Why You Should Use Algorithmic Approach To Budget Management
What is Budget Management?
Budget is the money a company is willing to set aside to accomplish its marketing objectives. Budget Management can be broken into 4 steps: forecasting, planning, execution and analysis.
This can be extremely complex when we consider multiple moving parts of the marketing engine – channels/sub-channels, Ad types, campaigns types, categories of products, transaction volumes, attribution models.
Since we are faced with this complexity, it is common today to have marketing budgets finalized at the start of each month. No real surprise that these digital marketing strategy sessions reflect the age-old “media planning” sessions.
Are your monthly marketing budgets “locked” by channels? Do your Budget Management decisions give an impression of being based on Roulette wheel outcome?
Read-on if you think machine learning is super cool and will help you deliver ROI against your spends.
Explore Vs. Exploit Trade-Off
What often gets missed in locked/pre-planned budgets is the experimentation. This is a classic explore vs. exploit trade-off problem.
On one hand, there could be a particular campaign (or targeting/ad type) that has potential to perform really well but hasn’t garnered enough performance since the budget allocated to it is too low.
On the other hand, increasing the budget might fail spectacularly if the campaign fails to perform.
What you need in such case is a smart experimentation which can minimize the experimentation cost and differentiate winners from losers sooner. In other words, there are potentially infinite ways to split your overall digital marketing budget between multiple campaigns and you need to choose correctly.
Let’s elaborate few approaches below:
You can use a simple probabilistic technique like simulated annealing. Technically, it is a used for finding approximate solution to an optimization problem.
In each iteration, the algorithm either chooses to stay at a particular state or move to a neighbouring state with some probability and re-calculate the objective value.
ROI/Volume Of Transactions Driven
Let us move from this technical jargon to our budget application. Say we have scored all launched marketing campaigns based on performance metric of our choice like volume of transactions driven or ROI on ad spends.
Based on this metric a newly launched campaign which brings lower volume has no chance of getting additional budget allocation. But we can instead launch an experiment (of increasing budget on a particular campaign) with some probability. The experiment iteration repeats and we have new scores.
The other approach is to take route of classic optimization problems. Linear programming in operations research for example, models payoffs from some decisions as an objective function subject to some constraints. In our case, we have payoffs defined as the score of each audience or marketing campaign.
The constraints are based on your maximum marketing budget and the spend capacity of each campaign. But as always, practice is much harder than theory.
The payoffs are not deterministic but stochastic. The cost (estimated spends) is not deterministic but stochastic and shows diminishing returns beyond a point. In other words, you can’t keep adding more budget to a particular group and see linear increase in performance.
To deliver the bad news further, the diminishing nature of returns means that we aren’t solving linear program anymore. Add to that the unknown scores for newly launch campaigns and you are looking at earning a Ph.D. in optimization algorithms and not running your marketing campaign.
However, some careful modelling, industry level benchmarks combined with Bayesian approach can help you make some progress.
Bottomline is, although there are many algorithms in machine learning or operations research literature which might look similar, the devil lies in the details when it comes to solving budgeting problems.
You need a deep understanding of the theory, industry experience to decide where a blind application of theory falters and then engineering expertise to run the optimization at scale correctly.
Previously I’ve explained what are the pros and cons of writing your own product recommender. This too falls in similar lines.
At OnlineSales.ai we have carefully tuned our budget optimization algorithms based on the experience of thousands of marketing campaigns and seen excellent result.
We are convinced that the algorithmic approach to budget management is far superior and scales really well. To find out more about our approach, connect with us!
- Writing Your Own Product Recommender? You Need To Read This First
- Cede Control – Accelerate with Marketing AI
- Integrated Campaigns Deliver a [Marketing] Super Bowl!
Rohit Kelkar | Harshad Saykhedkar