Artificial Intelligence (AI) and one of its approach – Machine Learning (ML) – are not new. They exist since the 1950s but recently it really boomed the industry. There are three main reasons for that: Data abundance, Research Investment and Cheap computational power.
In essence, ML learns from historical data to detect and exploit behavior patterns automatically. Building efficient algorithms requires data, the more the better. Over the past few years, the creation of new digital data has been extremely rapid which varies from an ML perspective, opens tremendous opportunities to gain insight and, consequently, money.
Leading this gold rush, Tech giants invest a lot in Artificial Intelligence. Strategies combine hiring top algorithm innovators and acquiring promising AI companies. This results in the creation of powerful open-source ML libraries like TensorFlow (Google) or algorithm such as LightGBM (Microsoft).
These ML algorithms require a lot of computation power to process huge amount of data. While at the beginning of AI hardware did not allow it, today’s large computing networks make big data models possible. Whether using CPU, GPU or even specialized chips like the Google’s Tensor Processing Unit (TPU), training model with high computational power is cheap.
How does it affect your business?
Mastering Machine Learning might be for tech giant or specialized companies, but ML can be applied to any industry and most problems. Fraud detection is an obvious example of ML but it’s only the tip of the iceberg. Soon all sectors will start leveraging ML to solve problems and your business is probably no exception.
You have unique data. If not, wake up and start to collect them. The more you have, the more you can learn from it and make changes that will drive success to your business.
Cheap computational power is easy to access. To build your first model, you don’t need a datacenter full of TPU as a multicore Personal Computer is sufficient. The day computational power becomes the bottleneck, simply rent power on cloud platform on an hourly basis.
Democratization already started. Years ago, you had to hire a web specialist to set up a website for your company. With WordPress or any other online solution, you don’t need to have a tech guy in your team anymore. Machine Learning is still far from this stage but is on its way. Data Science Platform like RapidMiner on your computer or Azure Machine Learning Studio on the cloud are GUI-based integrated development environment allowing you to create model without coding any line. We see also more and more new AutoML projects allowing end users without expert ML knowledge to automate the design of ML models using state-of-the-art ML techniques.
You now have everything you need to start identifying behavior patterns and trends of your consumer, predicting price product change, improving your logistics or detecting a machine about to breakdown.
Where to start such initiatives?
- It might sound basic but top-management support is mandatory to get the right decisions made at the right time. Make sure key stakeholders actually understand what Machine Learning can bring to the company.
- Clearly define the problem you are trying to solve. The goal is not to use ML but to answer real questions that have an impact on your business:
- Which prospect has the highest chance to turn into a customer?
- Which customers will stop using your paid offer in the coming months?
- What is the most effective marketing initiatives for each type of consumers?
- You don’t need to hire a Ph.D. Your goal is to leverage Machine Learning, not to develop it. Find motivate and passionate staffs in your engineering teams who will be responsible to lead the development of your first models. Ideally, the team should have competencies on data analysis and have a scientific background but most importantly, they need experience in your business area. It’s a must to understand what AI can do for your firm.
- Forget the sexy word: Neural Nets, Deep Learning or whatever. You are not developing a driverless car or creating your own chatbot from scratch, right? Your journey just begins, so start with the basics: A Linear Regression if you are trying to predict a continuous value (what is the good price for this product?) or a Logistic Regression if you are trying to predict a discrete value (will this customer buy this product?). Only then you will start to use more powerful ensemble learning methods like Random Forest or XGBoost.
- Start with GUI-based integrated development environment or directly with open-source library in Python (scikit-learn) or R. Let your team use the tools they are comfortable with and learn the pros and cons of the different solutions.
- You need mentorship. Find someone in your network with experience in ML project or address consultants with Machine Learning expertise that can help your team setting up an AI strategy in your company.
The journey will not be easy and you will initially face deception with poor predications but the objective is worth it. Start experimenting right now, before one of your competitor gains competitive advantage. It is the right time to position your company a step ahead as Machine Learning is part of the future.
About Author:
Thibault is the CTO of Popety.io, a Lead Generation Platform in Real Estate which leverages Big Data and Predictive Analysis. He holds a Master’s degree in Computer Science & Engineering from the University of Technology of Compiegne (UTC) in France.
