Jonathan Walton | Head of Data Insight
Measuring the effectiveness of digital advertising is an increasingly difficult challenge, with an ever-growing number of channels and devices to monitor, technical challenges from advert viewability, and on the horizon, increased end-user privacy constraints on data matching.
There are a number of ways to evaluate the effectiveness of a digital campaign in isolation, but the process of simultaneously evaluating all active campaigns falls under a modelling approach known as “Digital Attribution”. This is an umbrella term for a range of techniques whose collective aim is to [fairly!] allocate credit to online media advertising for generating a sale, be that in the form of paid-for inventory such as search clicks, affiliate referrals, banner advertising, social media interactions or client hosted site content. We use this credit to evaluate return on investment, and it provides an indication of campaign strategy effectiveness.
A good example is our work we’ve done with our client Hillarys
Work on attribution began in late 2015 with a need to better understand the value of display advertising and generic PPC clicks. Measured using the simplest of attribution models, “last click”, there was consensus between the planning team and client that neither channel was being apportioned credit relative to the incremental sales they generated. As in any media plan, this put their investment at risk but moreover with a Data Management Platform in development understanding their impact was crucial.
Starting entirely from scratch, we began a full research project with help from Dr Stevens of Manchester University, reviewing and evaluating all relevant written material we could find – all in all some 50+ research papers and book excerpts. We tested, re-tested, and in most cases discounted many proposed approaches as generating unrealistic results, or results that simply didn’t resolve back in to “real world units” – i.e. conversion pounds and pence.
We distilled the research in to three potential solutions, ultimately selecting a new approach from academia for full development. For us, the challenges of working with an untested commercial model was offset by a highly flexible algorithm, a method that was easily understandable without the ‘black box’ restrictions of many data mining techniques and, most importantly, credible initial results.
There followed a series of monthly client meetings where the results were critiqued against realistic expectations. The output of these honest and open discussions led to us further innovation on an already new model: we used a simulation of media touchpoints as a connected “network” to credit drivers of Brand Clicks (i.e. prior steps), and in doing so brought much of our modelled channel contributions in line with Media-Mix Modelling results and an independent A/B (TrueLift) display test.
This reallocation revealed some notable differences in how generic campaigns drive subsequent brand clicks; the net result of which triggered changes to PPC bid strategy and our understanding of their value. From a display perspective, we could see we had been undervaluing activity by a factor of four through the use of a last click model.
We really wanted develop our understanding of the digital journey, and not only get the data but also act on it too. Working with MediaCom on this project we have made big strides forward to a more personalised journey for the customer and the results have been really strong. It is very exciting to be working with an agency that is driving things forward and looking to the future!
Richard Petrie – Head of Digital at Hillarys
We have extended our attribution work to several other clients, and in the process developed additional models more commonly utilised in the industry: game theory and statistical regression. This allows us direct comparisons between model results, and gives better appreciation of error margins where there is variance in allocation across model results.
If you’d like to find out more about how we use bespoke digital attribution modelling for our clients, please get in touch.