Artificial intelligence — in particular, machine learning and deep learning — is becoming a competitive advantage across a range of industries, making them more resilient and allowing growth through the application of innovation.
Facing artificial intelligence from an enterprise perspective can be an overwhelming task, because not all artificial intelligence problems, technologies and solutions are alike, and choosing the right approach can make or break the introduction of AI.
Artificial intelligence solutions come in all shapes and sizes, from chatbots to face recognition systems, from algorithmic prediction in stock markets to generative imaging solutions, from world chess champions to automatic fruit sorting robots. You do not need the same resources, the same level of understanding of the technology, nor the same tools to build each of those solutions, and still, a lot of the time, companies try to use the same approach to these problems. The first thing to do is to choose the role your enterprise will play in this brave new world of artificial intelligence.
In the same way that you do not need experts in electronics capable of understanding the transistors that you use in the computers where your ERP runs, you probably do not need experts on the algorithms on which AI is based. Yet many times this is one of the first AI hiring decisions a company makes: let's bring in an AI expert, let’s hire a top data scientist. It would be like hiring a PhD in microelectronics to customize your ERP Sales and Distribution module: it may come out fine, but it will probably not be the best move.
An AI solution is a Business Solution
What we recommend is a different approach: start from your experience, find your pain points and your market position and try to find a way to integrate AI within your experience. Every AI journey should start from the same place: a business problem. That means you should find the right agent to lead the AI initiative: not the CIO but the CEO, not the technology experts, but the business experts.
After that, find the AI experts to start a conversation with your business specialists. That conversation should have always the business problem as a leading theme, and focus on what AI can do to define the questions AI needs to answer to solve your business problem. Then you can start thinking about the technologies and the strategy for implementation.
When it comes to Artificial Intelligence Strategy - the trick is to be nimble, be quick.
The approach to implementation should be as agile and modular as possible, taking into account the speed of change in artificial intelligence technologies. A difference of more than six months between the idea and the implementation can be fatal, because that is the speed of change in the technology. You should consider a staged approach, trying to test and decide on technologies as fast as possible:
- Proof of Concept - Minimum Viable Model. A Minimum Viable Model is an artificial intelligence based model that can answer your needs successfully. It should be the first step in the implementation and it should answer the following questions:
- Is your data good enough to use AI in the problem?
- Is there an AI model and architecture to solve the problem?
- Is there an AI algorithm to train the model in an acceptable time?
- Data pipelines: The data used in the model should arrive in the same conditions that the model training took place: data cleansing, data preparation and feature extraction should also be put into production.
- Retraining: The conditions for the process that the model addresses may change over time, and the model must adapt to those new conditions. You should monitor the model performance and retrain as soon as it changes, especially if the introduction of the model changes the process: Imagine a predictive system that allows you to vary the market conditions. Its mere introduction is changing the process that generated the data that was used to train the model. The model may no longer reflect the reality.
Artificial Intelligence Strategy: Where to start?
In spite of all the hype about Artificial Intelligence, we are still mastering narrow AI capabilities: solving specific, well-defined problems using specialized architectures with limited generalization capabilities. The main current applications of Artificial Intelligence in business do not carry out complete business processes but are integrated into the process: machine learning automates repetitive tasks in which there is human intervention and that cannot be done with a set of rigid rules. Artificial intelligence also allows to add steps in the processes that normally are not considered due to complexity or excessive cost.
Keeping this in mind, you should look for steps in current processes that are suitable for fast AI introduction:
- Image recognition
- Sentiment Analysis
- Natural Language Processing: Semantic understanding of unstructured texts
- User experience augmentation using chatbots
- Invoice matching
- Automated text translation
- Fraud detection
- Trend detection
- Risk assessment
- Customer segmentation
- Anomaly detection
- Power consumption optimization
- Logistics optimization
Any of these can be a very good starting point to start the journey to artificial intelligence, allowing you to find the right Artificial Intelligence strategy for your company.