targeting digitally), and (2) whether that data has enough relevance or scale to be truly useful. As stated above, one of the biggest limitations with techniques such as retargeting is the inherent lack of size of audience available for targeting.
The use of third party data is an option but, by definition, this data is available from third parties and so is ubiquitous. The ready availability of this data to anyone willing to pay for it erodes the uniqueness and therefore value of its use. Third party data can be useful in assisting an advertiser to put some definition on an audience but because of the generic aspect to the data there is little competitive differentiation to be gained from using it.
One of the reasons why so many DSPs emphasize the use of third party data is that the DSP does not have historical data (the DSPs or the advertisers) to draw upon. This is because many DSPs have tended not to store historical data after a certain period of time due to the cost implications for their data storage.
Intelligent Optimisations believes that the only practical way for advertisers to achieve both reach and relevancy when using programmatic bidding is to work with a solution capable of managing huge volumes (millions) of data points in real-time in order to prospect for new, responsive audiences and to then relentlessly optimise campaign performance to achieve the desired ROI. The few solutions currently in the market with the capability to provide these prospecting services are using Machine Learning.
Rise of the Machines?
Machine learning may be defined as a branch of artificial intelligence that studies and learns from data. Some algorithms recognise patterns in data and based on those patterns, make predictions. These predictions are the core of real-time decisioning systems and are integrated into a wide range of software applications, such as real-time buying. Increasingly, forward looking companies are making strategic and tactical advertising decisions by leveraging machine learning, coupling Big Data technologies and data science.
With the number of touch points (devices, different channels, and different times of the day) now available to advertisers, interactions occurring across multiple dimensions in real-time (for example, IO processes data in 40+ dimensions) are constantly changing. The volume of data now available and the corresponding speed of learning needed to keep pace and make decisions in real-time means that effective targeting is firmly beyond the realm of human assessment.
By uniting programmatic bidding and machine learning, advertisers can create campaigns which learn and evolve; continually enhancing performance by building always on what's working in the campaign and eliminating what's not.
For advertisers, the major advantage of machine learning is the opportunity to prospect relevant new audiences beyond the scope of retargeting, by analysing millions of data attributes over multiple dimensions and advertise to these audiences in real-time. The real value of machine learning is the ability to automatically identify the right blend of attributes between too few and too many. In other words, to find the sweet spot for optimum audience identification.
Intelligent Optimisations, for example, uses machine learning to work outside of the sales funnel. We work at 'pre-awareness' levels for brands, to drive net new customers to our clients products; increasing sales, improving engagement and driving ROI. We call this capability True Prospecting.
True Prospecting
The emerging opportunity for advertisers, and indeed for programmatic advertising, is in using machine learning (and prospecting solutions like IO's) to create and sustain brand engagement.
To do this, advertisers will need to go after the tougher customers, the ones who aren't aware or already engaged with their brand, the ones who need convincing. This is where the strengths of combining machine learning and programmatic really come into their own.
The full guide has been made available for free download from IO's web site - click here (free registration required).
More Info: |