Four ways to unlock marketing return on investment with machine learning

Despite the emphasis placed on data-driven marketing over the past decade, 54% of senior marketing practitioners reported in a 2020 Gartner survey that their organisations’ investment in marketing analytics hasn’t had the expected impact. They cited poor data quality and unactionable results, as well as the difficulty of tying data and analytics efforts to return on investment (ROI).

For analytics to drive measurable outcomes, the right conditions need to be in place so that an organisation’s data scientists and analysts can query data quickly and produce results. The first step is unifying customer data in a scalable platform that supports an up-to-date, 360° view of customers and ensures that teams can simultaneously get access and query data in near real time instead of being at the mercy of nightly data loads. Only then can investments in data science realise their full potential.

Here are four ways that data science can drive marketing ROI after establishing a single source of truth for customer data:

1. Personalise offers and reduce churn with more precise customer segmentation 

Data scientists can group customers and prospects in audience segments based on known attributes and behavioural similarities, so that marketing teams can deliver personalised content, experiences and offers to increase sales and conversions. High-value customers who regularly purchase products or services are one target segment, as are at-risk customers who may be in danger of churning out. The more characteristics used to segment customers, the more targeted campaigns can be developed. But segmentation becomes more difficult as the variables increase to dozens or even hundreds of attributes.

This is where data scientists can deploy clustering, a method of grouping objects by common traits. Clustering uses machine learning to identify how different data points (in this case, consumers) are related, and then groups them based on those relationships or similarities.   Unlike multiple regression models, where data scientists use a predictor variable to build models that predict a target variable, clustering uses several variables to discover statistically significant groupings by trial and error of each combination of variables. For example, a clustering algorithm may identify seemingly unrelated correlations, such as a group of consumers who are more likely to purchase a new pair of shoes based not only on their shoe purchase history, but also because they streamed the latest episode of a new hit show.

This enables marketers to tailor messaging that drives increased engagement and sales and deliver that message to the right audience, at just the right time, using the most effective distribution channels.

2. Uplift conversions with near real-time campaign optimisation  

By analysing data in a low-latency platform, marketers can see how campaigns are performing in near real time and filter that view by audience segment. From there, messaging and content can be adjusted quickly to optimise performance. For example, if a shoe retailer’s high-value customers convert at a higher rate when shown red sneakers compared with white or blue ones during A/B testing, the digital marketing team can optimise the creative to use red shoes when targeting that segment. What seems like a small change for a single digital campaign can translate into thousands (or even tens of thousands) of rand in incremental revenue. In addition to dynamically changing content and messaging based on what specific audiences respond to, ad placements and bids can also be optimised in near real time, making media spend more efficient.

3. Optimise spend with attribution modelling

By determining how much credit for sales or conversions should be assigned to each media channel, marketers can optimise ad spend, improve personalisation, and more. Many marketers launch their attribution initiatives with relatively simple first-touch or last-touch models, which assign 100% of credit for sales and conversions to the first or last touchpoint on the customer journey. There are even-spread models that assign equal weight to every known touchpoint, and W-shaped models that assign more credit to touches that are deemed to have the greatest impact on a purchasing decision, but these still treat audiences as if they are all alike.

Since no two customer journeys are exactly the same, an attribution model that assigns fixed weights to touchpoints won’t account for the nuances of how individuals learn about products and services and engage with your brand. Marketers should eventually aim to have their data science teams build custom multi-touch attribution (MTA) models that are dynamic and evolve over time as more is learned about how specific audience segments can be moved down the marketing funnel.

4. Increase customer lifetime value with predictive analytics 

By modelling historical data, marketing organisations can identify the likelihood of future outcomes and act on those findings. Specific applications include lookalike modelling, which identifies prospects who behave like another segment (for example, prospects who resemble your highest-value customers) to increase your customer base and overall revenue. Another example is using affinity-scoring models, which gauge people’s interests based on their browsing history, to improve product recommendations, which can drive incremental sales and greater customer lifetime value. These predictions can become more accurate over time through ongoing iteration of machine learning.

To have a flourishing data science practice that delivers maximum impact, marketers need to attract and retain top talent. In addition to seeking data scientists with technical prowess, marketers should also aim to hire data scientists who are good business communicators, as they will need to work cross-functionally with nontechnical teams.

Marketers should invest in increasing data literacy across their organisation, which helps ensure that technical and nontechnical employees alike recognise the value of data and are working toward the common goal of a 360° view of the customer. This is a change management exercise that requires a detailed assessment of existing levels of data literacy and proactive communication about data goals for the future.

Julien Alteirac is the regional vice-president, Northern Europe and emerging markets, at Snowflake