Attribution is absolutely critical to the sustainability of the display market. It underpins everything we do in the industry. Implementing an attribution model that rewards all touch points en route to conversion is a difficult and costly process. Agencies are investing a lot financial resource and manpower to make sure their client spend is returning the best possible results - as well as ensuring the right media is being probably rewarded. Ciaran McConaghy is Group Head of Data & Analytics at Havas Digital, and works on attribution using the agency's own proprietary modelling tool, Artemis. Here McConaghy speaks about attribution in Display, moving beyond the CTR metric, and the sensible agency approach to attributing conversion credits in media buys.

Can you give some detail on how Artemis addresses the area of attribution for Havas advertising clients?

CMC: Attribution is core to what we do at Havas. We make it available to all of our digital clients. It’s really the starting point for all customer insight we do for clients.

For media, each attribution model will reflect the advertiser’s business goals, so it will vary by advertiser, by vertical. We capture all touch-points in our clients’ digital campaigns. We measure all campaign data and look at the digital journey paths for both converters and non-converters. We consider a number of different factors including; the position of each event in the journey path, whether the Search click was a branded keyword, whether it was high or low volume, it’s position on the SERP, the recency of the interaction, the density of interactions and so forth.

The key thing for us is that each attribution model calculates the incremental value of each impression, click and keyword for a campaign. It uses machine-learned insight rather than an intuitive model.

We also incorporate offline media spend, GRPs and client sales data to give an integrated picture of channel performance and the value of particular customer segments. For us the wider attribution piece is not just about customer acquisition but also customer conversion, retention and the role played by each media channel.

Can attribution really deliver on transparency in terms of how media performs as per the goals of a campaign?

CMC: Outside of blind media network buys, attribution is absolutely necessary to show the transparency in conversion contribution. All agencies should be evaluating a publisher’s share of voice against the share of campaign cost to see which publishers are under-performing and which are punching above their weight.

In terms of Media & Data buys attribution won’t, on its own, solve the transparency issue. Agencies need to normalise the taxonomies we get from publishers, data exchanges and data aggregators to see the true value of the data we’re buying. For example, what are the differences between “auto intenders” from Publisher A to “auto in-market” from Publisher B. It’s apples and oranges, to a degree. We can model from behavioural and contextual data of course but there is a real opportunity for publishers to facilitate attribution take-up by giving more visibility into how their segments are constructed.

What is Artemis doing that is so different from similar offerings from the likes of WPP and Publicis?

CMC: We were one of the first players in the market to start addressing the attribution question when we starting developing Artemis over 10 years ago. Back then and until
the last couple of years attribution was a nice-to-have for agencies but now it is really a pre-requisite for all digital campaigns.

The same could be said for the quality and accuracy of the attribution models them-selves. Many agencies are still using subjective intuition for their attribution weightings rather than machine-based models. This will become more of a differentiator going forward as the number of media channels, platforms and volume of data grows.

With Artemis attribution is machine-learned. It is bespoke to clients and is responsive, not just to each client campaign, but also to time - for example the attribution model that is optimal for a travel client in early January may be very different to the one that is applied to a late summer campaign.

Artemis also incorporates customer offline sales data and offline spends to calculate the role and value of digital. For example we re-attribute value to digital campaigns based on their effect on offline sales.

By including offline media spend and actual customer sales data into client’s attribution models we are taking a much wider view of the value of all acquisition channels – not just in the number of touch-points and sales they deliver but also the quality of a customer (LTV, retention costs etc).
Again this will be something that agencies will need to do more of in future.

Are agencies and clients often overlooking the role of attribution in the media buying process?

CMC: Some are, without a doubt. They’re still set in the “impressions, clicks, conversions” mentality of if my spend increases and so do my sales then the campaign is performing. Agencies should re-attribute conversion credit not just at a channel and site level but also at a creative and keyword level. The difficulty for some is in doing this is the heavy lifting involved.

