If I told you that machine learning could complete a task in one second that would take 10 years for 200 people to
complete, would you invest in that system?
Gone are the days when Machine Learning (ML) and Artificial Intelligence (AI) were buzzwords that no one quite
understood. While some automation systems require people to hand-code each rule, these systems are built to
automate at scale. This is critical for growing companies to understand.
At SSG, we’ve witnessed the wonders of machine learning firsthand. In fact, we used ML to de-dupe four million
records and accurately categorize one million unique job functions within our database. We know why ML is so big
and so important: because it’s infinitely scalable.
Machine Learning for Marketers
On average, marketers spend more than five hours per week attempting to make sense of data. Yet with the help of
machine learning, marketers will get those five hours back and more. According to a recent study, 66% of marketers
agreed that automation and machine learning will allow their team to focus on more strategic initiatives.1
This means that ML will free marketers to do what they do best and allow the machine to do what it does best: make
sense of the data.
Additionally, ML is just one of many ways to personalize the customer experience by using clean data. Depending on
what customers look for, machine learning allows marketers to infer better recommendations. In the data-driven
world of demand gen, ML should empower marketers to give customers better leads. Imagine a world where
contracting sales leads is as easy as browsing your recommended shows on Netflix. That’s what data scientists are
trying to build. By creating new data points that were not previously tracked, the machine learning model will
improve its accuracy. That is one way you can train the model to learn better.
Fun fact: Netflix saved $1B in one year by using machine learning to
make personalized recommendations2
According to Forbes, “Measuring marketing’s many contributions to revenue growth is becoming more accurate and
real-time thanks to analytics and machine learning. Knowing what’s driving more Marketing Qualified Leads (MQLs),
Sales Qualified Leads (SQLs), how best to optimize marketing campaigns, and improving the precision and
profitability of pricing are just a few of the many areas machine learning is revolutionizing marketing.”3 In addition,
many marketers believe that AI and ML will increase customer satisfaction, improve overall customer experience, and
aid in lead scoring. Plus, machine learning will help marketers make better sense of the data they’ve collected and
reduce the time it takes to craft personalized experiences based on all that data.
Another real benefit of ML is that it strips away many of the frustrations and complexities we face in our jobs. How
can we make a salesperson’s life easier? If they need to fulfill a quota of 1,000 leads but only find 800, they might
have to manually find 200 more. But if we can train our models to do that, it saves our salespeople precious time and
helps our clients meet deadlines. At Selling Simplified, we first pursued machine learning in order to automate parts
of our business. The real win is that it makes life a lot easier for our people and our clients.
Some (but not all) Ways That SSG Uses Machine Learning
Early in SSG’s machine learning days, the team worked on the “De-Dupe Project” in which machine learning models
were applied to remove duplicate records from the database. This is a very common application for ML, aimed at
improving data quality and speed-of-delivery for clients. The project has been largely successful since deploying in
early 2021, making 7% more of the database accessible by merging duplicate records.
In fact, the main way that SSG uses machine learning is to cleanse data from man-made errors. For as long as manual
input exists, duplicate records will exist. For example, if salesperson A typed in “Apple” and salesperson B typed in
“Apple Inc.,” this would result in two separate profiles (records) for the same company. In order to write instructions
about something like naming conventions, there would be thousands of pages of instructions. So, data scientists
began teaching the machine to pick up matches by assigning “weights and biases” to the problem. SSG’s de-dupe
application can view those separate records as the same, much like the way a human would focus on the “Apple”
portion and ignore the “Inc.” portion.
Another model recently created by the SSG team is the “Job Function Model.” This model is totally unique because
the technology was developed completely in-house. The Job Function Model reads job titles, predicts what sort of job
function they perform, and groups them by that category. This was created when SSG’s database amassed almost a
million unique job titles and found many errors in the entries, some of which completely lacked details about the job
function. Without machine learning, it would take roughly 12 years for the data team to fix the Job Function field
manually! This model does it instantly and with about 90% accuracy. It is also able to label bad or junk job titles for
manual cleaning or review.
SSG is truly on the cutting edge of all that is possible in ML right now. Soon, the team plans to move into deep
learning models, which goes a step further. Instead of trying to solve a math equation, these models try to replicate
the human brain through artificial neurons. This is necessary because it is very difficult to read human data.
Making Sense of the Madness
To understand machine learning in simple terms, it helps to look at the difference between regular programming
and ML programming.
