Opportunities for using Artificial Intelligence through the sales cycle

Using machine learning and predictive intelligence through the B2B Buyer journey

AI (artificial intelligence) has become a business buzzword this past year, even featuring in the mainstream media. Many martech vendors are competing to launch AI features for their product or use an AI tool to do something that would previously have required human input.

Yet AI is such a broad term, you need to go beyond the hype to review its opportunities for your sales and marketing processes. That’s what we’ll do in this article. At a top-level, AI includes any kind of computer program which actively seeks to mimic a human capability, such as understanding speech, recognizing images or responding to questions. When it comes to using AI in the sales cycle, there are two technologies which are particularly useful, and it’s worth drilling down and understanding them rather than focusing on the nebulous term ‘AI’. These are Machine Learning and Predictive Intelligence.

The two technologies can work in tandem to provide your sales team with a way to target the hottest and most qualified leads, and thus save time and bring in more revenue.

What is machine learning?

Machine learning allows you to ‘train’ a program to make accurate predictions using data. A machine learning algorithm runs through a database of historical customer data and establishes the trends in the data, in order to create a prediction model.

A key benefit here is that once, only very large businesses with the capacity to employ large in-house teams with access to large quantities of customer data, now many medium or small businesses also have access to large quantities of data on their customers. This is thanks to the digital revolution, which has created many marketing automation techniques for targeting customers, which predictive analytics can be extremely useful for.

B2B sales teams can benefit from getting automatically alerted to the propensity of a customer to purchase a given product, thus allowing them to make decisions about making unique offers to that customer. As the simple graphic below shows, creating accurate models of a customers’ propensity to buy and their future value lets you make smarter decisions about where to invest your marketing budget or when to offer discounts.

It’s not just about knowing when to offer discounts. The B2B sales process, particularly for businesses whose product is low volume and high margin, is often highly consultative and involves tailoring the solution to the client. This means it takes up a large amount of time to properly follow up on a lead, and if these leads have very little propensity to buy, this wastes large amounts of time and money. By applying accurate propensity models to establish which leads are the most likely to purchase, you can target your sales team’s time at those prospects.

To be effective, machine learning for predictive analytics requires large amounts of historical data. This is because, without large numbers of test cases, it may be that extraneous factors could be having a major outcome of the results. Because the machine-learning algorithm can only work with the historical data it has, if your data set is too small it may be inferring propensity from non-relevant events. For example, if you had a small set of data on customers who bought a certain product, and one customer mistakenly bought one product when they intended to buy another, the model would extrapolate that propensity to buy and apply it to future customers, when in reality these customers would not hold that propensity. Therefore, you need a large data set in order to allow the machine learning algorithm to recognize what the underlying trends in the data are, and what are just noise in the data.

The diagram below shows the basic steps of the machine learning process. First a training data set is compiled, in which a number of descriptive features relate to a target feature. The algorithm is applied to this data set, and a prediction model is created. This prediction model can then be fed real-world query instances (e.g. is this customer likely to convert) and make predictions. These predictions can then influence business strategy. These may be done manually or programmatically, depending on the way your businesses systems are configured.

Therefore, using machine learning you can build an accurate prediction model which allows you to score leads for your sales team and target offers at exactly the customers where they will be most effective. This is predictive analytics in a nutshell.

Case Study: Propensity modelling of Smart Insights’ leads

To illustrate the business potential of machine learning creating more efficient B2B sales process, here is how we used this type of technology in our own sales process. SmartInsights has engaged in a data analytics project that involved using machine learning to predict the propensity of our subcribers to upgrade to our paid learning package. This propensity could then inform how we target offers at our freemium members.

By sharing how we ran this project and what we gained from it we hope you can learn important lessons about how to implement machine-learning projects.

The Business Problem:

We get several thousands of people signing up to our free basic membership option every month. Some of these members are happy to remain on our basic plan indefinitely. Some join the basic membership and then very quickly upgrade to our expert plan. And then another group remain basic members a long period, and then eventually decide to upgrade to expert membership.

It would be impossible for us to contact all the thousands of new basic members that sign up every month, so we wanted to grade users on how likely they were to convert, giving them a ‘heat score’ based on the interactions they had with the site.

We assumed more instances of a user reading one of our guides, visiting the pricing page or looking at our blog would increase their likelihood of upgrading. But we had no idea what were the key factors in determining this, and what the effect of time was on the changes (e.g. is someone who has been a member for two weeks more or less likely to convert than someone who signed up one day ago?). In addition, we had no idea how to correctly weight these events to build the most accurate prediction model.

To answer these questions and create the most accurate prediction model based on our data, we used machine learning to establish what the most effective way of scoring our leads.

The Data

For our training data set we look at the journeys of 53,499 of our basic members, analysing what pages they visited and how long it took them to convert from basic to expert membership (for those that did upgrade). We had data on 9,889,930 unique sessions from the 53,000 members, and 85 different unique event types that could be tracked and analyzed for each user.

Modelling

Out of the 85 event types we had data for, the machine-learning algorithm used identified the same 6 metrics as key conversion drivers, using 3 different propensity models (one modeled over 3 days, one over 7 and one over 30). This indicates they are likely to be due to real characteristics and not due to noise in the data.

Deployment

We implemented the 7-day propensity-scoring model by building an integer heat score into our CRM for each of the various key metrics identified. This could then be tracked by the CRM and then we could take actions when a basic members score rose above what the model identified as the required threshold for the lead to be likely to convert. We could contact a prospect that had become ‘hot’ via email to nudge them into converting.

The project also revealed that basic members who had completed our customer on boarding processes were more than twice as likely to sign up for expert membership as those who had not completed the process. This lesson meant we placed a greater emphasis on our on boarding process to future customers, and also tried to simplify the on boarding process so more would complete it.

Machine learning and AI across the customer lifecycle

There are a wide range of AI techniques which can be applied across the B2B customer lifecycle, many of which can support the sales process and enable your sales team to concentrate only on the most relevant leads, whilst automation other tasks.

The graphic below shows the range of different artificial intelligence, machine learning, and propensity modeling techniques which can be applied and different stages of the customer lifecycle.

Of particular relevance to sales teams will be those technologies in the ‘Act’, and ‘Convert’ stages, such as propensity modeling, lead scoring and dynamic pricing. Lead scoring is particularly useful for B2B sales processes, especially if you have a consultative sales process. Because each sale takes a considerable amount of the sales teams time, accurately scoring leads can help to focus time where it is most productive, allowing the sales team to make more sales and prevent wasting time on leads who would never covert.

Summary

The B2B sales process has been transformed over the past five years by ‘inbound’ marketing strategy, where new marketing tactics like content marketing changed the sales process to be more about consulting with interested parties than attempting the ‘hard sell’. The AI techniques represent the next step in this evolution, which will allow sales teams to target the very best leads. AI is not just about automating the sales function, it’s about freeing up human operatives to use their time doing what they are best at: selling.