How to manage your own marketing resource allocation
“Half the money I spend on advertising is wasted, and the problem is that I do not know which half”
How big is the budgeting problem?
What is the right approach for estimating the impact of marketing on demand?
Why an optimisation model is needed?
The benefits of taking control
When it comes to how to best spend your marketing budget, you are probably not short of advice!
Agencies, media planners, product managers, finance directors and channel managers can all try to influence your decisions, and in some cases dogmatically state what is the ‘right’ approach. The question however is whether they are giving you the best, impartial, and objective advice, and how to know whether they are right or wrong?
There is no doubt that optimising marketing resource allocation is often a minefield for most CMOs, and their planning support team.
However this does not need to be the case. We expect to be able to save you a substantial chunk of your marketing budget, and deliver the same results.
This short paper describes the problem, and explains an approach that any company concerned about how they are spending their budget, on above and below the line marketing, can adopt.
As far as we know this approach is unique, in that it utilises two streams of historic marketing performance information when working out what the impact of marketing spend is on sales; both actual responses to campaigns that can be counted, such as source coded orders, or clicks, and models of how above and below the line channels impact sales at a macro level.
By not discarding any information, and using a pragmatic decision engine that forecasts into the future, we have a built a solution that could put you the marketer back in control.
What do we mean by ‘marketing resource allocation’?
Marketing resource allocation is the process of deciding how best to divide and spread a marketing budget across multiple activities, each of which will impact customer awareness and purchase behaviour.
Companies we deal with typically have between 100 and 1000 communications activities in their market budget, and the problem we are trying to help them solve is how to get the right spend in each row, to achieve their sales objectives most economically, and to live within business constraints that may for instance require a minimum amount spent in a particular channel, or perhaps a maximum in another channel, such as outbound telephony where resources are finite.
We also need to allow for a range of semi-fixed costs such as developing content, agency fees, database management, research, insight development, and sponsorship. Although these cannot be included in a budget optimisation model they do effectively reduce the amount that can be spent.
An optimal marketing resource allocation meets all your business objectives and constraints at the lowest possible cost.
Fig.1 UK advertising expenditure in 2013, Advertising Association
The total UK advertising spend, which excludes ‘invisible’ areas such as telemarketing, email, mobile or salesforce came to £17.6bn in 2013 according to the Advertising Association. It is forecast to reach £20bn by 2015.
Of the advertisers the largest spender was BskyB at £264m, and the 100th in the list, Saga, still managed to spend £23.7m.
But the budgeting problem is not just one of size, it is equally one of complexity. Once could say that when Lord Leverhulme (or it may in fact have been John Wanamaker) complained “Half the money I spend on advertising is wasted, and the problem is that I do not know which half’, he was having it comparatively easy.
In today’s world there are at least five main drivers of complexity:
- Rapid growth in the number of channels with recent additions being areas like app and voucher marketing
- The interactions between all the channels. It’s no longer a question of just looking at how TV halos may impact other channels
- The issue of how spend in one period impacts spend in subsequent periods, often known as the adstock effect
- The problem of attribution for direct campaigns when often the immediate impact is found in increasing the volume of natural search
- And, finally the wild card, the impact of creativity. Some argue that variations in the impact of ads due to creative differences is greater than variations caused by different levels of spend.
No approach to marketing resource allocation is going to get all of this right, but we can chart an approach which at least tackles the top four of these areas.
At the start of any marketing resource allocation process is the need to estimate the impact on demand of direct and non-direct advertising (the impact of direct also needs to be unravelled as it can be masked by recipients using the internet to browse or purchase).
Two academics Sunil Gupta and Thomas J Steenburgh wrote a paper for the Harvard Business School in 2008 in which they identified three main methods that are available to marketers for estimating the impact of nondirect advertising.
The first is what they call “Decision Calculus”; in layman’s terms this is managerial insight where the cumulative experience of people in the business is used to estimate the impact of marketing.
Next is experimentation. This approach is often overlooked where time is of the essence, and the luxury of embarking on a range of tests cannot be enjoyed, but in reality where external conditions are relatively stable, experiments undoubtedly provide quality hard evidence. Different experimental techniques have been used from picking matched cities and advertising differently in each, to using cable TV to field AB tests.