Normal programming:
- 1.you have a program
- 2.you give it an input (data)
- 3.you provide a set of instructions
- 4.the program gives you an output
Machine learning:
- 1. you build the model
- 2. you give it an input
- 3. you give the desired output
- 4. it creates the instructions
In other words, as you give the model inputs—with an assigned output—it will learn the “instructions” and pick up
patterns to make predictions. You tell it the end goal, and it tells you how to get there. This is truly the simplest way to
explain machine learning. It requires data scientists to train the machine—i.e. to help the machine “learn” to pick up
patterns—so that they can make better sense of the data.
Another important distinction lies between machine learning and Artificial Intelligence (AI). According to Northeastern University associate dean Bethany Edmunds, “Where intelligence is the overall appearance of being
smart, machine learning is where machines are taking in data and learning things about the world that would be
difficult for humans to do.” While AI uses data to solve problems and analyze patterns, ML “identifies anomalies,”
allowing the machine to address problems that a human might not find or identify. Some understand machine
learning as a deeper layer within AI.
Not Without Challenges
Since machine learning is rapidly evolving, there are expected growing pains for data scientists and marketers alike.
The biggest challenge that we face is getting all businesses—and people—to agree on structured data. This includes
internal and external data. For example, sales profiles are set up by people, not computers, and it’s difficult to find
patterns in human data. You may even need a million copies of data to teach the machine. If the copies aren’t
consistent, it becomes very challenging to apply successful learning models.
Another challenge is perhaps less obvious but still common: it is tremendously difficult to represent words in
numbers. As marketers, we care about data in the form of words (email addresses, company names, job functions,
etc.). But computers don’t think in words. The challenge is to represent these fields in numbers for the machine
learning model to understand.
Looking Ahead
Despite its complexities and challenges, machine learning—and AI—is the future of data. Because there is so much data out there, our task is to find a better way of processing it.
At SSG, one goal is to make an advanced ABM tool for our customers. Imagine that you could take an ABM list and teach the ML model to recommend companies to include in the list. In the past, ML models could not pick up patterns because there were not enough descriptive statistics in the database. But our team is working scrupulously to collect more data and statistics to improve the model.
We’re also seeking another way to gather more information about the companies in our database, including intent, revenue growth, employee tenure, brand loyalty, etc. Once we get data behind these attributes, we can teach the machine to pick out patterns. The idea is to generate more usable data to teach SSG’s models to learn better and deliver better results. “We’ll never be truly finished,” said one team member. “We’ll always be improving.”
The Unsearchable Wonders Of
Machine Learning
Feature Story
Abby Nickols
Corporate Content Writer
April 1, 2022
If I told you that machine learning could complete a task in one second that would take 10 years for 200 people to complete, would you invest in that system?
Gone are the days when Machine Learning (ML) and Artificial Intelligence (AI) were buzzwords that no one quite understood. While some automation systems require people to hand-code each rule, these systems are built to automate at scale. This is critical for growing companies to understand.
At SSG, we’ve witnessed the wonders of machine learning firsthand. In fact, we used ML to de-dupe four million records and accurately categorize one million unique job functions within our database. We know why ML is so big and so important: because it’s infinitely scalable.
Machine Learning for Marketers
On average, marketers spend more than five hours per week attempting to make sense of data. Yet with the help of machine learning, marketers will get those five hours back and more. According to a recent study, 66% of marketers agreed that automation and machine learning will allow their team to focus on more strategic initiatives.1 This means that ML will free marketers to do what they do best and allow the machine to do what it does best: make sense of the data.
Additionally, ML is just one of many ways to personalize the customer experience by using clean data. Depending on what customers look for, machine learning allows marketers to infer better recommendations. In the data-driven world of demand gen, ML should empower marketers to give customers better leads. Imagine a world where contracting sales leads is as easy as browsing your recommended shows on Netflix. That’s what data scientists are trying to build. By creating new data points that were not previously tracked, the machine learning model will improve its accuracy. That is one way you can train the model to learn better.
Fun fact: Netflix saved $1B in one year by using machine learning to make personalized recommendations2
According to Forbes, “Measuring marketing’s many contributions to revenue growth is becoming more accurate and real-time thanks to analytics and machine learning. Knowing what’s driving more Marketing Qualified Leads (MQLs), Sales Qualified Leads (SQLs), how best to optimize marketing campaigns, and improving the precision and profitability of pricing are just a few of the many areas machine learning is revolutionizing marketing.”3 In addition, many marketers believe that AI and ML will increase customer satisfaction, improve overall customer experience, and aid in lead scoring. Plus, machine learning will help marketers make better sense of the data they’ve collected and reduce the time it takes to craft personalized experiences based on all that data.