Re-attributed conversion credit should appear in campaign reporting as a default so media planners and buyers can engage their analytics teams as soon as they see a difference between the sales a publisher is awarded (last ad credit) and what they should have been awarded (re-attributed credit)

But this has its challenges when it comes to trading. We need to help clients sell-in attribution methodologies to their senior leadership teams as a more accurate measure of campaign performance. Similarly, more agency trading discussions need to incorporate attribution models to give a truer picture of the ROI of each publisher in the plan.

Should all advertisers be attributing some of their display budget to touch points higher up the sales funnel? Is there too much emphasis on the last impression/click wins?

CMC: Around 93-95% of all digital touch-points occur before the last ad. So relying on that last ad alone as an indicator of how media is working could be viewed as being a mixture of guesswork and luck. We have seen time after time that mid to upper funnel activity can have a significant effect on the last touch-point. A great example is brand keywords which are quite often the last ad in a path-to-conversion. Around 60% of all PPC clicks are a single branded keyword. If we look at just this last ad it suggests that consumers are very brand aware despite the fact that large numbers use price comparison sites, affiliates or generic keywords earlier in their paths-to-conversion.

How would you suggest the display industry moves away from this approach?

CMC: All sides have a part to play. More advertisers should accept that the contribution of display to a sale goes beyond the last ad and CTR. This should be agreed with their agencies, pre-campaign, with the business goals and attribution models being put in place beforehand.

Agencies need to be able to show clients the contribution of each part of their media plan whether it’s re-targeting buys or richer media placements such as video. The journey-to-conversion shouldn’t be a shown as a strictly linear path but more along the lines of which audiences did you reach and at what stage of consideration where they.

This is, of course, dependent on publishers giving more visibility into what audiences they can deliver.

There’s been a lot of debate recently around the PV window in display – and how effective it is in attributing the conversion to the right party? Is this working for advertisers? Is display pulling its weight in the path to conversion?

CMC: Too much of the post-view debate has been around the worries advertisers have about awarding the whole CPA credit on just one impression (last ad). Clearly they’re right to say the credit should not go to just one ad. But to say there is no value at all in post-view conversions is throwing the baby out with the bathwater.

If we look at all consumer touch-points in a digital campaign we see on a consistent basis that some publishers do a better job than others in driving conversions, be they post-view or post-click. Equally some conversions are better than others in terms of transaction revenue. Clearly, some impressions work better than others. This has definitely been reinforced by the growth in ad-verification and brand duration analysis where we can tell what % of an individual impression a consumer saw and for how long.

Post-view windows need to be set in agreement with the advertiser and they must consider their business and campaign goals.

Artemis allows us to track post-views over the cookie lifetime and then set limits (the post-view cookie window) based on actual observed behaviour. This is bespoke to each client.

Can you give some overview of the roles media, site and CRM data play in the attribution model?

CMC: For us attribution is not just about which media channel should pick up the conversion credit. It should be viewed in the wider ‘joined up’ context of; a) seeing which media drove the customer to the site (acquisition) b) which elements of the site contributed to a sale (conversion) and c) how did we deploy the client’s own CRM data to retain and up-sell to these customers. It’s only when you connect these data pots that you can begin to evaluate the net value of the media you’ve bought and the quality of the customer you’ve acquired.

Are there any case studies you can share on successful attribution modelling developed for clients?

CMC: Yes, we have quite a few but one of the most interesting ones was for a travel client who was facing a difficult business challenge of improving media efficiency across multiple channels whilst also sustaining consumer revenue. The client was making cuts to their media spend of around 40%. Their targets remained the same despite these cuts.

Through Artemis we looked at user paths to conversion to calculate the correct attribution of conversion credit and the optimal media mix.

As a result we smashed the revenue target by a 333%. The ROI increased by 24% which was amazing, considering the economic climate. In the face of a significantly reduced budget we made an improvement of 55%, YoY.

We have also done a number of attribution case studies across a number of sectors including Utilities, Finance, FMCG, Gaming, Mobile Telecoms and Broadcasting, using both online and offline data.