Last year we predicted that AI would start having a real impact in the enterprise in 2018. This is already happening for DMI’s customers. Primarily it’s machine learning and not the more general AI that is driving success.
Most categorizations of AI and Machine Learning come across as either too technical or too theoretical, so here’s an overview of the top 5 categories where we are seeing real applications and results:
1. Image/Voice/Video Recognition/Classification
The most common use case so far as many of our customers have images (product catalogues, photos, invoices, etc.), voice recordings (customer care) and videos (e.g. security cameras) that have been manually analyzed by people to date. Machine Learning is rapidly replacing this entirely.
Solutions: We primarily use out-of-the-box cloud services such as Google Machine Learning, Azure Cognitive Services or Amazon Rekognition to automate some kind of process of recognizing digital content.
What if computers could provide better forecasts for demand, inventory or pricing? Now they can with sufficient amounts of historical data and the right models. Using data and ML, we can predict when a machine needs maintenance (IoT), forecast demand and supply, and improve personalization, pricing/discount models, loyalty, CLV, fraud, etc.
Solutions: The range of tools is broader since it’s often combined with where and how the data is stored. Therefore, we’ve used SAP Predictive Analytics, Microsoft R, Google Cloud Prediction API, Azure Machine Learning, Amazon Machine Learning, Shogun and Tensorflow.
Clustering can be a great way of finding micro segments, patterns in text, search results and categorize products or sales locations. The advantage is that clustering is unsupervised in the sense that the algorithm identifies data that can be grouped. We often combine this with predictive analytics in second stage.
Solutions: The best tool we’ve found is Carrot2, which is open source.
4. Conversational Interfaces (NLP & Chatbots)
We’ve been writing a lot about conversational interfaces as the demand for chatbots continues to grow. Thankfully companies are now beginning to become more realistic about how they work and their path to success. We also urge customers to look at the wider scope of communication automation with humans beyond chat, leveraging e-mail, voice and messaging.
Solutions: We’ve worked with api.ai, Pandorabots, wit.ai, springbot, Mindmeld, Chatfuel, Watson, Azure Cognitive Services and Artificial Solutions.
There are R&D type project requests that fall outside of the categories above, but it’s still in the early days.
A common denominator when talking to businesses, and especially technical teams, about machine learning or AI is that they spend too much time on the solution and not enough time understanding the problem. By applying human-centric design, design thinking and lean UX we’ve found that we can save a lot of time, resources and money on technical development. Thanks to some of the great tools mentioned above, the technical implementation is actually the easy part.
Magnus Jern, Chief Innovation Officer