Being used to sudden shifts doesn’t make managing them easier. After multiple tidal waves of disruption that have (mostly) swept away third-party cookies, added many new platforms and channels to activity remits, and repeatedly reconfigured consumer habits, marketers have come to expect the unexpected. But keeping performance steady is still a hard task. 

Resilience will obviously remain vital amid the latest swell of challenges. The ‘are we or aren’t we?’ questions around economic recession continue, alongside downsized spend estimates that could see teams struggling to drive greater results with even fewer resources. More than just continued hardiness, however, what marketers need are smarter ways of managing change and, ideally, getting ahead of it. 

To be specific, they must work on building better systems for identifying early warning signs of sliding effectiveness, in addition to emerging opportunities, instead of simply scrambling to respond after the fact. And an essential element of doing so is embracing valuable yet often overlooked elements of proactive analytics, especially anomaly detection.

Ghosts in the machine

Even in the current age of technology-powered marketing, anomaly detection might sound more like a sci-fi invention than an everyday tool. Not as complex as it seems, the process is about uncovering patterns in data outside the norm, using artificial intelligence (AI). 

The basics tend to involve setting machine learning algorithms to assess given datasets and 

spotlight outliers that don’t tally with what they expect to find. Trained on huge stores of data about past performance and consumer behaviour, algorithms run comparisons by harnessing knowledge of what counts as typical for particular types of audience segment, campaign, seasonal offers and more. With semi, fully or unsupervised options for analysis, marketers can leave labour-intensive evaluation to machines if they choose, or apply closer levels of control.

Great, so what does that mean?

For marketers, the ultimate outcome of smart analytics is simple. When any unusual patterns are detected, they receive instant warning signals or alerts. As assessment is often focused on monitoring pre-defined metrics, this means they get immediate insight into specific areas of inconsistent performance. For example, analysis may show a product explainer video that has enjoyed high viewing completion for months is now losing audiences mid-way, or highlight sudden spikes in click through rates (CTRs) for recently released display ads.

How does it benefit marketers?

In the days marketers only had a handful of campaigns to juggle, manually monitoring for abnormal data points was feasible, although not always precise. Vast expansion in the scope of modern communications has produced data that’s increasingly difficult to organise, let alone scrutinise for discrepancies. In fact, our research reveals most CMOs (99%) are using at least 10 data sources, compared to six or fewer just three years ago, while six in ten (67%) unsurprisingly admit to feeling overwhelmed by growing volumes of available data.

Where mistakes caused by human error were already a risk, it’s becoming much more likely that marketers drowning in data will miss crucial insights: stopping them from swiftly moving with developing changes and potentially leading to unchecked performance issues.

Outlier detection tools can go some way to help reduce overlooked discrepancies and speed up insight activation. With the ability to rapidly assess large-scale marketing data — including information about cross-channel spend, delivery, and interaction — analytical engines stand a better chance of accurately capturing all possible anomalies, enabling marketers to take swift, informed action. For instance, on top of directing budget away from ineffective efforts to limit wastage, they can identify which activities are over-performing and decide where adjustments should be made to improve both in-flight ROI and future outcomes.

It’s also worth noting that unusual events can indicate the beginning of bigger shifts. Keeping an eye on anomalies once detected will allow marketers to see whether certain patterns grow into fully-fledged trends; meaning they can observe, and respond to, the unfolding effects of external factors such as consumer living costs and market flections, as they happen.

What does adoption involve?

As with any form of analytics, there are crucial ingredients for effective use beyond the evaluation tech itself. Before adopting solutions with real-time detection abilities, users first need to ensure there is a solid foundation of unified, precise, and usable data to work from. 

Almost every marketer will be familiar with the data processing mantra of “garbage in, equals garbage out”, especially in terms of AI solutions. Whatever their assessment sophistication, intelligent analytics are only as reliable as the data feeding them. If data is flawed and fragmented, it’s highly probable that choices guided by the insights they produce will fail to hit the intended mark, or worse, result in negative impacts for brands. 

While the work needed to fine-tune data fundamentals will clearly depend on individual needs and circumstances, marketers can make a start by checking whether systems meet a basic list of criteria. Data stores need to be comprehensive collections of all relevant sources, accurate, and up to date. Analysis will also be a lot smoother, and reliable, when inflowing information is automatically integrated, cleansed, synced, and merged, ready for evaluation. 

Not only can anomaly detection assist with identifying any patterns but it can also provide critical intelligence to maximise performance, which can be replicated in future campaigns. Anomaly detection also tends to work better when companies already have a higher level of data maturity, including use of machine learning (ML) that helps them pinpoint where to focus detection analysis. For example, data mature companies know what KPIs are relevant for their business, how to track them and what KPIs are relevant for which dimension. The AI and ML algorithms can then help classify and analyse these two parameters, indicating how significant an outlier would be for this combination. For instance, a small outlier for an important KPI in the most important market should ring the alarm bells more than a huge outlier on a vanity metric.

Despite acute awareness that their remit has become more complicated, many marketers are hastily reacting to each new curve ball, instead of rolling with and turning fresh developments to their advantage. By providing immediate indicators of deviations from the norm, anomaly detection can enable marketers to see change coming and make proactive decisions about where, how, and when their approach should be adapted for better results. All of which means anomaly detection is a key agility asset for the modern analytical toolkit.