At the start of 2022 I sat down to write my MBA research paper on how artificial intelligence could be applied to solve a variety of common marketing problems. The term Chat GPT meant nothing to 99% of people and web3 was the keyword in every trend-hopping digital person’s LinkedIn bio. I felt ahead of the curve.
1 year and 1 research paper later, the world of A.I and marketing feel very different, unfortunately not due to the seismic impact of my paper. But despite the recent frenzy of commentary surrounding Chat GPT, there has been little focus on what I found to be the most powerful application of artificial intelligence for marketers, its ability to predict consumer behaviour.
The introduction of machine learning into marketing offers the promise of unlocking greater value from customer data by improving the speed, affordability and precision of predictions relating to future behaviour.
However, there is currently a knowledge gap between marketing leaders and artificial intelligence tools. In addition, most marketing professionals struggle to effectively evaluate these technology solutions or have a meaningful dialogue with data science teams.
This article aims to bridge that knowledge gap by providing marketers with a clear overview of how A.I can be deployed, specifically to improve customer lifetime value and provide a set of recommendations on how they can build the capabilities required within their organisations which enable them to successfully utilise the technology.
- Why CLV is more important than ever.
- How marketers should think about artificial intelligence in relation to CLV.
- A.I applications for CLV growth.
- Key Considerations for marketing managers adopting Artificial Intelligence.
Why CLV is more important than ever.
The last decade has seen the democratisation of e-commerce and digital technology, lowering the barriers to entry and leading to an explosion of E-commerce focused DTC brands. Shopify’s 2022 State of E-commerce Report concluded that “advertising costs are skyrocketing across platforms meaning digital advertising costs are eating up marketing budgets”.
Rises in media costs combining with increased competition and stagnating consumer demand in many categories puts huge pressure on cost-per-acquisition, brands must react by finding effective ways to improve their CLV in order to keep their business models viable
However, a recent Criteo study found that only 36% of marketers surveyed were “completely aware of the term CLV and its connotations” and only 24% of respondents felt their company was monitoring CLV effectively. This presents a significant opportunity for those brands who do understand the drivers of their CLV and take effective steps to improve it.
If a brand can achieve a higher CLV than its direct competitors it can afford to pay a higher CAC, allowing it to outbid its competition in auctions and grow at a faster rate or enjoy greater profitability.
How to think about artificial intelligence in relation to CLV.
“At its simplest form, artificial intelligence is a field, which combines computer science and robust datasets, to enable problem-solving.” IBM
Most AI-based solutions can be classified into the following categories of machine learning; supervised and unsupervised machine learning.
- Supervised learning: The dataset being used has been pre-labelled and classified to allow the algorithm to see how accurate its performance is. Supervised learning includes solving classification problems and regression models.
- Unsupervised learning: The raw dataset being used is unlabeled and the algorithm identifies patterns and relationships within the data without help from users. (Tamir, 2020). Unsupervised learning is used to identify hidden patterns within datasets, this includes clustering and association which are two types of techniques used in recommendation systems.
In both cases the goal is to make predictions, and this is a useful way to think about the technology. At their core A.I & ML models are simply making better and faster predictions about unknown outcomes than humans and previous statistical modelling techniques are able to do and in turn, marketers can use these predictions to make better decisions and build better customer experiences.
As described in their book Prediction Machines “the new wave of artificial intelligence does not actually bring us intelligence but instead a critical component of intelligence—prediction.” (Agrawal, Gans and Goldfarb, 2018)
A.I applications for CLV growth.
Artificial intelligence can be applied to a broad range of marketing problems, but the most common marketing use cases revolve around segmentation, recommendation systems and attribution.
- Predictive CLV
To gain an exact view of a customer’s value you need them to complete their relationship with the company, although this is precise it takes the duration of the relationship to realise and is of limited value to real-time retention and acquisition efforts. Businesses should therefore make predictions about the future lifetime value of customers as early as possible and the more accurate and fast these predictions are the better.
This is where machine learning can be applied, historical statistical models take into account simpler data structures and require multiple observations to make accurate predictions. As shown below, ML models are able to take in more data, including available customer data combined with granular behavioural data and use this to make an immediate prediction, these ML models will also continuously adapt to changes in real-time as more historical training data is fed back into the model for reinforcement.
