⬅️ Back to posts
Filter By Categories

Democratizing ML

Sami Kaipa
August 16, 2022

Voting. Finance. Information. Drones. Software Development. AR/VR…


This list could literally go on and on with examples of things that have been democratized.


So, how will we democratize Machine Learning (ML)? AutoML.

Wait, what is AutoML?

Ok, AutoML may not be all that is needed to fully democratize ML. But it helps. A lot. It certainly starts us down the path of democratization.


AutoML automates how machine learning builds models. It streamlines the time-consuming steps in the ML lifecycle. Specifically, it simplifies data prep, feature engineering, model selection, and model training.


Seems like a lot? Sounds amazing? Yes, it is…


If done correctly, AutoML lets the implementor focus on the first step of acquiring data and the last step of evaluating predictions. All other steps in between are done for you. It helps to see this visually:Traditional ML vs AutoMLThe potential impact of this technology is HUGE.Huge

The future of AutoML is so bright its gotta wear shades 😎

First, as stated, it sets us down the path of democratizing ML. The end goal is to empower line of business workers to access the power of AI without requiring data science resources.


In this future world, Parker, our SalesOps Lead with no ML skills, could easily automate lead scoring and prospecting.


You see, Parker has a spreadsheet with multiple years of manually entered customer data. With AutoML, Parker could just upload this spreadsheet and click a button. And voila! Just like that, Parker would have a ready-to-use lead qualification list based on a sophisticated ML model.Impressive

But is AutoML really that easy?

Many data scientists argue that AutoML is overhyped. And as of now, it might be.


This is because AutoML currently requires a nuanced understanding that only a trained engineer has. For example, a trained machine learning engineer knows how to optimize a data model. They can easily evaluate test results and select the best way to tune parameters. These are not things one can intuitively deduce.


And for these reasons, AutoML currently is better suited to be a tool to assist data scientists with their development efforts. Meaning, it’s not quite ready for line of business workers.


Also—as a preview for what will likely be a future blog post—AutoML alone is not enough to operationalize machine learning models. Line of business teams will also require help from ML Ops / IT.Hmmm

These teams will need to deploy and manage things like the model lifecycle and workflow execution tools. This will ensure others on the GTM team are alerted about timely customer insights.


Turns out democratization takes a village…

Sometimes the future is now

Of course, every so often the stars align and you find a use case where a promising technology can be put to use right away.


That is the case for how Tingono is leveraging AutoML.Magical

We’re using AutoML so you can easily retain and grow customer accounts. To do this, we’ve:

  • Defined an appropriate scope. AutoML is most effective when the problem space is narrowly focused. Turns out the optimization of NRR for SaaS businesses is the right amount of focus.
  • Identified the right type of data. To make predictions on recurring revenue, we mostly use tabular data like product usage metrics or customer journey data. This is a great match for AutoML.
  • Identified the right amount of data. With the explosion of business data, enterprises now have loads and loads of both customer and product data. And it so happens that this amount of data is required for an AutoML system to effectively build, test, and iterate its models.
  • Shared our data science expertise. Our staff of ML engineers have the knowledge and experience to properly evaluate test results and build data models. So, we used this ability to build a proprietary ML engine. We've made this engine easy to implement and access, especially for those without an in-house data science team. Now you can quickly and easily access data models that help you generate revenue.

The result of this effort is your line of business workers are empowered by data models they didn't need to build (or understand). Specifically, this helps your GTM teams scale their efforts to retain and expand revenue.


So, while we’re all waiting for AutoML to democratize ML, there’s an interim solution that helps you drive more revenue. And we’d love to help you implement this. Let’s talk!