5 Market Segmentation Challenges That Are About to Be Solved

5 Market Segmentation Challenges That Are About to Be Solved

By Emily Webber

It’s one thing to create a segment, but understanding why you’re segmenting, if it’s effective, and measuring it—these are all challenges we face.

Machine Learning and Market Segmentation

Market segmentation is one of the most basic arms of business strategy. Firms bundle customers to understand their preferences, manage relationships with them, improve product and service offerings, and assess risk. From media to telecom, retail to financial services, industries today are heavily invested in static segmentation and yet there are major problems with most segmentation approaches. Classic statistical analysis requires months of work, resulting in discrete customer groups that are too outdated to match the dynamic body of people they are supposed to represent. Furthermore, the segments often fail in granularity, leading to market portions that closely resemble each other. This lack of precision means that firms are unable to tailor messaging that is relevant and compelling enough to specific groups of customers; the bottom line is that the true customer context of why someone is compelled to respond or purchase is often left wholly out of the picture. Without rich granularity, precision, context, and dynamism our digital age, firms are not able to meet their customers’ changing needs.

Having spent the last fifteen years investing in a strong digital presence, marketers have an abundance of data sitting in their laps with shockingly few tools that give them the granularity they must have to communicate effectively with their customers. They already think about their audience in terms of segments, but what marketers need is a technology that will raise these segments to the next level, equipping them with the full power of AI. See our blog post Similarity Learning Cracks the Code of Explainable AI, demonstrating how similarity based machine learning provides unmatched capacity for explainable AI in marketing.

simMachine’s recently commissioned a Forrester Consulting study1 that revealed while most of the 155 responders in retail, media, financial services, and telecom employ a type of segmentation, 98% of firms agree that static segmentation is no longer adequate, and 100% of firms recognize the opportunity cost of not evolving to its more granular, precision-based, inheritor: Dynamic Predictive Segmentation.1 For each of the issues outlined by Forrester, Dynamic Predictive Segmentation provides a unique ability to overcome and provide value back to the business specialist. Dynamic Predictive Segmentation (DPS) is about to unleash a new breed of marketing campaigns with granularity and specificity that could only evolve as the by-product of computational power and human ingenuity.

DPS is an approach that segments customers based on propensities to take a specific action grouped by the most important shared characteristics revealed for each segment. Segments and their associated characteristics change dynamically in real-time based on new data from customer interactions.

#1 : Static Segments Fail to Provide Enough Actionable Detail

As the chart suggests, the overwhelming majority of segmentation users in industry agree that static methods are no longer adequate. The first issue outlined by the Forrester Consulting study is the failure of segmentation to provide actionable detail. Imagine spending months investing in your teams’ segment build, only to discover that the analysts weren’t able to incorporate the very data elements that you care the most about. Or even worse, they included the data elements, but the configuration is lost in the intricate modeling at such a deep level that there is no take-away, no insight upon which you can make a business decision.

There are two main reasons for this failing of detail: first is machinery, and second is transparency. A segmentation system that is driven by manual statistical modeling simply will not be able to capture the thousands of fields and rows that originate from multiple data sources. On tasks that can be automated, machines will out-perform humans. They can capture more complexity, reduce it to a digestible format, and their methods are easily repeatable. Furthermore, a system that does rely on thousands of data fields in an automated fashion, (i.e. a machine learning pipeline), typically does not provide the transparency that is necessary to drive decision-making. These problems are overcome by Dynamic Prediction Segmentation.

Capable of ingesting thousands of columns and dynamically weighting the factors for each segment, DPS is a new type of machine learning modeling that not only provides transparency, it relies on it. You can upload a data file with just under 2000 columns, and with DPS build a predictive model in several minutes. The dynamically weighted factors behind this model are piped into a “sunburst” cluster with interactive visualization capabilities to provide unmatched granularity for business decisions.

# 2: Static Segments Aren’t Updated Based on Changing Customer Behaviors

The second key problem identified for static segments is their lack of ability to be updated based on changing customer behavior. In a world where financial transactions are measured in milliseconds, and smartphone customers change their plans when their provider is no longer capable of handling latency requests, keeping business strategy up to date and market awareness i tact is a must-have. The manual creation time of static segments is the primary barrier to having timely, relevant information at the nexus of decision-making.

Imagine having all of the strength of your current segmentation team, but with results delivered in minutes instead of months. With one pull of customer history inside DPS, you can generate predictive models and segments with incredible granularity and precision. These segments are easy to update; with an automated segment builder you have the capacity to meet your customer’s behavioral changes in real time.

# 3: Static Segments Lack Precision & Are Too Undifferentiated

The static segments built manually by a team of statisticians will simply never be able to capture the precision, granularity, and dynamism of real customer life. Models built by humans simply are not able to integrate thousands of data elements in a robust and reliable way.

Now let’s take a look at dynamic predictive segments in comparison. Here is the result of an algorithm that is classifying customer data into people who are likely or not likely to want to purchase Apple cell phone accessories. Individual predictions are grouped into these dynamic predictive segments. This produces extremely granular clusters of people, that you can select and use to design your campaign.

Below we can see that our cell phone analyst is selecting the largest cluster of people who are likely to purchase cell phone accessories. What’s interesting to note here is that based on the data set used, this segment of people on average is reluctant to adopt new technologies. They subscribe to traditional TV  services, they do not like online banking, they do not rely on peer accommodations, and they do not like wearable devices. Furthermore, they are not likely to use a smartphone. This segment, it turns out, are individuals who typically do not adopt new technologies for personal use, but who love to provide the latest and greatest to their children or grandchildren. Equipped with these insights, our sales analyst now knows more about who to target for his upcoming product release, and what relevant message and offer to make for this segment.