Another real benefit of ML is that it strips away many of the frustrations and complexities we face in our jobs. How can we make a salesperson’s life easier? If they need to fulfill a quota of 1,000 leads but only find 800, they might have to manually find 200 more. But if we can train our models to do that, it saves our salespeople precious time and helps our clients meet deadlines. At Selling Simplified, we first pursued machine learning in order to automate parts of our business. The real win is that it makes life a lot easier for our people and our clients.
Some (but not all) Ways That SSG Uses Machine Learning
Early in SSG’s machine learning days, the team worked on the “De-Dupe Project” in which machine learning models were applied to remove duplicate records from the database. This is a very common application for ML, aimed at improving data quality and speed-of-delivery for clients. The project has been largely successful since deploying in early 2021, making 7% more of the database accessible by merging duplicate records.
In fact, the main way that SSG uses machine learning is to cleanse data from man-made errors. For as long as manual input exists, duplicate records will exist. For example, if salesperson A typed in “Apple” and salesperson B typed in “Apple Inc.,” this would result in two separate profiles (records) for the same company. In order to write instructions about something like naming conventions, there would be thousands of pages of instructions. So, data scientists began teaching the machine to pick up matches by assigning “weights and biases” to the problem. SSG’s de-dupe application can view those separate records as the same, much like the way a human would focus on the “Apple” portion and ignore the “Inc.” portion.
Another model recently created by the SSG team is the “Job Function Model.” This model is totally unique because the technology was developed completely in-house. The Job Function Model reads job titles, predicts what sort of job function they perform, and groups them by that category. This was created when SSG’s database amassed almost a million unique job titles and found many errors in the entries, some of which completely lacked details about the job function. Without machine learning, it would take roughly 12 years for the data team to fix the Job Function field manually! This model does it instantly and with about 90% accuracy. It is also able to label bad or junk job titles for manual cleaning or review.
SSG is truly on the cutting edge of all that is possible in ML right now. Soon, the team plans to move into deep learning models, which goes a step further. Instead of trying to solve a math equation, these models try to replicate the human brain through artificial neurons. This is necessary because it is very difficult to read human data.
Making Sense of the Madness
To understand machine learning in simple terms, it helps to look at the difference between regular programming and ML programming.
Normal programming:
Machine learning:
In other words, as you give the model inputs—with an assigned output—it will learn the “instructions” and pick up patterns to make predictions. You tell it the end goal, and it tells you how to get there. This is truly the simplest way to explain machine learning. It requires data scientists to train the machine—i.e. to help the machine “learn” to pick up patterns—so that they can make better sense of the data.
Another important distinction lies between machine learning and Artificial Intelligence (AI). According to Northeastern University associate dean Bethany Edmunds, “Where intelligence is the overall appearance of being smart, machine learning is where machines are taking in data and learning things about the world that would be difficult for humans to do.” While AI uses data to solve problems and analyze patterns, ML “identifies anomalies,” allowing the machine to address problems that a human might not find or identify. Some understand machine learning as a deeper layer within AI.
Not Without Challenges
Since machine learning is rapidly evolving, there are expected growing pains for data scientists and marketers alike. The biggest challenge that we face is getting all businesses—and people—to agree on structured data. This includes internal and external data. For example, sales profiles are set up by people, not computers, and it’s difficult to find patterns in human data. You may even need a million copies of data to teach the machine. If the copies aren’t consistent, it becomes very challenging to apply successful learning models.
Another challenge is perhaps less obvious but still common: it is tremendously difficult to represent words in numbers. As marketers, we care about data in the form of words (email addresses, company names, job functions, etc.). But computers don’t think in words. The challenge is to represent these fields in numbers for the machine learning model to understand.
Looking Ahead
Despite its complexities and challenges, machine learning—and AI—is the future of data. Because there is so much data out there, our task is to find a better way of processing it.
At SSG, one goal is to make an advanced ABM tool for our customers. Imagine that you could take an ABM list and teach the ML model to recommend companies to include in the list. In the past, ML models could not pick up patterns because there were not enough descriptive statistics in the database. But our team is working scrupulously to collect more data and statistics to improve the model.
We’re also seeking another way to gather more information about the companies in our database, including intent, revenue growth, employee tenure, brand loyalty, etc. Once we get data behind these attributes, we can teach the machine to pick out patterns. The idea is to generate more usable data to teach SSG’s models to learn better and deliver better results. “We’ll never be truly finished,” said one team member. “We’ll always be improving.”