Full text access here
This document analyzes recurring mistakes and problems organizations face when conducting internal Machine Learning projects.
Machine Learning (ML) is a combination of technologies with the capability of providing insights from large amounts of data faster than a human. Its business value is currently in a state of inflated expectations, which can cause problems when organizations undertake internal projects without prior knowledge.
Thingbook.IO personnel have analysed the lessons learned during dozens of ML projects and undertaken a review of available literature, followed by a set of workshops with a sample group of customers. The following conclusions were drawn:
1. For companies who decide to implement ML projects, there is a direct correlation between a miscalculation of Machine Learning value and the level of experience the organization has in it. Inflated expectations come from companies who lack ML knowledge without realizing that lack, at both the technical and managerial levels.
2. Data related mistakes, such as not enough data, low quality of data, or biased data most severely impact the accuracy of results. This is a key differentiator with standard software development projects, where data related problems are usually identified during the development phase and are difficult, if not impossible, to foresee during project planning.
3. It is difficult to accept the value of long-term Machine Learning projects, because there is no obvious Return on Investment (ROI) or the business case is unclear.
This document is structured as follows:
An exhaustive review of the machine learning literature is presented including a list of both technical and managerial recurring mistakes. A managerial view of Machine Learning is also provided, as well as a short comparison between software development projects and Machine Learning projects.
A “de facto” standard procedure for implementing machine learning projects is presented.
The companies examined are split into groups based on their level of Machine Learning adoption, its maturity and the role of Machine Learning in their portfolio and processes. Note: This document does not include companies commercializing Machine Learning Services or Data Analysis Products as their core business but focusses on how non-analytical companies can benefit from Machine Learning.
A conclusion is made about the correlation between identified mistakes and their impact on companies based on their categorization.