DPS offers a technology that most marketers find exciting when they see it for the first time. The fact of the matter is that in real life your customers are as granular as these segments.  Just think about how different you are from your neighbor. You have the same zip code, you probably have a similar income range, and you might even shop at the same grocery stores. You probably have completely different music preferences, vacation history, domestic lives, and leisure activities. And even more importantly you have different motivations and interests in terms of why you purchase a product vs. why your neighbor purchases the same product.  This granularity reveals what makes you unique.  DPS models the granularity, highlights and contours of unique groups within your customer base, with a predictive edge unmatched on the market today. You can use this technology to interact with your customers in their own way, knowing what channels they like, what their online preferences are, and what devices they have a natural affinity towards.

# 4: Unable to Tailor Messaging to Specific Groups of Customers

With static segmentation, a major issue is the lack of capacity to tailor messaging and outreach to groups of customers. A core advantage of DPS is the ability to reveal the preferences of each customer group, closing the communication gap and enabling highly specific communication plans.

Dynamic Predictive Segmentation gives you relevant information upfront so that you can design campaigns, creatives, and schedule your product release. On top of looking at sets of predictions, you’re actually getting clusters of dynamically weighted factors behind those predictions that give you everything you need to know about your segment to design the perfect creative for them. Clustering is another word for putting them into a segment, but in this case the differentiator is a trained algorithm, not a biased person.

# 5 : Lack of Transparency Loses Customer Context

Machine learning is a data-driven technology. It works in this case by joining customer data with product purchase history, a process known as labeling, and feeding it into an algorithm that learns to discreetly differentiate customers.  The field of designing these algorithms, perfecting, optimizing, and applying them is machine learning. In machine learning, we are teaching a computer to classify the difference between objects, and we’re using data in order to do that. In the example of our customer, the classification is whether or not they purchased your product.

After you’ve trained your classification system, you can use it to find new customers. This means giving your trained algorithm examples of new people, new rows and columns of data without your product purchase history label, and asking it to score each person for their propensity to buy. 

When you use a model to generate predictions, you are essentially feeding new records into your algorithm, and it will automatically generate a number between 0 and 1 telling you how likely this person is to purchase your product. After you generate predictions, you can limit your potential customer pool down to people who are already likely to be very interested in your product. Once you already have the raw predictions, limiting them is as easy as sorting a column in Excel.

The next step is to group your predictions. In marketing, this is known as segmentation. Traditionally segmentation analysis meant looking at small sets of descriptive variables and using summary statistics to determine how one variable changes in relation to another. While still relevant, summary statistics do not drive home the insights necessary to compel a buyer to make a purchase in the context of their motivations and interests.  Instead of descriptive factors, the drivers of predicted behavior, or predictive variables must be understood.

What marketers need to drive contextually relevant messaging is a technology that is just as transparent as it is predictive. Dynamic Predictive Segmentation is right now the only tool on the market capable of revealing the weighted predictive factors behind each prediction which are the motivations and drivers of each customers’ predicted behavior.  The word “dynamic” here refers to the fact that DPS can dynamically weight and display all of the factors behind each prediction based on all current data provided. This means that every time a unique prediction is made, DPS has the capacity to provide the local feature importance driving that prediction.

For specialists in machine learning, this is somewhat of aberration. A field that is so strongly driven by theory and methodological implementation cannot afford to place too much importance on context. As a consequence, the heavy-duty lifting in data science happens mostly in algorithmic implementation, experiment design, and pipeline optimization. These are transferable skills, appetizing to the mathematician and logical-minded, that can be applied in many correlation scenarios. While this does lead to progress in the mechanics of prediction, the losers in the game of theory development are business users.

For business users, features are their bread and butter. They live and die on the context of their customers. The success of companies weighs in the balance of how much they understand and sympathize with the real-world setting of their most valued customers, and their ability to turn this understanding into actionable product that makes their customers’ lives easier. Dynamic Predictive Segmentation relies upon a lens of similarity that can reveal the context behind every prediction. This means that on top of being able to find new customers, DPS explains the factors that provide context and clarity for every prediction.

Dynamic Predictive Segmentation Is About to Disrupt Marketing

Marketing is due for a massive overhaul. The amount of money wasted on irrelevant marketing is astonishing, and the prevalence of data should serve as an antidote. The missing factor here is simply the lack of a technology with the maturity and granularity necessary to scale to the business needs of the day. Dynamic Predictive Segmentation provides the transparency necessary to bring contextual relevancy to life, and transform marketing.

Even today, most companies wait for a 2- 3 month manual process of updating segments. They’ll submit a request to analysts who spend hours manually digging through databases, sifting through summary statistics, and compiling extensive reports. These reports are still useful, but DPS is able to provide far greater breadth and depth within minutes. The platform built by simMachines provides your team with Explainable AI applications that build models in minutes, frequently outperform the accuracy and precision of other vendors, and generates predictions at 1 million rows per second. DPS is better segmentation, hundreds of times faster, that continuously learns, and dynamically updates in real time is the future of marketing.

1 The Customer Moment With Dynamic Predictive Segmentation, , a January 2018 commissioned study conducted by Forrester Consulting on behalf of simMachines