Data + AI = BOOM!¤

The two most important words for the future of businesses happen to also be the two of the most hype words at the moment: data and artificial intelligence. Research done by Narrative Science shows that 62 per cent of organizations will be using Artificial Intelligence by 2018. The number is way higher for startups: many of the tech-startups, regardless of industry, have business models based on identifying a problem and applying some sort of artificial intelligence technology to solve it.

The number of companies that brand themselves as artificial intelligence companies has exploded lately — it is clear that using some sort of AI technology in your business has become a branding tool that can attract both investors and the best team.

Looking beyond the fad, artificial intelligence is becoming more prominent for businesses as the data growth continues. Companies in most industries use significant resources on managing their data and most expect this to rise further:

It’s a fact: we live in times with exponential data growth. Take a moment to reflect on the following:

• 90% of all the world’s data has been produced in the last two years
• Forecast: the amount of data we will produce in 1 week in 2025 will be equal to all the data we produced up until 2015

If you respond to visuals better than text, the following chart should help put things into perspective:

It is clear that what we call big data today will seem very small in the near future. Data is the super-fuel for business growth, but only if you handle it right. The winners in the future will be the companies that manage to build unique data sets enabled to provide continuous feedback to their businesses. Companies with established “feedback loops” do not need to guess what their users prefer or which part of their product they should develop further — they use the collected data to innovate and provide excellent service to their customers. Establishing the “feedback loop” can be an expensive affair, often requiring changes in the core of traditional organizations with pipeline business models, but they might just be necessary to survive in the long term.

The exponential growth of data (and the power of processing) has made breakthroughs in artificial intelligence possible. So far, machine learning has received most attention within AI, both within research work and business application. The idea of creating an algorithm that can read unstructured data, learn from it and then use its knowledge to predict an actionable insight is highly attractive for businesses in all industries. The effect of machine learning on a business can be profound. Think about the development process in a traditional company:

Business development identifies opportunities, product managers write detailed specifications that the developer uses to programme the desired output. The code is then continuously maintained and each change or extra feature is implemented manually based on a prioritized back log.

When UBER team was about to launch UBER Eats they wanted to inform their customers about the expected delivery times. They took a traditional approach and setup a linear regression model to compute the expected delivery time using the distance between the customer and the restaurant, the average speed and the time it usually takes to prepare a meal:

Expected delivery time = distance*average speed + 12 (average time (min) it takes to prepare a meal)

The model, even though statistically significant, did not consistently provide trustworthy estimates. The team could try to further adjust the model by adding or dropping variables to try to improve the calculated estimates. The problem is, it would probably never become good enough.

That is when the team decided to apply machine learning to solve the problem at hand. At that time, over 20 000 deliveries had already been made by UBER Eats. This historical delivery data was fed into the machine learning model in order to predict the expected time it will take for each delivery in the future. Overnight, that improved the delivery time estimates by 26 percent. Further, every time a delivery is made, the data collected is fed back into the model providing continuous feedback that further improves the estimates. Note that this process happens with zero involvement from humans.

To take it one step further, if your model is good then the problem you are trying to solve with it should eventually disappear. Establishing a feedback loop that not only amends your current model, but also identifies new areas for improvement is critical. Think again of UBER Eats: predicting delivery times based on historical delivery data is a good start, but the real magic happens when you start feeding your models with other types of data like real-time traffic data and high-resolution satellite imagery.

Lastly, in order to scale innovation one needs to collaborate. Collaboration enables exponential development that no organization alone is capable of. Think of Amazon’s intelligent voice assistant Alexa. When Amazon opened up Alexa to developers a year and a half ago, it had 14 skills. Today Alexa has more than 10 000 skills and considering the development speed at the moment, it is probably not long before the milestone of 100 000 skills will be passed. And since we used UBER as an example throughout, they just released a huge amount of historical traffic data to general public for everyone to innovate on.

Many traditional organizations are starting to realize the power that data can have on their business. Many will take the needed steps to understand their data and enable it to create actionable insights. Others, will be forced to do so or risk going out of business in the near future. One thing is for sure, these will be exciting years for many of us embracing the change and pursuing the opportunities we discover on our way.