This allows businesses to create future customer value segments, such as likely “high Value” and “Low Value” segments immediately and take real-time actions through CRM. In practical terms, this can mean focusing more resources, providing a superior experience or avoiding giving away unnecessary margins.
It also allows brands to pass those high-value groups into ad platforms to acquire more people who are similar to those customers, this is a powerful way of improving CLV in the long term as it is much easier to acquire good quality customers than change behaviour later on. This process is visualised below.
2. Segmentation & Targeting
Segmentation is commonly conducted to group customers based on demographic, geographic, behavioural or value-based attributes. This is useful but limits the segmentation to predetermined criteria. Unsupervised machine learning methods such as k-means clustering allow computer algorithms to find patterns in the data and segment customers into distinct groups based on all available data points.
This is done by using Principal Component Analysis which reduces large numbers of observations into two distinct variables that can be plotted on a two-dimensional graph. A K-means clustering algorithm can then be applied to data to cluster the population into groups that are distinctly different to one another. An example of this is visualised below where x1 and x2 represent measures of difference, and the customers have been clustered into 3 segments.
This approach takes into account much more data points than normal segmentation models, the number of clusters can be adjusted and the model re-ran in a matter of minutes. This allows marketers to discover previously hidden trends within customer or market data and allows for new ways of segmenting an audience to discover customer groups with high levels of similarity that would otherwise be impossible to identify.
3. Churn Detection
The goal of churn detection is to predict which customers are likely to churn so that marketers can; identify high-risk customers, identify customer pain points that lead to customer churn and identify strategies/methods that can be applied to reduce churn. (John, 2021)
The signals customers emit ahead of departure are often buried in the noise of overall customer activity. Preventing a customer from leaving requires marketers to have some amount of advanced notice which is obtained through the careful examination of large volumes of historical data, something for which machine learning models are ideally suited. (Smith, 2020)
AI performs exceptionally well at pattern recognition and can detect patterns humans may not have considered. Machine Learning models can identify patterns of behaviour preceding churn, effectively learning what early warning indicators to look out for. (dotData, 2022)
The typical output of these models can be a churn risk score and then high-risk customers can be segmented based on their behaviour, marketers can then step in with targeted retention campaigns to the right people.
4. Recommendation Engines & Personalisation
The best-known and most widely adopted application of A.I in digital marketing is recommendation engines and personalisation tools. Product recommendations have been around long before A.I, these systems previously used sales data to display popular products and surface products that were often bought together.
The application of A.I has flipped this from being a product-centric system to a user-centric system where a combination of previously discussed methods, clustering and association rules are combined to deliver highly personalised messages or product recommendations created for a specific individual. Google Recommendations AI visually represents this below:
A well-documented example of this was IKEA switching to Google’s A.I based recommendations, which resulted in a 30% increase in CTR across recommendations and a 5% increase in revenue per session.
Although each of these 4 applications of A.I use different types of algorithms or processes to improve CLV there are commonalities which make all of them superior to human decision-making or previous statistical solutions, primarily through their ability to generate better and faster predictions:
- They remove natural bias in human decision-making.
- They can process vast amounts of data and complexity.
- They process this data and complexity incredibly quickly.
- They are adaptive rather than rule-based.
Key Considerations for marketing managers adopting Artificial Intelligence.
- Ensure solid CRM strategies are in place before looking to deploy A.I.
CMOs need to ensure teams have a good understanding of their current CLV and have clear strategies in place to improve before looking to adopt a technology solution. Teams should not try to run before they can walk, adopting A.I is an expensive and complex process which requires a lot of resources, marketers should ensure they are doing the basics correctly and have a solid existing foundation on which to build. Teams should have high-quality, useable customer data in place, be familiar with segmentation, cross-sell, upsell and personalisation and have the ability to produce enough creative assets to serve different messages to different customers.
2. Understand how A.I can be applied to improve CLV.
Client-side knowledge of A.I is currently one of the main barriers to adoption, so CMOs must take responsibility here and ensure their knowledge is up to date in order to lead organisational change from the top and have informed strategic discussions on the role of A.I, its capabilities, requirements and limitations within a marketing context.
