Models they’ve used:
Email Cleaner & Last Reply Extractor
Extract the last reply from an email thread. A cleaning mode …
Outbound Sales Response Classifier
Classifies outbound sales email responses based on their…
Drift is a high-growth SaaS company leading the conversational marketing and sales revolution. Founded by David Cancel and Elias Torres, Drift recently raised a $60M series C and continues to grow and expand.
Its hypergrowth implies that they need to be able to quickly evolve their product and offering.
In January 2018, Drift announced Sequences: a new way to send sales emails, which gave sales reps a revolutionary new way to connect with potential customers.
Drift Sequences aimed to help companies:
Connect sales emails to real-time messaging on their websites.
Avoid sending unwanted emails.
Use behavioural data and responses from emails to have smarter conversations.
To avoid sending unwanted emails, Drift sought to use machine learning and natural language processing to automatically allow recipients to opt out of emails based on how they reply.
Drift’s customers are actively trying to sell their own products and services. With tens of thousands of emails being sent, some recipients are bound to opt out. This required filtering through email responses and managing requests to unsubscribe. Instead of having sales reps take on this job, Drift looked to save their time by using machine learning models to automate the process.
Although Drift could have chosen to develop this feature internally, working with MonkeyLearn allowed Drift to focus on their product priorities, while still being able to release an important feature more quickly to their customer base of more than 100,000 businesses.
In House vs MonkeyLearn
Had Drift decided on pursuing an in-house development path for this feature, they would have required:
A data science team with relevant experience in Natural Language Processing and Machine Learning to work on three major problems:
Building tools to prepare, curate and tag training datasets.
Choosing the right machine learning frameworks to develop the model.
Executing on the testing and experimentation necessary to find the right model.
Building the underlying infrastructure to handle training models, which usually involves working with parallel computing, clusters of servers, managing huge amounts of RAM, as well as managing multiple CPUs and GPUs.
Deploying their own infrastructure, using either raw servers or using a PaaS to run the models.
Investing in maintenance, both on the engineering side (to ensure infrastructure is working properly) and on the data science side (model maintenance, adjustment and improvement of machine learning algorithms, as well as retraining models).
This would have required thousands of hours and a ramp up time of more than a few months to get to the first production level MVP.
With MonkeyLearn, Drift not only saved thousands of hours of their internal team’s time, they were able to put a significant feature into production in just a matter of weeks, and one that has a huge impact on their customers’ productivity.
“Working with MonkeyLearn allowed us to quickly and easily create a new feature for our customers, without having to dedicate internal resources or spend months on custom development.”
VP Growth @ Drift