The third and perhaps most popular technique is econometrics. The econometrician will take a time series of marketing spend history, usually at channel by week level, and use this to predict variations in sales for a product or brand. The more sophisticated models include changes in external market conditions, the behaviour of competitors, and shifts in pricing. Econometrics needs significant variations in spend and sales over time; if all KPIs are flat it’s impossible to build a model.
Of these three we like econometrics, provided that it can take proper account of the impact of all off-line channels on natural search and web traffic. The approach we use builds a cluster of models to echo the multi-channel ecosystem (see Fig. 2).
Fig.2 Using a cluster of models to describe a typical marketing ecosystem
For direct marketing where a response is sought from targeted individuals, our approach is to develop saturation curves; these describe how, as more spend is put into a specific media or activity, its effectiveness declines.
Every channel’s performance will saturate as budgets increase. This saturation is effectively the cannibalisation of media, i.e. an individual who is already going to buy is still marketed to. After reviewing multiple clients’ data across multiple sectors we have seen that there are consistent and durable relationships within historic saturation curves.
The graph below illustrates how as the spend increases, the incremental value decreases:
Fig. 3 Illustrating how value per unit spend declines as spend increases
The graph also shows that whilst the two series are different, the overall relationship is similar. In this case we can see that the return in the second year is 20% better (across the board) than the first year. This could essentially be saying that after the first year’s learnings, processes are better and hence marketing is more efficient.
Having achieved a means of estimating demand, both for above the line advertising and below the line, these need to be combined into a single model in order to optimise marketing resource allocation looking forward.
The model has to select the best combination of spends across all channels; this will be in the light of understanding of how each unit of spend in any channel has the potential to impact the performance of all other channels.
The model also needs to accommodate specific business targets like minimum levels of recruits, or sales of specific products. There will also be constraints such as minimum spends in certain areas to reflect existing commitments to e.g.suppliers or partners.
The model will need to be able to build forward scenarios e.g. for next year’s budget, and show how the spend is to be allocated across the weeks, and months, in that period.
The model may need, within defined limits, to project spend that is outside the boundaries of any spend that has previously occurred; this means in effect extending curves beyond known parameters.
And finally the model may need to work back from business targets; a scenario may need to answer a question of how much budget is needed to hit a specific sales target.
All this goes to show how far we have moved outside the range of what human ‘Decision Calculus’ can normally be expected to achieve, and into an area where optimisation tools that can perform complex calculations are required. Particularly when the business is going to want multiple iterations of scenario projections until they have found to an acceptable outcome.
Gupta and Steenburgh sensibly define the requirements of a good model as being:
- Simple to understand
- Easy to control
- Adaptive to change
- Easy to communicate
We have developed a four step approach;
- Import campaign performance history and build saturation curves
- Build econometric models
- Set up the budget allocation software tool (our tool is called BAT)
- Set parameters and run optimised resource allocation scenarios
Where a company’s knowledge and experience is most needed is in setting the parameters within which the scenarios can be run. These include adjusting historic performance metrics to take account of new market conditions (what we earlier called Decision Calculus), fixing the overall budget, setting maxima and minima for specific areas of spend, and setting targets for e.g. sales or customer recruitment.
Fig 4 Importing campaign performance history and building saturation curves
That done our budget allocation tool BAT will take account of the campaign histories, saturation curves, and econometric effects, in conjunction with your parameters, to provide an optimised and detailed budget allocation plan.
It is your call as to whether you want external support in preparing the inputs for running scenarios. The software is designed to facilitate the import of campaign performance histories and the development of saturation curves. With training this can be managed internally.
You may have existing econometric models or you may need to have new ones built for you. We are able to develop them for you if required.
Taking control of your marketing budget allocation using BAT has a significant number of benefits:
- Objectivity (no risk of recommendations being biased by supplier self-interest)
- Credibility (with all input marketing metrics and outputs being auditable, you will get fewer challenges to the scale of the marketing budget)
- Efficiency (once set up, you can run new budget scenarios in minutes, and respond to changing market and competitor environments)
- ROI (you will be using science to optimise the return on your marketing investments, unhampered by vested interests)