3. Analyse your marketing operations and prioritise specific decisions where A.I. can help.
Attempting to build A.I into general marketing processes is the wrong approach. CMOs should start by decomposing their customer journeys and identify specific decisions where A.I can be used to predict the outcomes better than a human or existing statistical model. CMOs must ensure the value of improving a decision through better prediction leads to clear commercial gain, in most cases the incremental value will be created by improving customer experience which in turn leads to higher spending or by avoiding margin erosion through improving targeting before discounts are given away.
4. Use the A.I Canvas for project scoping & planning.
The A.I canvas is designed to help structure thinking and ensure that all the necessary requirements are identified before embarking on deploying A.I. The A.I canvas is a highly practical and useful way to start the planning process and can be easily communicated with your team. The idea of decomposing decisions in this way was strongly supported by the experts I spoke to during the research phase who possessed prior experience of introducing A.I in organisations.
5. Invest in a Customer Data Platform.
Unifying your customer data into a single usable platform is an essential enabler to unlocking value from first-party data. A customer data platform is the simplest way to connect web, CRM and ad platform data to build a marketer-managed single customer view which can be connected to A.I applications in order to enhance decision-making. It puts customer data into the hands of marketing teams with direct connections to ad platforms and A.I technology in a way that empowers marketers to take action.
6. Evaluate internal strengths and set a clear path for progression
The digital skills gap was one of the most prominent barriers to adoption in both primary and secondary research. CMOs need to do an honest appraisal of the digital skills in their team, invest in retaining digital talent and upskilling to provide employees with relevant skills before they are required. As a minimum aim, employees should have enough knowledge to manage third-party specialists where specialist knowledge is required and be able to have a meaningful dialogue around data & technology. As well as individual skills, CMOs should also analyse where their teams sit on both the analytics maturity curve & A.I maturity curve. These frameworks help CMOs articulate where they sit in terms of capabilities and show the common paths forward.
7. Be prepared to hire and champion third-party specialists.
A.I is still a specialism and most DTC brands are unlikely to possess the internal talent required to manage the end-to-end process of A.I adoption. Leaders should, therefore, be prepared to bring in third-party specialist knowledge to support internal teams. This should be done on a project basis with the aim to upskill employees. The CMO role here should be to champion the process and gain buy-in from other internal stakeholders whilst ensuring that as much knowledge is imparted from the specialists to their internal team over time.
8. Consider the implications & opportunities at a strategic level
Although starting with a limited use case is suggested, the adoption of A.I may have wider impacts across the marketing department so should be discussed at a strategic level. Wider strategic impacts include creating efficiencies by replacing or augmenting human roles, creating a greater need for complementary skills such as judgement and creativity, and changing the value of existing resources i.e a greater value placed on data acquisition due to the increased value of customer data. A.I should therefore be a topic on the agenda for the C-suite with the CMO being capable of grappling with and presenting the trade-offs of adopting A.I to key stakeholders.
9. Reduce risks by ensuring human oversight
The introduction of new technology comes with inherent risks, the algorithms and automation that A.I adoption introduced through A.I present risks to brand image and reputation and this is something which CMOs must manage. Human oversight is required to monitor the output of ML models, to ensure they make logical sense and are in keeping with brand guidelines. Once the technology has been adopted, it requires constant oversight from marketing professionals, especially with regard to segmentation and targeting decisions and any generative content produced.
Final Thoughts
Reflecting on the topic there are two main takeaways, the first is that despite the low level of current adoption, there is a clear use case here where A.I can be applied and make a significant impact. After in-depth interviews with experts who are developing this technology, I see the potential in the technology and am left with no doubt that A.I will be transformational in marketing operations over the next decade.
The second takeaway which has been striking is that the technology already exists and has already been packaged into viable SaaS products, the barrier to entry for DTC brands now is simply the low level of in-house knowledge, the industry is still trying to catch up with the explosion of new digital technology over the past decade and it feels like currently, the DTC industry is not ready for the next wave of technological disruption and the opportunity it presents.
