Machine Learning As A Service:Integration Of Machine Learning Into Businesses

Machine Learning has become an important technology for almost all industries.Applied in businesses and combined with Big Data,Machine Learning already creates a measurable return on investment and is seeking real productivity increases which are leading to competitive advantages within markets.Although Machine Learning became relevant to the market competitiveness of companies,many businesses are still struggling to adopt Machine Learning.Many studies have already analysed the phenomenon of Machine Learning in businesses,concluding that companies are facing various kinds of challenges.With the introduction of Machine Learning as a Service(MLaa S)expectations for overcoming integration challenges are rising.This study aims to analyse whether MLaa S can indeed support companies to overcome Machine Learning adoption challenges and to identify which factors influence this adoption.We evaluate MLaa S as highly relevant for business managers who are in charge of a company's IT-innovation strategy.While multiple studies are covering MLaa S from a technical perspective,no business-focused study has been conducted.Concurrent,in our literature review we have identified 10 Machine Learning adoption challenges which are discussed by other studies,none of them taking MLaa S as a possible solution into consideration.Consequently,we have seen the research gap in this field and defined our research questions as ”What factors influence the business adoption of Machine Learning with MLaa S?” We aimed to answer this question by defining four sub-questions:SQ 1: What are Machine Learning requirements from companies?SQ 2: What are MLaa S market provider and product categories?SQ 3: What is the MLaa S market situation and what are key trends?SQ 4: When is MLaa S suitable for companies?We base our qualitative study on the diamond organizational model which serves as the underlying research framework.Our data collection follows a multi-method approach,one data source being company case-studies,another being a comprehensive MLaa S provider market analysis.While the five company cases are selected by non-probability sampling due to the special nature of this research topic,are the five MLaa S providers selected based on their market leadership.The case studies aim to analyse the Machine Learning adoption situation of each company,while the market analysis aims to capture today's MLaa S product portfolios.We followed a thematic analysis approach to processed the qualitative data of both sources.We aim to answer our research question by comparing company requirements with the current market situation.First,we focused on Machine Learning requirements from companies by looking at the case studies.In total,we identified 13 requirements,most significant being the requirements that MLaa S should be able(1)to close the Machine Learning knowledge gap,(2)provide a portfolio of algorithms,(3)adapt to the individual company situation and(4)handle forged data.Second,we looked at the MLaa S market and identified categories for providers and products.We distinguish between three MLaa S product categories(1)Machine Learning Cloud Instances,(2)Machine Learning IDEs and(3)Machine Learning APIs and decided that this study will not further focus on(1)due to the similarity to general cloud computing.Furthermore,we classified the five market leading providers into categories based on their market position and customer target group.The three providers categories being(i)innovation with technical focus,(ii)market speciality and(iii)simplified for general user.Third,we summarized the insights we gained from the market analysis and concluded which trends we see in the MLaa S market.While we see industry standards,almost every provider is offering special services which are focus on specific customer target groups.A significant trend is a provider-driven development of innovative software and hardware solutions with the ambition of gaining a competitive market advantage.Such solutions may reduce operational costs(reduce energy consumption),improve model performance(faster interference)or aims to enable general users to develop Machine Learning models(web-development suite).Four,we focused on aspects that predestine companies for the MLaa S suitability.We compared previously defined company requirement aspects from our research framework with the available MLaa S product portfolio.We come to the conclusions that the aspects of Predictable Costs,Scalability and Data Security can be fulfilled by MLaa S.Not supported are the aspects of Data Quantity and Data Quality.Aspects where a further differentiation between MLaa S products and individual business situations is necessary are Gaining ModellingInsights,Reducing Costs,Automatable Access,Close Knowledge Gap and Individual Business Case Adoption.We used these results to develop a suitability model,which aims to support companies in selecting a suitable MLaa S product category.In our discussion,we argue that the most critical factors for Machine Learning adoption with MLaa S remain in a company's Data Foundation and the workforce's Machine Learning Expertise.We have seen,that companies with supporting management can build up Machine Learning infrastructure relatively easy with MLaa S solutions.However,building a solid Data Foundation may take longer,as product cycles for sensor integration or adaption of business processes have to be waited for.Furthermore,we have seen that ML APIs can substitute lacking Machine Learning Expertise,however,our research also shows that such standardized APIs could in 4 out of 5 companies not cope with the required task modelling complexity.In contrast,ML IDEs can cope with this task modelling complexity.Our research shows that the data analytics teams say their productivity got increased by using ML IDEs,still they also that Machine Learning Expertise remains required for using ML IDEs.Therefore,we conclude that companies with Machine Learning applications in place,and consequently a Data Foundation and Machine Learning Expertise available,can profit more from MLaa S than companies without.This is the first business-oriented study covering MLaa S.While we see limitations in the sample size of the case-studies and the MLaa S providers we focused on,this study can serve as a reference for business managers who have to decide about a company's Machine Learning strategy.We recommend Business Managers to evaluate their company's MLaa S suitability with the models we proposed.We emphasize that MLaa S should not be looked at as a tool that can substitute lacking adoption factors like Data Foundation or Machine Learning Expertise.However,in a suitable environment,we have seen that MLaa S can have a positive influence on the business adoption of Machine Learning.