Free ML Consulting

No contracts to sign, no payment needed, no time-wasting meetings

What We Offer

We are the only company in the world that offers machine learning consulting for free. Just like a lawyer who takes a case at no cost and only gets paid if they win, we do the same with machine learning.   Our offer is simple:  You only need to pay us if we increase the accuracy of your ML model.  And, you decide how much to pay us for it. 

If we can't do better than your current ML model, you pay nothing and get all the work we did for free (we will show it to you), proving that what you have now is the best. It is a win/win for you. We are the ones taking all the risk.

How can we make such an offer, you ask? Like most things in life, it is complicated. But a quick summary is:
1. We use an automated machine learning (AutoML) program that we developed to create a baseline model 100 times faster than usual.
2. We then have a team of data scientists from all over the world try to improve the score the AutoML program gives us, by making manual improvements to our AutoML platform. Don't worry, though, we never give any of them access to your data. We basically give them fake, but similar data to work on, and then we take the best solutions and try those ourselves with the real data.

Here is the longer, but more interesting explanation:

Data science is very inefficient. If you ask 10 machine learning experts to build a model for your data, they will almost all do it differently, and all get very different results. Any data scientist can spend weeks or months coming up with a model for you, but it is almost impossible to know how good that model is compared to what else is out there. Most people are happy just to have something that works well for their business. But what if that model could be 20% or even 50% more accurate?

In ML competitions (such as,,,,, and, almost all of the improvement over the baseline accuracy score comes from feature engineering and trying out new ways to handle the data. It does not come from traditional AutoML methods such as:
Testing more algorithms/neural networks
Testing different feature selection methods
Testing different data cleaning methods (scaling, handling missing data, etc.)
Testing various ensembles (voting/stacking/blending).

All of that experimentation is helpful but insignificant in the end result. That is why most people say AutoML will never replace a good data scientist. On the other hand, machine learning experts waste a huge amount of time with each new project just doing all that basic grunt work to get it working, instead of spending time looking into the cutting-edge methods that might help them the most. There is almost always something better out there. In ML competitions, for example, which usually last for months, there is always a frenzy of improvement even in the last few minutes before it closes. Nobody ever runs out of ideas for something new to try.

Typically, with a data scientist or ML consultant, you would be paying them for many weeks of work, to do what may take us just 1 hour (plus having our AutoML program crunch the numbers for a day or two).  We will then have multiple data scientists try integrating more advanced solutions with our AutoML baseline. We do not give them a copy of your data, instead, we use a GAN (generative adversarial network) to create synthetic (fake) data that is similar to yours, but does not actually contain any of your original dataset.  Here are 2 recent papers about that topic:  Generating Artificial Data for Private Deep Learning and Privacy-Preserving Machine Learning Through Data Obfuscation.  Also, these data scientists will still be using our AutoML program so as not to waste time, but they are able to customize it to do whatever they need to increase the accuracy.

We do not care about the specifics of your business. We do not want to meet with you.  We simply want to make you more money, so we can get paid. If that is something you are interested in, please email us at