How to get started with AI in your business without breaking the bank?

AI-as-a-Service offers turnkey solutions for start-ups and SMEs

When it comes to guiding improvements in every sector, artificial intelligence offers a great deal of potential. But how do you get started? Should you hire a whole team of developers, or should you choose AI as-a-service (AIaaS) as a fast – and above all affordable – way to innovate? This was the topic of a fascinating evening organised by the Vlerick Entrepreneurship Academy.

Picture4

The V-Entrepreneurs alumni community of entrepreneurs was hosted by Orsi Academy, a pioneer in the field of medical innovation and robotic surgery. Our alumni listened to practical examples and received tips from Philippe Baecke, Professor of Digital Marketing & Big Data Analytics at Vlerick Business School. He is convinced that investing in AI is also a viable approach for smaller entrepreneurs and can create a great deal of added value for their company.

Picture0

The difference between AI, machine learning and deep learning

The term artificial intelligence was already launched in the 1960s by the American computer scientist John McCarthy. He described it as the development of systems that are able to simulate human behaviour. Initially, this mainly involved rule-based AI. Based on the available data, you convert certain human behaviours into rules as effectively as possible; these rules are then manually programmed in order to automate certain things. Take a chess program or a robot that can solve a Rubik's cube, for example. One application in the business world is Robotic Process Automation (RPA). This technology allows certain repetitive tasks to be programmed via software as successive instructions, allowing you to automate simple processes.

Today, it is possible to develop much more advanced AI models by intelligently linking underlying systems. An absolute prerequisite for taking the step towards machine learning (ML) is that you must have sufficient historical business data (CRM, ERP, etc.) to be able to train the ML model. This kind of model automatically detects the relationship between the input data and the desired output data, and it is the machine – not the human – that develops rules on the basis of this relationship. You can then apply these rules to new input in order to make predictions and take business decisions.

Possible fields of application:

  • Immoweb has developed a tool that makes an automatic estimate based on historical data regarding the sales prices, region and characteristics of your house.
  • Predictive policing allows predictions to be made on the basis of regional data about where a certain type of crime could take place, thus allowing police officers to be deployed more optimally.
  • Netflix is working on a model that will make it possible to detect fraud – in particular, sharing accounts with people outside your family – using AI tracking.
  • Insurance companies are developing models that can predict the likelihood of an accident based on sensor-driven data relating to driving behaviour, location, etc. These predictions can help to determine the premium.

Deep learning takes things a step further. Just as a human brain develops rules through the connection between neurons, a deep learning algorithm combines input data to develop new data, which is then recombined to form predictions. Initially, the model makes random connections that lead to random predictions, but you can use mathematical techniques to feed back errors, train connections and allow the model to learn for itself. Just like babies or children learn step by step.

Possible fields of application:

  • In a medical context, photos that have been annotated by radiologists worldwide can be used as input for the development of a tool that can help young radiologists with their diagnosis.
  • Using annotated photos of brand logos, GumGum helps marketers to measure the visibility of a logo during events and sports competitions and thus determine the value of a sponsorship contract.
  • Okay is testing a cashless shop in which purchased products are automatically recognised and charged for.


Rule-based AI

  • The rules and instructions are manually programmed by a person
  • Quick and easy to implement
  • Suitable for simple problems and processes

AI based on machine learning

  • You must have sufficient historical data
  • The rules and instructions are developed by a machine, and are sometimes so complicated that humans can no longer understand them (black-box model)
  • Suitable for highly complex processes
  • Offers endless possibilities

AI based on deep learning

  • Little data preparation required to make predictions
  • A very large amount of annotated input data is required
  • The model trains itself and makes autonomous decisions

As an entrepreneur, how do you get started with AI?

Even without large budgets, you can already achieve a great deal yourself as a start-up or SME and create value for your business. There are two possible routes.

1/ Off-the-shelf solutions

Building a model is the most expensive component. Many cloud providers such as Salesforce, Microsoft, Google or Amazon Web Services offer models that have already been trained. All you have to do is connect them via an API and you can start making predictions immediately, without the need for additional coding.

  • For example, the Einstein component in Salesforce allows you to use a food image model based on images of your own products.
  • SkyBiometry is a facial recognition and detection tool that produces a data file for uploaded photos.

The biggest drawback is the limited flexibility. You need to use the model and predictions that are available. However, it is a good way to find out whether there is value in a particular model before you start training models yourself based on your own data and for company-specific challenges.

2/ AI platforms

Platforms go a step further. They allow you to develop initial models relatively easily and without a background in data sciences. You can enter your own data, add annotations and create models with a high degree of accuracy. A good example of such a platform is Chatlayer, which develops AI-driven virtual assistants that allow companies to interact without the customer feeling like they are talking to a chatbot. Another example is Robovision, which develops AI applications for agriculture, health care and production that you can continue to feed and adapt as a client. Or you could join an ecosystem such as Sirris, which helps companies to make technological choices.

Get in touch!