Being digital is no longer a luxury, it’s the need of the hour. As more and more organizations become digital, large volume of data gets generated from customers, operations, manufacturing, supply chain etc. Monetizing these data is an absolute need as organizations are looking towards newer streams of revenue and profitability. Adoption of AI has been a major trend across organizations in realizing these sources. However, a preview into the adoption of AI and how it has helped realizing these objectives has given a mixed response. While very few organizations have realized benefits through production deployments and embedding AI in their business processes, many organizations are still in a research and development (R&D) mode. Some of the reasons attributed to this are lack of talent, lack of data, data quality issues, poor metrics baselines, and organizational culture. There is also an inability to identify the right use cases which create value and to measure value. Many AI models are not compatible to existing legacy systems/architecture, and firms do not have scalable platforms to cater to AI workloads and AI projects. There is a lot of noise around the real issues hampering successful AI adoption.
What are the scenarios of AI adoption across organizations?
Scenario 1: In this scenario, AI is being adopted in a siloed and tactical way. Some functions like customer management, marketing, and/or operations define a use case and engages the analytics team within the organization or a service provider. Data is provided in an offline mode. Data engineers and data scientists prepare data, perform exploratory analysis, build models, and provide output in an offline mode. Business presentations are done depicting potential benefits and the project is closed. In these kinds of AI projects/PoCs (proofs of concept) it is unlikely that the business would have realized any value. The project would not have created any positive vibes in the organization and stakeholders would conclude that AI has not created the intended impact and is therefore, not very beneficial. AI investments are not coherent and a clear strategy on AI adoption is missing here.
Scenario 2: Here, the organization has graduated from PoCs and has implemented some use cases in production. Models are exposed as APIs, web services, or other services. The use cases have shown some initial results. However, there are challenges in the management of AI models, model performance with new data, model refresh issues, active learning, poor inference response times, and infrastructure scalability issues. Though minimal value is realized here, that does not justify accelerated investments in AI. So despite increased AI investments, the organization goes through a slow process of AI adoption.
Scenario 3: The organization has matured in AI adoption. Models are built on premise or on the cloud and deployed as APIs or services. However, there are issues in having the right model pipeline which helps in real time processing of data from hybrid environments. This creates real time decisioning and actioning. Value realization is there, however operational challenges are morphing the positive vibes.
There are many more such scenarios which are not leveraging AI to its full potential and are operating at suboptimal level. Organizations will be at different levels of maturity in their AI journey and it would be good to look at the level of maturity at which they are operating. The diagram below provides a view of various maturity levels in the AI journey. Based on these maturity levels, Scenario 1 would be at Initial level of maturity, Scenario 2 at Defined level of maturity and Scenario 3 at Repeatable level of maturity.
How to overcome these challenges and make AI mainstream?
AI adoption is at various stages of maturity across organizations. To overcome some of the challenges highlighted above, especially the systemic ones, it is required to adopt an enterprise view to AI adoption. A platform approach to AI is essential to ensure mainstream AI adoption from a technology and business process perspective. These platforms have inbuilt features for end to end model pipeline consisting of data ingestion, transformation, model build, validation, visualization, deployment, versioning, security and AIOps (the equivalent of DevOps in AI world). These platforms help in productive collaboration among analytics leaders, architects, data engineers, data scientists, business analysts, and operations specialists. The focus is on innovation, rapid iteration, and bringing in use cases to reality. The hard part of wiring things together is handled by the AI platforms. A well-defined strategy aided by a robust platform and an AIOps approach would help organizations to mature their AI capability and move up the maturity levels to operate at the Optimized level of maturity.
Per leading industry analysts, due to widespread AI adoption, the platform market is growing at a CAGR of over 30%. AI platforms are provided by various providers and they are targeted towards specific customer groups based on their preference for on-premise or cloud deployments. Following are some of the prominent offerings in the space:
- Cloud-based AI platforms and AI solutions from Google, Amazon, Microsoft, and IBM among others
- OEMs like Philips, GE, Siemens, AT&T, and Nokia having their own AI platforms. They bundle AI solutions along with their products and services
- AI platforms from Global IT service providers like HCL Pangea, DryIce etc.
- Startups providing AI platforms like Datarobot, Dataiku, H2O, Rapidminer, Algorithmia, Seldon
- Open-source ML and AI platforms like Acumos, Kubeflow, and many more
Choosing the right platform
While various AI platforms are available, it is important to evaluate these platforms on various dimensions and look at their suitability to the organization’s needs. A strategy consulting exercise can be conducted in conjunction with an AIOps approach as a precursor to define AI strategy, potential use cases, “as is” business processes, data availability, “to be” processes, roadmap definition, and business cases among others. This would serve as a key input to evaluate various platform options and test the feasibility of implementation of use cases. Finally, the most suitable approach can be selected to fulfill the organization’s needs.
It is important to understand how the capabilities come together to meet the technology and business expectations. As depicted above, an AI platform should have most of the following building blocks:
- Data exploration
- Model building and optimization
- AIOps - Model deployment, version control, monitoring and governance
- Orchestration layer for infra-scalability
- Authentication and authorization
- Model zoo/marketplace for reusability/ transfer learning
- Open data APIs for hackathons/innovations
The AI platforms are engineered in such a way that it allows organizational focus on the business use case realization rather than worrying about the operations around it. It helps in the standardization of tools, processes, and automation complex tasks, thereby moving the organization right up to the Optimized level.
The journey from a PoC to production would be challenging for any enterprise. AI projects are different from traditional projects and would need the engineering rigor and the right set of tools to be brought in for seamless integration with architecture, data sources, and existing systems and processes. AI platform providers have understood these unique challenges and have packaged relevant features for quicker onboarding and value realization. These platforms are maturing at a rapid pace keeping in pace with the needs of the industry. While AI platforms are only a technology enabler, it must be supported by leadership sponsorship, cross functional teams, a well-defined process, cultural acceptance, and a shared vision to make the transformation a success. Platform adoption would accelerate AI deployments, reduce time to market, and create favorable business impact. Mature your AI capability and accelerate your ROI from AI initiatives with AI platforms!