Alena Gaybatova and Ekaterina Stepanova, experts in the direction of analytical solutions of KORUS Consulting Group .
Everyone wants to make money and save with data: applying ML methods even on one project helps to achieve significant savings or even revenue growth. But in order to feel the effect and not fail the implementation, you need to take into account the difficulties and avoid managerial mistakes. Using an example, we will tell you how to make sure that machine learning algorithms do not make mistakes.
Machine learning - only 5% of the project resources. But the complication of ML logic can lead to an increase in implementation time, and improper planning of data collection can lead to inaccurate analysis that can become useless and expensive. Why is this happening?
The problem of expectations
The company heard somewhere that a neural network is the solution to all problems. At the same time, the quality or volume of data leaves much to be desired - it is simply impossible to implement a model on such conditions. For example, it takes about a year to accumulate data in retail or manufacturing, and if the necessary equipment is not available or some of the processes are not digitized, then even longer.
To avoid confusion, we recommend negotiating specific results with a caveat to the requirements for launching the system, rather than a vague income opportunity. Such illusions are experienced not only by business, but also by the developers themselves. Sometimes business analysts expect models to work well based on what they read in complex technical articles. Unfortunately, such texts write about algorithms developed on model data, not real data.
Therefore, the results of the project should be useful, easily interpretable and validated by metrics and business experts.
False premises
- ML-. , 2019 90%. 2020 , — c , .
. .
, , , . : . , , . , . 2020 - , , . .
: , .
: - , . , , . , , .
. – . .
.
1. ? , ? ?
value . , . , 100%, , «» . : 50% , «».
2. ML? ? , – ?
, . - . , .
3. ?
– , ML-. , . , -.
4. ?
, ML – . - ML, , . , , . , . , , .
ML is a great tool for optimizing work, solving non-standard problems and analyzing large amounts of data. It is important to consider many aspects for its use before and after implementation. Hopefully, this brief overview will be able to warn against unwanted situations and minimize the problems with using ML, so that you can enjoy the additional benefits of the technology.