Ebiquity’s Chief Client Officer, Andrew Challier, and Head of International Effectiveness, Mike Campbell, reflect on i-com, the “Cannes Lions” of data analytics.
The Harvard Business Review – no less – has declared ‘data scientist’ to be the sexiest job in the 21st century. The organisers of i-com have definitely received this memo. For at the end of last month, the leading lights of data science in the media and marketing community came together for the 8th annual i-com global summit for marketing data and measurement. The event was held in suitably glittering venues across the city of Porto. Move over, Cannes Lions, there’s a more evidence-based party in town.
Ebiquity took a strong and experienced team to i-com, all of whom took part in the event; as presenters, panel members and, most excitingly of all, as part of the i-com data science hackathon. Over the course of the summit, we found our thinking very much aligned with the cutting edge of an industry looking for smarter, more rigorous, and better use of data – both in terms of our approach to digital attribution and the quality of our in-house data scientists.
While most of the conference delegates were still arriving, looking forward to sitting in comfort and being refreshed at the breaks by fine European fare, teams of data scientists were already working away in rooms with no air conditioning or natural light for 24 hours, as they wrestled with the data challenges posed in the i-com data science hackathon. Two tasks were set, one by Intel and the other by Unilever, to two tiers of data scientists: experienced (Tier 1) and less-so (Tier 2). Ebiquity fielded a senior team in Tier 1, addressing the Intel challenge, looking at the impact of advertising and Twitter on sales in the U.S. Our four-strong team comprised Ed Dickerson, Miles Wood, Patrik Sahlin, and Wojtek Kostelecki. They had never worked together as a team, and it was the first time that we – or they – had entered this kind of hackathon. But they did not disappoint.
Presented with the data sets and the objective – of determining the relationship between advertising, social media, and sales – the team got to work. Unlike some of the other teams, our team was able to multi-task on the various jobs involved: pulling the data together to make sense of it, modelling the data and analyzing it, generating insights, and building these insights into a coherent and compelling narrative for client storytelling. So, none of the Ebiquity team was waiting around for – say – a specialist modeller to do their job before – say – looking to extract insights. This lack of a conveyor belt approach meant that, if one member of the team went down an unproductive ‘rabbit hole’, another could take over and help them out.
This way of working enabled our team to produce very specific commercial insights for Intel. We were the only team to be able to isolate the impact of Intel’s advertising from PC manufacturers’ advertising which also promotes “Intel inside”. This enabled Intel to understand the actual ROI of their own advertising separately from the added benefits and synergy of PC manufacturers’ advertising.. In this way, they could see how separate and combined campaigns reinforced each other’s messages.
The hackathon culminated with all finalists presenting their findings to the entire conference and also the judging panel. Ebiquity’s team won both the audience vote within their tier and the support of the Intel representatives on the judging panel, although in the end we came a very close second to Analytic Partners – to whom many congratulations are due. But to do so well – and to win the sponsor’s approval – in our first time entering a hackathon was by any measure a very strong performance.
Understanding multi-channel attribution
Presenting at conferences can be a lottery, particularly when you’re speaking with other industry leaders. The risk is that they say something totally different from what you were planning to say, or that they merely deliver the same message. Both authors of this article found themselves in the latter camp, speaking between US-based Sequent and Dentsu Aegis. But we took this violent agreement as a positive sign, and that we’re reassuringly at the cutting edge of thinking on how multi-channel attribution should be conducted and applied.
After years in which digital-only attribution solutions overstated the contribution of digital media to multi-channel attribution models and market mix modelling was deemed unable to account for the granularity of contributions that digital can make, the industry consensus now points towards a blended approach that uses a combination of both techniques.
We explained that the right blend clearly depends on the type of business you are, the type of physical and digital assets you have, and the extent to which you both advertise and trade offline, online, or both. For purely digital businesses using only digital advertising, a pure-play digital attribution solution is likely to be sufficient. But many – particularly retail – businesses have significant offline elements in their consumer journeys; physical stores as well as online retail operations, traditional as well as digital advertising. For these businesses, a hybrid approach of both digital attribution and econometrics is required to attribute inputs (marketing) to outcomes (sales). And for purely bricks and mortar businesses who just use traditional media, then market mix modelling still generates the best ROI insights to help shape future investment.
We also explained that many of the pure-play digital attribution solutions have fallen short because they’ve applied arbitrary rules to cookie level data, when the right – and more productive – approach is to let the data speak and decide what has caused what. “Rules-based” attribution – imposing rules to force an answer out of the data – leads to a predetermined, often wildly inaccurate answer. That’s why we’ve built our own multi-channel attribution solution that scales the overall impact of digital media in an econometric model. From there, we dive into the detail of digital, using an attribution model that’s not rules-based. In this way, we allow the data to tell its own story, giving appropriate weighting to the value of each touchpoint.
We closed our session by emphasizing that the maths and modelling is only part of the task; to make multi-channel attribution work requires smart engagement, explanation, and storytelling that senior marketers can understand and champion. Marketers get to be senior because of good commercial instincts and by making strong commercial decisions. If they don’t understand the data and conclusions they’re presented with – as many CMOs have failed to do with pure-play digital attribution solutions – they’ll ignore them and use and trust something they do understand.
Talking to delegates as part of panels and during breaks and at i-com’s receptions, we heard several examples of senior marketers who were just unable to trust black-box digital attribution solutions and were actively looking for white-box or clear-box answers; transparent solutions that aren’t rules-based and that let the data decide. It seems like we – and the industry – are onto something.
The ability to tame and mobilize huge data sets in order to optimize marketing performance is maturing fast, and the modellers, agencies, and marketers taking part in i-com’s latest global summit gave plenty of evidence that the HBR got it right in singling out data science at this century’s sexiest job. The only significant discordant note sounded in a symphony of otherwise harmonious music was May 2018’s arrival of the General Data Protection Regulation from the European Union and the threat that GDPR poses to organizations looking to make intelligent use of first-party data for personalization and targeting purposes. Doubtless this will be a dominant theme of i-com 2018.
We – and our hackathon team – can’t wait.
Ebiquity’s approach to attribution is detailed out in our new Viewpoint paper, Understanding Multi-Channel Attribution. Download your copy here http://bit.ly/2pFxRn7