Today, I’m excited to announce the launch of Primer Automate — the world’s first no-code, end-to-end artificial intelligence solution for NLP model creation and deployment. Primer Automate allows you to encode your tacit knowledge within the machine and essentially teach it to understand your world.
Our latest addition to the Primer product suite extends our world-class machine learning technology to put NLP models into your hands. We make it easy for anyone to quickly build their own models, or train and deploy our proven NLP engines, to automate the most demanding text-based workflows with human-level precision.
Organizations use Primer Automate to power their daily workflows in the following ways:
• Build a custom model: Want to create your own model from the ground up? Our easy-to-use interface guides you through the steps — no coding or technical skills required.
• Retrain our models: Need to tweak our models to meet your specific needs? You can quickly retrain any one of our models against your custom data sets.
• Use our models off-the-shelf: Want to get started using best-in-class models right away? Simply access dozens of Primer’s own models and deploy them with a single click.
We are launching with three of our most popular models: Classifiers (including multi-class classifiers), Sentiment, and Named Entity Recognition. Our selection is constantly growing, and new engines are coming online every month, including: Summarization, Key Phrase Extraction, Headline Generation, and Topic Identification.
The idea for Primer Automate started with our customers. Some of our Fortune 50 customers needed custom NLP models and would come to us asking for help. Our team of machine learning experts would build them on the customer’s behalf, and while the results were good, it was prohibitively expensive for the vast majority of our customers.
We set out to democratize NLP model customization and give everyone the ability to build their own models without having to know any code or needing a whole team of engineers working beside them. With Automate, we’ve achieved something pretty unique: analysts now have the tools to train machines to automate all those mundane, repetitive tasks that they are doing each and every day. Not only that, but automation at scale can surface things that would have otherwise remained hidden using manual workflows, which gives analysts new insight into their world. We believe that Primer Automate is going to profoundly change knowledge work — literally everything that you do with text will be transformed — and we can’t wait to see what our customers achieve with it.
Primer Automate sits on top of our proprietary NLP technology, which is built on transformer-based neural networks, and fully integrates active learning into the user experience. This capability makes our machine learning models faster, more accurate, and importantly, easy to retrain.
Active learning makes sure that the examples that you’re asked to label are the ones that will best improve the performance of the model. This means that you spend less time labeling, and more time working with the output of your models. Active learning can reduce the amount of labeled data needed to train an NLP model by up to 30 times versus conventional methods.
In a way, active learning is more respectful of the expertise that humans bring to the table. Instead of spending your valuable time labeling thousands of repetitive text examples, Automate shows you examples of where the model is least certain, so you can teach the machine what you know and directly improve its performance. And the machine teaches you as well, prompting you to think through complex questions and edge cases, so you come away from the training experience better knowing your own data.
Using Primer Automate is quite simple. You log in and create a reading or writing task that you want to automate. You then start training your model, which involves labeling data until it is able to reach the right level of performance. Some people start with a few dozen examples to get things running and then move to a few hundred examples, which brings their model closer to human level performance.
When this is done, you can output the results of the model into a spreadsheet, or with a single button you can generate an API key to deploy the model on your own data. If you want to update the model to make it better, you can go back and do that easily. The model and the training data are available for you to audit; it’s all versioned for you.
We’ve made it incredibly easy to build these advanced natural language processing models. No coding or technical expertise is required — you can do it all within the user interface. Automate guides you through all the steps: uploading data, labeling, training, exporting, and deploying at scale. You can even collaborate with team members or tap into external labeling resources. The training process is not only easy, but it’s actually pretty fun. Check it out for yourself in our Automate webinar.
There really is a huge number of ways that organizations can use Primer Automate. We’ve seen finance analysts building models to identify supply chain risks from news data, customer support teams building models to identify incoming emails from customers at risk of churn, and researchers building models to classify scientific papers about COVID-19 into a set of topics. And that’s only the tip of the iceberg.
Here’s a real-world example. Let’s say that you are a business analyst at a home exercise equipment company and you need to analyze customer sentiment trends across social media platforms. You would first upload all the text from Instagram posts about the company. In a couple of hours, you could build your own classifier model that identifies posts from people who love their equipment because of the community versus those who love the equipment because of the instructors. You can take the same posts and build a classifier to help you understand user frustration by identifying posts from people who can’t get their equipment to function properly versus those who think it’s too expensive. The entire workflow can be done in seconds, giving you time to focus on next steps.
This is an important release for analysts across sectors and industries. Compared to other solutions on the market, Primer Automate stands alone. Some NLP platforms allow you to label data, others are just starting to allow you to train models or optimize things like hyper parameters to improve model performance. But these solutions are all very complicated to use or require days of work and considerable expertise to integrate together.
Primer Automate is really a breakthrough product. We haven’t seen anyone else launch a product that allows users to go from data to fully built and deployed models all in one application. Automate is one more way that Primer is transforming knowledge work, and we are incredibly excited to bring this capability to our customers.