In order to tackle these questions Adroit reviewed the available data, including cost information and an exercise was undertaken to align financial information to the actual period when marketing activity took place. A longitudinal dataset was assembled with inputs at as granular level as possible, with external data such as Nielsen spend data and external economic factors added.
One of the fundraising areas we were asked to assist in is to help optimise channel spend to support acquisition of new supporters, and to understand how different media channel spend interacts and what “halo” contribution each media makes often in combination with other factors e.g. emergency response appeals and events, competitor spend, economic factors etc. Whilst this is often seen as the realm of standard conventional econometric approaches e.g. ARIMA/time-series or Bayesian approaches, in fact often insufficient or incomplete data exists to deploy these approaches
To report on the findings Adroit developed the “Media Trend Visualiser” – a simple but highly effective application in Excel.
The application allowed analysis and visualisation of the dataset, with sponsorships mapped overtime against a combination of up to 3 factors, and uses the coefficient of determination (R2) to look at the amount of explained variation. Different datasets are allocated to a normalised distribution to allow cross-comparisons to be made. Significance on volumes of sponsorships generated by a media combination of up to 3 types with separate weightings and “lags” (so allowing for genuine media lag or known differences between accounting recording of cost and actual usage).
A further “analyse” function within the tool uses the gradient to look at expected volumes. A “scenario” feature allows the user to project the number of expected acquisitions from a given spend, providing a more accurate estimation for planning purposes.
There were a number of project challenges in interpreting multi-media channel usage and impact attribution. These include:
- Ensuring users with different levels of expertise were “au fait” with the approach, and understood what could be derived and where the interpretation needed to start. As with many scenario tools creating the ideal ‘real-world’ scenario is sometimes a mix of science and art – a balance between empirical data and marketing instinct & experience.
- Assembling and determining the dataset that has sufficient information at a granular level e.g. media schedules by day, web visits, social activity for a reasonable lengthy period to judge impacts, is no easy task. The dataset did stretch back as far as 10 years for some items, but a much shorter length for other channels e.g. many digital channels were shorter. Cost data in particular is typically not usable directly from a finance system which records invoice dates, so these need to be married back to the actual media schedules (n.b. often tv slots change/get pulled so intended schedules are not the same as actual).
- Activity over time is not consistent. i.e. the client had “bursts” of TV spend, or other channels and the application needed to allow the analyst to isolate periods where there was an appropriate combination of channels, and compare this to other periods. (n.b. we developed a way to do this and also to shunt the dataset to allow for different analysis points and also respective “lag”.