Customer segmentation, and the similar and related field of market segmentation, are particularly relevant to the field of business-to-business (B2B) e-commerce. B2B customers often have a higher Customer Lifetime Value (CLV) than those in consumer markets, so companies are much happier to invest much larger sums in acquiring them, because they know they’ll recover those costs later on.
While there’s a clear trend towards digital transformation, whereby B2B e-commerce sites want most of their customers to transact online, where costs are much lower and profits proportionally higher than orders taken offline, some do also invest heavily in providing sales teams to offer personal service through telesales teams and account managers.
While the cost to serve a B2B customer through an account manager can be significant, the hope is that the service will be targeted specifically to those who require it, and those account managers develop personal relationships with their customers to grow their account value by more than it would have done if they had a purely transactional relationship through a B2B e-commerce site.
Customer segmentation can, therefore, help B2B companies in a number of ways, from simple market research and identifying potential key accounts, to discovering the best target audience to focus marketing efforts upon and guiding the customer acquisition and marketing strategy.
In this guide, I’ll explain a range of customer segmentation and market segmentation approaches data scientists and marketing teams can use to segment customers in B2B markets to help improve marketing strategies and increase sales.
B2B customer segmentation is broadly similar to B2C customer segmentation. Many concepts used in B2C segmentation are exactly the same in B2B segmentation, but there are some key differences. The main one is that firmographic data is used to help understand and segment the customer base via market segmentation.
Firmographic segments are those based on variables that describe the business itself - they’re the most common type of market segmentation variable. These are obviously very powerful features in most segmentation models used in B2B markets, but to acquire them you’ll need to send your customer data away to a specialist data house, such as Experian, to have it segmented with your chosen firmographic information and then re-imported into your CRM or data warehouse.
The most obvious market segmentation approach used to segment business customers is to use qualitative data on their business sector. The Standard Industry Classification (SIC) code has long been the accepted way to do this, and it has been adopted by the UK government in its Companies House data to segment businesses.
While it’s useful, SIC codes need to be taken with a pinch of salt by B2B marketers. In the UK, the SIC code is defined by the business, not by the data provider, so it’s very common for the wrong one to be applied. You may need to do some data wrangling with your market segmentation data in order to extract value and use it to guide strategy or make predictions.
The other common approach used in firmographic segmentation is to examine business size. There are, of course, several ways to do this and you may benefit from creating several different customer segments to examine this, as each can have its advantages and disadvantages.
Company turnover or annual revenue is one way to do it. In the UK, turnover data can be obtained for most companies via the government’s Companies House service, or it can be purchased from data providers such as Experian. Since it changes annually, you’ll need to spend money on keeping the segmentation variables up to date if they matter to your market segmentation approach.
Number of employees
The number of employees can also make a difference in some markets. If you sell products that are more associated with people than revenue (for example, workwear, PCs, or office chairs) then it would make sense to acquire this segmentation variable.
Number of sites
For some businesses, the number of sites is a useful quantitative data point to acquire on B2B clients. For example, let’s say you sell CCTV systems for businesses, and they typically have just one system per site. Identifying businesses with multiple sites might be a better way to identify a lucrative customer than simply focusing on revenue, which could easily be skewed if a single site has a very high turnover, but a low requirement for your product.
Demographic segmentation data is, in general, better suited to B2C segmentation than it is to B2B. However, there is one common demographic variable that is often worth collecting - job title. This could be considered a firmographic variable, but you can often collect it yourself during the ordering process, so there may not be a requirement to purchase it separately.
Simply targeting would-be customers using firmographic data only is rarely enough in B2B marketing. If there are 1000 staff at the business you’re trying to target, you need to ensure that your marketing activities reach the right people.
In many cases, for this to work, you need to collect data on your customers’ job titles and identify the sort of roles most commonly linked to purchasing. To complicate matters, the job title responsible for decision makers or those purchasing can vary according to the business sector and the business size.
For example, in smaller businesses, it might be the managing director who places orders, but in larger ones, the responsibility for decision making may lie with the marketing manager or procurement manager, or with a decision making unit made up of several staff.
Once you’ve identified the right job titles, you can approach a data house, such as Experian or Dun and Bradstreet, and purchase contact details for, say, marketing managers who match your company size and company sector requirements.
Needs based segmentation (or benefit segmentation) is pretty much unique to B2B retailing and is rarely seen in B2C settings. Needs based segments look at specific customer requirements that might make a difference to a company acquiring a certain business or placing a certain account under dedicated account management, instead of encouraging a more transactional B2B e-commerce experience.
Probably the most commonly seen needs based segment looks at the way customers order. The buying or purchase process used in B2B markets can be different from that used in B2C companies. The aim is to get as many customers as possible to have a transactional relationship where they purchase online to reduce management costs.
