By Dr Thomas Oosthuizen, Global Consulting Director at Acceleration UK

The most important issue when dealing with programmatic buying is to be aware of algorithm bias. As data sources and points grow, this problem will decrease. But in most statistical techniques, the most salient data points often dominate and will continue to do so. We have to design to avoid this.

Pure common sense will tell us that as long as some marketers have more resources than others, "average" algorithms will benefit the large and jeopardise the small. This means that, even in algorithms, creative thinking is important. One size will not fit all.

Here are some tips to ensure maximum benefit from data-based algorithmically driven buying:

1. Know your product or service category

We need to understand whether our category is growing fast, maturing or declining? Different conditions and requirements come into being once a category develops from a fast-growing initial phase into a mature condition.

2. Is the market saturated? 

In a saturated market we need to look for niche segments that will enable wider expansion? To do so, we need to find out who these consumers are. Once we have done so, we need to understand how to expand the algorithm to enable us to identify and target them.

3. If growing, differentiation is less important. 

However, if the segment or category is not growing differentiation is key. We need to discover what are the signals that will enable the algorithm to detect these variations. Without this, our brand will simply fall into the trap of algorithm "same-ness", where less is, in fact, less.

4. Is our brand a leader or a challenger? 

Leading brands can leverage all the economies they can access. However, smaller brands need to work far harder at being different.

5. Is our brand properly differentiated? 

If so, how? This may include features, benefits, emotions, personality types, symbols, words, statements, slogans, colours, iconography and communities. It's clear that a small brand will have a vastly different profile than a smaller one. Hence, using the same algorithms a large brand uses is simply a waste of money. We then need to build in bias our differentiation "bias" so that it becomes a focused tool.

6. Are results declining over time? 

If so, why? Can a changed algorithm assist or does the problem lie outside of that? It's always tempting to constantly adjust algorithms, but we need to be aware that the problem may be something completely unrelated. Keeping an open mind is crucial when working at a granular level.

7. Can we segment algorithm groups? 

If so, can we learn more about what separates algorithms that are greater or lesser predictors of sales results? Understanding how they explain a category is very useful, particularly when this is a significant factor in planning exposure.

8. Can we build in "bias"?

Our algorithms need to contain enough granularity that we are able to fine-tune them to match whatever it is that differentiates us. By following trendsetters or up-weighting data from groups that demonstrate differences we can build this necessary "bias" into our algorithms.

9. Can we test different options and assess results? 

This is usually resource-dependent. The fewer resources we have, the more we will have to rely on testing to provide the data we need.

10. Can we expand diversity? 

If so, will the incrementally deeper and more creative messaging give us an above average return on investment?

This is by no means an exhaustive checklist, but by applying these tips we will be able to apply our programmatic buying algorithms more effectively.