Now we present the second part of the Algorithmia study on machine learning. The research shows the status of ML and the current possibilities for improvement.
Another finding of the study relates to the development time of an ML model. For 40% of the companies, this is more than one month. Only 14% develop a model in up to one week. For such a sophisticated program, this period initially sounds very short. However, it should be borne in mind that specific methods, such as rapid prototyping, can be used to speed up the development cycle and can even be a competitive advantage in some industries.
A comparison of the different company sizes shows relatively quickly: Larger companies have longer development cycles more often than smaller companies. This is often due to the more pronounced testing and approval regulations of the companies, which take up more time. Another exciting but expectable fact is that companies that work longer in this field have shorter development cycles than companies that are new here.
Therefore: short development cycles are essential to keep up with the rapidly changing market. Nevertheless, you should have patience here, as much can be compensated with experience in the subject matter, as speed comes with practice.
The next topic concerns the challenges associated with machine learning. Here, a comparison between the years 2018 and 2020 is used to illustrate the development. This shows that the biggest challenge in both years was the scaling of methods. This is particularly problematic for large companies, as scaling must be well thought out due to the number of employees affected and the high number of available use cases. The most significant leap, as mentioned challenge was the versioning and reproducibility of ML models. This is due to the much more complex models and the shorter development cycles described above. Further trials are the duplication of inter-organizational efforts, organizational alignment and senior buy-in and cross programming languages and framework support.
Therefore: The challenges of ML development are manifold and also relate to the environmental impacts of AI. A holistic view when carrying out such activities is therefore essential.
The next statistic deals with a permanently relevant topic in companies: money. A pleasing figure for the progress of AI: Over 70% of companies have increased their budget for ML development in the last two fiscal years, and only 2% have reduced it. Notably, 66% of the early-stage ML companies increase their budget by up to 50% to push ML further. The already established players in the market even increase their budgets by 51% – 75% (9%) or by more than 75% (7%) per fiscal year. Here the benefits and opportunities of machine learning have been recognized and thus also receive the necessary subsidies.
The study also distributed the budget increase across the various industries. It shows that there is a similar distribution in financial services (33% did not change the budget, 40% increased by up to 25% and 27% multiplied by up to 50%) and production (28% did not change the budget, 43% improved by up to 25% and 29% increased by up to 50%). In contrast, in information technology companies, there is a greater willingness to invest more money in ML. In addition to 47% of the companies that increased their budget by up to 25%, 20% also improved their budget by up to 50% and 5% of each company even put up to 75% or more money into the development of ML procedures. Only 21% of the information technology companies have not changed their budget.
Therefore: The finances of a company form the basis for all entrepreneurial activity. However, if the budget is sufficient, investments in AI activities should not be considered as a last option. Investments in AI can be used in many ways and offer immense added value in many areas. Furthermore, AI applications can be established for small amounts of money.
The last insight is about the success of machine learning and how it is measured in the company. This is mostly done with traditional business metrics like ROI and more technical evaluations that affect performance. 58% of the companies have said that a good ROI, customer satisfaction and brand affiliation are essential characteristics of ML usage. Another distinct finding is that data scientists and software developers tend to look at the technical metrics rather than the business ones.