The company sells direct to consumers and has no retail presence. Enquiries are fulfilled either via the call centre, or via their web-site, or a combination of both
The channels used for recruitment are TV, Radio, and Google PPC. There is also an extensive DM campaign to warm prospects. The marketing budget is under £20m.
Digital attribution is last click based, as there is no individual level tracking.
The value of each sale is calculated at an individual level.
Marketing budgets have been set relatively arbitrarily in the past, with a mixture of spend levels and varying strategies within each channel.
The company has experienced great performance in the past, within a growing market. However, the market has been growing faster than the company; hence a shrinking market share.
Our aim was to fine tune the budget allocation to increase value for the same spend by 5%-15%, and to estimate the impact of increasing the overall budget by 10%.
Scope of work
Our remit was to build a structure to balance campaign performance across all channels. This meant that our approach needed to support budget recommendations at a micro level, whilst understanding macro impacts.
So we had to describe how to set a budget for say an Adwords campaign, as well as building cross channel impact models, for instance to predict branded PPC sales given TV spend levels.
We had one month to compose an initial optimised budget allocation, so that the company could implement the recommendations quickly.
Our approach was to:
- Have an initial cross (channel) department workshop to discuss our approach, and cover:
- Gaining an understanding of their marketing eco-system
- Data requirements
- A rough output structure
- Complete some initial analysis
- Arrange 1-2 mini workshops with each channel department:
- To review the findings of our analysis
- To finalise the level of modelling/outputs required
- To discuss how the recommendations would be implemented
- Deliver a first scenario*
*Our first scenario was backward looking; the next stage was to define a list of assumptions that replicate the changes from last year to this year. These assumptions can be included and update the outputs within a matter of minutes.
The client provided:
- read only access to Google AdWords and Google Analytics
- weekly spend data for TV and radio
- the contact history for their warm prospects
- a source coded response file (including value) *
*Source coding is not essential, however we were then able to operate within their reporting structure.
Given that individual level tracking was not set up, and that time was of the essence, we agreed to a simplified attribution structure:
Essentially we treated generic PPC as a totally “causative” channel, i.e. someone who Googles a generic term such as “XXXX” and clicks on the PPC link was mainly driven by that option (not by above the line media).
Whereas branded terms and natural search were regarded as being solely driven by above the line media.
There had been a mixture of DM activity to three main segments within their database; existing customer, previous customers, and previous applicants.
We reviewed the contact patterns for each segment and broke this down by a propensity model that was used for selecting customers. We could then clearly define the impact of contact density:
Above the line impact model
Once we had taken account of the attributed value, we could then model the impact of above the line media spend on branded PPC and direct traffic.
We did this in two stages:
- Locally to observe the impact of local radio
- Nationally, which observed the net impact of national spend
Interestingly, whilst we could directly attribute spend and value in the local model, in the national model it was skewed by the market share.
As part of this work we also considered other factors such as seasonality, or channel specific nuance’s like the costs per click vs budget vs impression share vs position in PPC.
We took these various inputs and imported them into our budget allocation tool (BAT). Our final presentation was an optimised marketing budget, layered with:
- the individual channel level models, with tactics being given a specific budget
- constraints such as maximum impression share in PPC, or TV channel spend
- versions of this with varying market share, and budget levels
As the channel managers found new limits, or gained new insight, these could be incorporated into BAT and a new optimisation run – re-calculating all the limits.
In our initial run we found that TV had previously had the best performance, however with a shrinking market share, direct attribution channels such as PPC were starting to outperform TV.
Local Radio had a mixture of really good and really bad locations, with more rural and poorer areas performing better.
PPC budgets were generally increased, however many were at their maximum impression share (hence could only improve position). It was also found that the more generic terms did not perform as well as niche terms (relating to this specific product).
Overall our % gain from the BAT scenarios run was dependant on market share and constraints; we produced optimisations with an uplift of between 5.9% and 17.4%. Our final champion model gave an uplift of 8.9%.
Berry Thompson completed this task on time, to budget, at the required level of detail, and producing returns within the estimated range (5%-15%).
As the market and unforeseen changes occur, assumptions and limits can be provided to Berry Thompson and new scenarios output in minutes.
Our solution enabled the channel managers to have a say in how the model was built and further support could be provided to support the implementation.
Given that BAT optimises the whole marketing budget, roughly £130 of additional value was added with every £1 spent on Berry Thompson services.