Machine Learning For B2B: Interview with 6sense’s Mio Alter

 In Data Science, Inside 6sense

“Can machines think?” 

-Alan Turing

Today we’ve reached a defining moment where companies are finally taking the leap into earnestly using big data and predictive intelligence solutions to grow their businesses. Yet machine learning, a key part of predictive analytics, remains an elusive and largely misunderstood concept.

 We interviewed one of 6sense’s lead data scientists, Mio Alter, to get to the bottom of machine learning and find out why it’s a big deal for B2B businesses.

Mio Alter

Mio Alter, Data Scientist, 6sense

What is machine learning?

Machine learning means training a machine to associate known patterns with known outcomes, and then when the machine sees new patterns it can predict new, unknown outcomes.

We have to represent all of our raw data mathematically. This is called “feature generation.” Essentially, we look at all the things a company or individual does and find a meaningful way to represent that as a list of numbers. Then for every company and contact, we have a list of numbers that represents things that happened (for example, a closed sale), and things that didn’t happen.

What’s the desired output of machine learning?

We’re essentially making predictions on the basis of numbers (a mathematical representation of the input data), and training our tools/machine to associate certain patterns with certain outcomes. Machine learning is really about adjusting the knobs of the machine to get it closer to the right answer. To achieve this, you have the training phase, where you know the answer to the data (i.e., this action led to this outcome). After training the machine, you can put data in that you don’t know the answer to, and it will predict the outcome.

What is the applicability of machine learning for B2B companies, specifically?

B2B companies sell to other companies, so there is a focus on account-based marketing, among other strategies. For a given company, our job as data scientists is to aggregate all the things that both the account and the contacts within that account did (i.e., behavioral data). Using machine learning, we’re able to now test whether or not specific actions will take place. Done correctly, machine learning can help to predict a company’s likelihood to buy and offer other data such as time to close or products reviewed. B2B marketing and sales teams can use these predictions to target the right accounts.

What are some misconceptions about machine learning?

Machine learning makes it possible to extract meaning from huge and chaotic piles of data. This is extremely powerful and valuable. But, it is not a magic wand that you can point at anything. It requires domain knowledge to build a machine that draws the most meaningful insights from the most meaningful pieces of data. In the B2B world, this means a lively back-and-forth between the math experts and the domain experts to hone in on the most valuable business insights.

What are some of the challenges when it comes to machine learning?

Data is usually a mess. There’s tons of missing data—but that doesn’t necessarily mean you throw data away or don’t consider it in your predictions. You need to be very clever with how you deal with data. We have to think: What is the meaning of this missing data? Does the missing information imply that something didn’t happen? Or just that there’s no information, for example, if it happened a long time ago?  It takes experience and finesse to tackle these questions.

Machine learning also requires that we take descriptive or qualitative data, for example, columns of notes from Salesforce and turn it into quantifiable math. On a daily basis, we have to make decisions to parse the text in some way and write an algorithm to determine that this piece of the text is the most important thing.

Also, it’s easy for someone to count. But when there are lots of possibilities, or each thing only happens once, how do you determine what’s meaningful? This is where business and customer leads come in, because what’s meaningful differs from company to company. Someone with a strong understanding of a customer’s business model, audience, and objectives is able to give insight on which patterns in the data are important. For example, for one company, a prospect downloading a whitepaper may not be all that meaningful, but a prospect downloading a whitepaper plus attending an in-person event might be. The significance of data varies from organization to organization, which makes it that much more important for the team delivering the machine learning insights to be well-versed in that organization’s pain points.

How does 6sense do things differently, when it comes to machine learning?

The fact that we combine a multitude of different data sources is a big differentiator. We don’t only use data from our customers; we also use behavioral data. By not being tied to our customers’ historical data, we’re able to predict net-new customers. That’s in addition to ranking existing customers throughout the whole funnel. 6sense is also unique in our methodologies because we’re able to aggregate data from both known and unknown contacts to shed light on an overall company or account’s readiness to buy, which is important for B2B companies specifically.

We’re also pioneering some exciting things when it comes to prediction, ideas which require deep knowledge about temporal influencers of data. For example, if we only have a current snapshot of the data, can we use that to infer the state of the world at other points in time? I can’t say more on this just yet, but it’s going to be big.


 Interested in learning what the 6sense data science team can do for your business? Request a demo.

Request a Demo

Recent Posts

Start typing and press Enter to search