For many senior marketers, managing the customer recruitment budget is one of the biggest challenges we face. This white paper explains a new approach, based on our practical experiences in a number of different vertical markets; it relies on first understanding the whole multi-channel recruitment ecosystem, and then modelling how the different components act and interact to bring new customers on board. Only at this point can technology that runs and optimises budget scenarios be usefully introduced.
A BtoC recruitment budget ecosystem will often cover a number of channels, some of which are broadly independent, (like TV whose impact is not the consequence of spend in other channels) and others that are very dependant (like Google’s PPC which often acts more as a fulfilment channel). Some campaigns are direct to known consumers, whose responses can be tracked, and others are targeted but indirect, like digital display where you don’t know whom you are going to attract. Some are off-line and some are on-line, although we find this distinction less important than the way in which cause and effect is to be tracked for each channel.
Fig.1 Showing typical components in a recruitment marketing ecosystem
At Berry Thompson we have developed a four step process that will deliver the capability to run realistic recruitment budget scenarios across multiple channels taking account of the inter-reactions between them. This brief white paper explains that process, the inputs required and the outputs expected.
- The four step process
1 Map the current recruitment marketing ecosystem
Every organisation involved in recruitment marketing will build their own unique marketing ecosystem; our first task is to map this, and to understand which channels support which others, so that suitable models can be built.
The models need to show both the customer conversion journeys, where they can be traced, and also the influence pathways, for example from TV advertising to branded PPC.
Fig.2 In this fairly simple example ecosystem, the blue lines indicate influence, the red line negative influence, and the green lines conversion.
In some cases more advanced modelling can be used to achieve a more granular level of results.
2 Understand the currently available spend and results data cross all channels
Each channel or box in the ecosystem needs to be understood in terms of what it costs and what it delivers. The cost side is relatively straightforward, and where possible we like to measure cost in weeks as that provides more data points for any subsequent model development.
However each channel will also need to be broken down into smaller units which may represent media or other forms of tactic such as groups of Adwords. These sub-channels are the level at which the budget optimisation will eventually be developed. For instance where Adwords can be formed into groups, so CRM to ex-customers may be differentiated by time since the individual was a live customer.
Where the role of a channel is to support conversion, then provided the right kind of source coding or tracking is in place, it is relatively simple to attribute a value to a channel (subject of course to a debate about how long a timescale a customer’s value is being measured over).
But where the role of the channel is to influence, such as is often the case with TV when used for brand development, then its value can only be understood by the relationship over time between that activity and some dependant activity such as branded PPC and natural search.
3 Build channel and tactic level spend and value attribution models
3a Where we have a direct link between a channel activity and a sale, we can build what we call saturation curves. These curves describe a series of points generated by mapping spend and value derived from different campaigns in a channel. The near universal rule is that the more is spent, the worse the ROI derived from that spend.
Our approach is to take historic marketing campaigns in a particular channel, where there is a known spend and trackable response level and value generated, and to plot the cumulative response and value as spend is increased.
This has worked well for such diverse channels as for instance press, door-drops, and inserts.
Fig.3 an example conversion channel’s saturation curve
As this chart shows, the value generated tails off as spend increases, and even reaches a point where increasing the spend has little impact on the value generated.
(There is one exception, which is when dealing with Google generic PPC. The issue here is that the relationship between spend and value is dependent on the bidding mechanism. Namely as the budget increases Adwords tend to increase the position and cost per click (CPC) rather than simply gaining more impressions at the same level.
If one can master these Adwords controls, maintaining a consistent positon and CPC, then the spend-value relationship becomes linear as the probability of any impression being clicked (in the first impression instance) is the same.
Where we do however find an interesting relationship between spend and value is when we look at the returns from a higher position. Here we find that the ROI may often decline the higher the position as the CPC increases; however underlying contribution can fall as the click rate reduces).
3b Where the direct link does not exist we may enter the world of time series model development. Here we are aiming to describe the relationship between a series of spends, usually at a weekly level, and a series of outcomes such as the level of calls to a call-centre or visits to a website.
When looking at these relationships it’s important to look out for external influences; for instance competitor advertising may reduce the influence of one channel on another and seriously skew the results.
3c By this point we will have a set of historic marketing metrics showing the connection between spend and value, although value may be measured either directly by means of a saturation curve or indirectly by means of time series model.
Fig. 4a The impact of market share on sales
Once these factors have been identified the value can be re-weighted to account for this impact, making the underlying spend-value relationship clearer.
Fig. 4b Weighted value vs sales
4 Introducing technology that runs and optimises budget scenarios
Our objective is to arrive at a budget distribution where the ROI achieved across all channels when increasing spend in them is the same; at this point any redistribution of budget would have a negative effect.
Naturally there will be a budget, within which we are trying to optimise, and there may be a number of constraints, such as maximum or minimum spends in a particular channel or treatment. The maxima often occur because there are limits to how much one can spend in particular channels. For instance there is finite amount of press available, or direct marketing lists are of a certain size. Minima more often occur because of pre-existing contractual arrangements such as commitments to media owners.
The business may also have specific sales objectives for particular products, or the need to use up as much of its call centre resource as possible.
At Berry Thompson we have developed technology to run optimised budget scenarios like these, called BAT (standing for budget allocation tool).
With BAT it is possible to introduce the channel saturation curves discussed above, as well as the inter-channel influences from our time series models. It may also be necessary to adjust the saturation curves for such external influences as market conditions.
Within BAT a budget is built incrementally in small units we call bricks. Each brick is a particular amount of spend, in any channel. BAT picks in turn the most valuable bricks from any channel, but at the point of making the selection takes into account the inter-channel effects created by the bricks chosen to date. So if there has been no TV spend to date, the value of the first TV brick will be comparatively very high, and the TV brick becomes increasingly likely to be chosen (as the ROI from other bricks is starting to reduce as they are selected from higher positions in their saturation curves).
Bricks may represent spend across a year, or may represent the value of a spend in a particular month, where seasonal impacts are strong.
By building a budget up from zero in this fashion, BAT achieves a budget allocation optimisation, and can present this at a level of tactic much lower than that provided by an econometric model. We often find ourselves dealing with say six or seven channels, and perhaps a 100 lower level treatments which may represent specific media below that.
BAT also allows the user to quickly run different scenarios with different budgets, and constraints. The time to run a new scenario is rarely more than fifteen minutes.
Fig. 5 An example BAT output comparing two optimisation scenarios
- What are the benefits of introducing recruitment budget optimisation?
The primary benefit is financial, either from getting more value from the same budget, or from being able to reduce budget at the least cost. Every business will have its current methodology for allocating budget, and the amount of improvement that optimisation of the budget can bring will depend on the success or otherwise of the existing approach.
That said we expect to be able to improve results in the range of 5-15%, which when dealing with budgets of £10 to £100m, and values of £100m-£400m, can represent a useful financial benefit.
In addition budget planning and re-planning is often time critical; businesses face demands to achieve more or spend less and an urgent call to re-plan goes out. With BAT in place this can be achieved in a matter of minutes.
The third benefit of using an organised approach to budget allocation is that you can maintain a full audit trail. All inputs to BAT are visible, and any adjustments to take account of market conditions are recorded. The scenarios themselves come with documented constraints. In this way marketing budgets can be fully justified, and hence less likely to be demolished by corporate politics.