However, some will require a quotation or purchase order (PO) before they can buy. Unless you provide these services online, the cost to serve these customers can escalate. You need to know which clients are happy to purchase transactionally, and which require quotations, POs, or worse, a laborious tender process that can only be achieved through account management. Segmenting customer based on needs is one of many ways to segment B2B customers in a more sophisticated manner.
Single customer vs. multiple decision-makers
In a B2C business you’ll generally be dealing with one customer per account, however, in B2B environments it’s common to deal with multiple decision-makers via a shared account. Large companies often have a decision-making unit, where the procurement process may involve a whole team of people. The company may need to adopt a different strategy for dealing with this.
The other needs based segment that varies across business customers is the level of customer sophistication. For example, larger businesses may require you to sell products through their online procurement platforms, while others (who presumably still live in the last century) might even want to fax you their order!
Buyer personas can be useful to understand the differences that may exist in customer sophistication within your customer base. Buyer personas are usually made by selecting a bunch of customers you know a bit about, and then identifying their various attributes and segments. Most B2B marketers can create them with ease.
Data scientists in e-commerce businesses have access to a wealth of behavioral data, allowing them to construct behavioral segments that help sales and marketing staff understand customers more easily, and target them with products or services based on their likelihood to respond.
Purchase latency uses order gap analysis to measure the number of days between each of a customer’s orders. Since customers usually place orders somewhere around the mean of their typical latency, this behavioral data can be used to identify whether a customer is due to order or not.
That can help solve pain points for both marketing and sales - it helps provide good marketing by contacting businesses at the right time, and it helps sales staff focus their attention on businesses who are going to order, rather than those who’ve already done so.
Finally, there’s value based segmentation, which attempts to assign existing customers to the appropriate segment based on their value or revenue contribution to the business. While it serves a slightly different purpose to some of the qualitative techniques, value based segments are among the most useful for day-to-day marketing activities, especially if (like older B2B businesses) you still do catalogue marketing.
The two most common ways to segment customers based on their value are ABC classification and RFM segmentation. I use both of these heavily in my work, and have created a Python package called EcommerceTools to make it easier for retailers to segment their customers using these algorithms. They’re covered extensively in the articles below.
Segmentation analysis is the technique of analysing your existing customers to gather information about the segments that are correlated with either responding to marketing or sales, growing in value, or becoming your best clients in the future.
Segmentation analysis is often undertaken on small subsets of the customer base, whose segmentation data may be augmented with a wide range of demographic data, geographic data, or firmographic data to help identify segmentation variables correlated with whatever it is that you are trying to predict.
In other words, rather than spending vast sums on getting a data provider such as Experian to segment all of your customers, you’d segment a smaller sample of them - such as your best customers or managed accounts - then use statistical analysis to identify common characteristics associated with sales or likelihood to respond. This initial work can save you lots of money in the long term, and should ensure you have the data you need to better target your marketing campaigns.
Does account management work?
One of the pain points for B2B retailers is that account management is an expensive business. In many businesses, not only do sales staff receive a salary for their efforts, but they’re often also incentivised with commissions for meeting sales targets. Therefore, it pays to ensure they’re being rewarded for generating incremental revenue, and not for simply being allocated high RFM accounts that are naturally growing and would purchase anyway.
Are you managing the right customers?
One of the most popular methods for identifying which accounts require dedicated account management is the Pareto Principle or 80:20 rule. The most practical implementation of this is ABC classification or ABC analysis, which separates customers into a specific segment based on their contribution to cumulative revenue for the business over a specific time period, usually one year.
The idea is that the Class A customers who generate 80% of your revenue probably deserve a more one-to-one service than those who just transact online like the Class B and C customers, which generate the bottom 20% of cumulative revenue. It’s a good general indicator and allows non-technical staff to understand a customer’s contribution to the business without the need to analyse custome data. However, it does overlook customer needs.
Not every Class A customer needs account management (or wants to speak to a person to order), so needs based segments are being used alongside ABC classification to fine-tune which accounts actually warrant expensive account management, rather than providing it where it may not be needed.
What target market should you focus upon?
It would be easy, but shortsighted, to assume that you should simply target the sectors which have the highest average account values. However, this overlooks the fact that some of them are easier to convert than others. For example, let’s say your customer data shows that major supermarkets have the highest account values, so your marketing team decides to target these businesses.
However, as there are relatively few major supermarkets, and they have thousands of staff, it is difficult for marketing teams to reach the right contact within the customer segment with their marketing messages, resulting in a much lower conversion rate, or quite probably, no conversions at all.
As well as their account value, and other behavioral segmentation data showing their spending behaviour, it’s important to consider the conversion rate achieved from your marketing efforts when trying to hit a given customer segment. You may find, for example, that it’s actually much more profitable to target your marketing at businesses with a lower average account value, but which are much easier to convert, such as smaller businesses.
