1 Introduction

Artificial intelligence (AI) is a buzzword with many strong business cases that most enterprises are ready to embrace and, thanks to the abundance of computing power, AI has never been so accessible to support routine business processes. Cloud computing is one way for enterprises to acquire the infrastructure and computing power to run AI solutions, but accessibility to infrastructure is not the only prerequisite to implementing AI solutions. The shortage of qualified programmers, data scientists and machine learning experts makes it difficult for enterprises to develop their own AI solutions, and this only adds to the already high upfront costs of developing an in-house AI solution. These other barriers still exist, and put AI realistically out of reach for most companies.

Before enterprise support tools for AI development came on the market, enterprises faced high upfront costs for an AI proof-of-concept. Especially when enterprises are only beginning to develop their AI capabilities, it is normal and even expected that many of these proofs-of-concepts may fail. This risk of not recouping R&D investments alone is enough to put a hold on many in-house AI projects, but the shortage of qualified people only exacerbates the issue. Professionals cannot be trained fast enough, and the competition is fierce between competitors to attract the limited number of those already trained. Without support to customize and implement AI solutions, they would remain inexcessible.

AI-on-demand platforms provide a workaround for enterprises by increasing accessibility to high-level AI services, designing user interfaces that can be navigated without high technical knowledge of machine learning algorithms and coding, and provide resources and training for programmers to smooth integration with existing systems. AI-on-demand platforms will likely become a dominant business model to allow enterprises to develop and customize their own AI solutions at a lower cost.

AI-on-demand platforms typically provide their services via the cloud. Until recently, a lack of sufficient computational power and storage prevented entrprises from managing AI implementations at scale, thus the rise of cloud computing has made AI implementations often ubiquitous with the cloud. Implementing most types of ML models requires access to large datasets, both for training and as constant inputs when the trained model is hard at work. This data needs secure storage and high quality management, and cloud services provide the volume of storage needed,classification and labeling, and access to that repository of data without being restricted to a single location. Cloud computing also allows for scalability of enterprise AI proof of concepts. Cloud computing is a powerful infrastructure element that enables enterprise AI, inclusive of CPU, GPU, FPGA computation.

AI-on-demand platforms also provide a variety of “finished” AI services and tools for enterprises to choose from when designing their AI solution. These commonly are a combination of natural language applications, computer vision, decision support, and predictive analytics. Robotic process automation is also often included, although it does not always fall under the definition of AI. More mature platforms allow these tools and services to be combined, extended, and customized to the exact needs of each implementing organization.

Key institutional pain points that drive the need for AI-on-demand platforms are:

• The development costs of in-house AI solutions are high
• There is a shortage of qualified personnel to develop AI solutions
• Enterprises are just beginning to build up internal knowledge and experience with AI, which means that many POC projects may fail