Why Auto-ML tools will drive up the cost of Machine Learning

Automatic Machine Learning (Auto-ML) is the toolset that enables people who aren’t specialists or engineers to solve problems with machine learning. And it’s on the rise. Tools like H2O, DataRobot and the APIs provided by the likes of Google, AWS, Microsoft and even IBM’s Watson are opening up the field of data science for many companies. It seems like the industry that was helping automate other people’s jobs is looking inwards.

People have asked me ‘aren’t you worried about losing your job?’. And the answer I give always surprises people - I think these tools are great! And for several reasons, main one being that they will actually raise the value of data science within an organisation.

Let’s get started with why these tools are amazing and you should be trying them out.

1. Makes you frame a problem correctly

AutoML tools are great at processing relevant data, showing you relationships between variables and making predictions. But in order to use these tools you will need to frame the right question to ask and gather the data into the correct format. This allows an expert in a subject to focus on where they can provide the most value - defining what the solution should look like and what information would be most relevant to the problem.

2. Provides a good starting point

Even if the solution provided by the AutoML tool isn’t the absolute best model, it will serve as a baseline that can be improved upon. A model doesn’t have to be perfect before it can be used. As a starting point the model should reveal new insights into the process you are trying to explain which you can then refine by including more relevant data or by hiring a specialist after an AutoML tool has provided a baseline.

3. New revenue stream for existing companies

One of the biggest advantages that enterprise companies have over incumbents is data. They have been operating for a long time and have data on what works and what doesn’t (even if by accident rather than design). A great way for companies to leverage this data is to apply advanced analytics on existing processes. Rather than a huge investment in a new division a different approach is to see what problems can be solved using AutoML tools with existing data and processes. Leveraging these tools in the right way can increase revenue (e.g. recommending new products to existing customers) or decrease costs (e.g. better forecasting demand to reduce the amount of inventory for a supermarket).

So why does all this raise the value of Data Science?

1. Replaces some of the “data scientists”

There’s a lot of ‘data scientists’ that won’t be able to beat the baseline established by an automated system. Now, I don’t mean the data scientists feeling imposter syndrome, I think any good data scientist has to feel that at some point. I’m talking about those that don’t feel imposter syndrome. The ‘data scientists’ doing k-means clustering in Excel using the linear regression function and iterating by hand (which I have sadly witnessed). Their days are numbered, which is a great thing for the industry on the whole. I’ve seen lots of money wasted on projects going to consultants that couldn’t solve problems. If you can get a decent enough result with an automated solution, I see the median salary for a data scientist actually going up (partly because the bottom falls out).

2. Commoditises the quick wins

These products are designed to be easy to use and very cost effective. As their use becomes ubiquitous you’ll see more and more companies using these tools which in turn drives demand for these skill sets. By commoditising quick wins the types of problems a data scientist will then work on will be the larger and more strategic problems for the business.

3. Non-homogeneous solutions win

This leads us to the final point, if you and your competitors are using exactly the same systems and coming up with the same insights you will need a way to differentiate. This is where the real value of data science come into play, where the market is competitive. This eliminates the race to the bottom of the commoditised toolsets as companies look for ways to add value to the baseline models.

The automated machine learning revolution is on it’s way which will open up a lot of doors for companies looking to adopt a data driven culture. Data scientists already working in the field should embrace this software as a productivity tool and promote their responsible use within organisations.

If you’d like to keep up to date with all that’s happening with Verge Labs be sure to follow us on Twitter.


Previous article Monthly Roundup - April 2018

Here's our roundup of AI and ML in the news we found wanted to share and offer...

Next article Video: Interpretable Machine...

Interpretability matters! Machine learning models have been getting more...

Get in touch

To contact us, please fill in your details.
We're fast responders.

Do you have data?