If your a given customer spends £100K per annum on your widgets and they’ve grown by 5% in the past year, is their account manager doing a good job, or not? There are various data science techniques you use to estimate the potential value of a business in the future.
Average account value for firmographic segments
If you already hold firmographic data on your customers, then it would make sense to examine the average account value and purchasing behaviour across customers in the same sector, or who have similar numbers of employees or sites, or a similar turnover.
How does their spend compare to similar customers? What proportion of their turnover is spent on your products, and how does that compare between similar customers? If other customers are spending more of their turnover on your products, or their annual spend per site or per employee is higher, perhaps there’s more that could be done to grow this account.
What products do they buy?
Category-level RFM is one of my absolute favourite user data analysis techniques for B2B. Rather than creating an RFM score for each customer at the business level, it creates an RFM score for each customer at the category level.
This can show you, for example, how their purchase history differs across different product categories, showing you where they’re no longer buying, where they don’t buy at all, and where they buy most. It’s a perfect way for account managers to see how customers interact with the business and lets them visualise customer data in a simple and logical way, so they can target people with products they perhaps should be purchasing but aren’t.
Machine learning has two main applications in B2B e-commerce, as far as segmentation data are concerned, which can both help improve sales and marketing efforts. Firstly, it can be used to create new segments based on combinations of various other qualitative and quantitative data, such as behavioral, needs-based, firmographic, demographic, or value-based segmentation variables, usually using a method such as k-means clustering. Secondly, it can be used to power machine learning response models.
Cluster analysis Clustering, or cluster analysis as it’s also known, is the process of creating groups of current customers who share common characteristics. Clustering is a form of unsupervised machine learning, so doesn’t require data to be labeled with a given class. Instead, you provide numeric data to the model (both quantitative data and numerically encoded qualitative data) and it creates clusters based on their underlying numerical similarity.
The k-means clustering model is by far the most widely used unsupervised machine learning algorithm for implementing customer clustering in B2B marketing. It’s quite practical because it allows you to create a specific number of clusters, which are all unique in their characteristics, which can then be targeted with carefully focused marketing messages.
Response models are supervised machine learning models that use classification algorithms to predict the probability of each customer (or each segment) responding to a sales call or marketing promotion. This is particularly important for B2B e-commerce teams because they commonly incorporate an outbound sales team to try and generate additional sales through telesales marketing. While they’re not all constructed in the same way, they’re all segmentation based in some shape or form.
A typical B2B telesales team will usually be given a list of companies to call and will be targeted with contacting each one to try to generate a sale. To demonstrate why companies probably want to listen to their data scientist and build a response model, let’s look at the numbers.
We’ll assume that the average salesperson gets paid £10 per hour (about £22K pa) and, over the course of a week, makes an average of 10 calls per hour. Over a 30-hour working week, let’s say they make 300 calls, at a cost to the business of £300, not including commission. To calculate the potential profitability of this activity, these are the things we need to work out:
Cost per call = Hourly rate / Calls Conversion rate = (Orders / Calls) * 100 Cost per conversion = Costs / Conversions
With a 1% conversion rate, our hypothetical salesperson would generate 3 sales in a week, giving us a cost per conversion of £100 per order, whereas a 10% conversion rate would generate 30 sales in a week and a cost per conversion of £10 per order. The average order value, therefore, needs to be pretty big to break even, otherwise, you’re paying simply to fill the time of your sales staff.
Building a response model
Savvy B2B e-commerce retailers who utilise outbound sales teams to cold-call potential or existing customers and offer products or services are now using machine learning response models to help retailers increase sales staff profitability.
A segmentation based response model uses segmentation data to identify relationships between customer segments and the likelihood to respond to a sales call. Since they use classification models that work via supervised learning, they require labeled training data. The easiest way to generate this is simply to do things the old-fashioned way and then get staff to record whether the customer purchased or not when called.
By creating a feature set based on a wide range of behavioral segmentation and market segmentation data, you can train a response model to predict whether a given customer will purchase when called. When fed a list of new potential customers, previously unseen by the model, you can get back a list of customers ranked according to their probability of responding.
My response models have often generated well over 90% accuracy, massively reducing labour, greatly increasing call conversion rate, and decreasing the cost per conversion, while also generating much more revenue for the business. They’re well worth building.
One of the other methods used in B2B marketing is Account Based Marketing or ABM. ABM is an extreme form of B2B customer segmentation in which accounts (whether they are prospects or existing customers) are marketed to in segments of just one.
To reiterate, rather than marketing to a target audience of thousands of potential business customers, account based marketing aims to market to just one business at a time - a customer segment is then a single customer. While it’s been used in IT and service sectors for some time, this marketing strategy is very uncommon in B2B e-commerce businesses, though some aspects of it do roll over into B2B account management.
To roundup, here’s my quick guide to the steps you need to follow to implement an effective segmentation strategy in a B2B e-commerce business:
Matt Clarke, Saturday, August 14, 2021