2019 was a year where the potential of AI entered the mainstream. According to recent analyst report, worldwide spending on AI reached $35.8 billion in 2019 with a 44.0% growth over the spend in 2018. Marked by innovation, experiments, and pilots, enterprises demonstrated the intent and mindset to harness the power of AI for business growth.

However, adoption was not without its challenges. As we move forward, it is evident that there is a need to refine, focus, and streamline the approach to AI adoption.

Enterprises have to think of the potential of AI in solving everyday challenges like being able to predict system failures, speeding up issue resolution or extracting and making use of information that remains locked away in datastores.

Explainability of AI solutions will become the cornerstone for enterprise AI adoption

Having outgrown its ‘emerging technology’ status, AI can no longer absolve itself of the liability for accountability, and transparency. Over the past few years, there have been discussions around ethics of AI, and the AI methodologies have acquired a “black-box” tag because of the apparent lack of transparency and explanation of results.

In the model building phase, the most significant change will be the increased focus on explainable and interpretable results. Apart from showing improvements in processes and operations, results produced by AI model deployments will have to be understood by human experts and make sense to users. Explainable and interpretable results and what-if scenarios will go a long way in improving the acceptance of machine-generated decisions.

Standardized processes for ModelOps will become integral to consumption of AI in the enterprise

2020 will see further industrialization and the consumption of AI in the enterprise. One of the key vectors that’ll garner attention will be the lifecycle management of models in an enterprise. ML Ops, as the industry calls it, will define the entire process of implementation from data preparation to model training to the governance of models in production and integration with the CICD workflows.

AI platforms will start to bring together the requirements of the entire lifecycle into focus and some will provide mechanisms not only to train ML jobs, set hyper-parameters, and auto-tune models, but also to oversee models deployment,
monitor metrics and versioning. Defined processes governed by a dedicated CoE, perhaps supported by IT departments, will begin to take shape as organizations develop scalable operational models backed by a robust framework. Consistency and reliability will be crucial to moving AI from an experimental workbench technology to a mainstream component of enterprise operations.

Standardized processes will evolve for testing AI models leading to increased reliability of AI workloads

Since AI models are different from traditional IT deterministic systems, that are deterministic, the testing and quality assurance requirements of AI systems are fundamentally different. Along with increased AI adoption, there is a growing
need to standardize the testing processes as well. Enterprises will look to create systems for unit and end-to-end testing while ensuring checks and balances for real-time monitoring as well. Third-party audits could become mandatory for compliance, regulation, and integrity, as cursory checks on AI output fall short of the diligence needed for consistency.

2020 will see a definite shift towards Outcome- based AI Spends

In all our client interactions and industry conversations, we have seen a clear trend emerge towards a preference to measure AI performance based on well-defined and explainable outcomes aligned with business goals. While AI itself could function as a layer of ambient intelligence, its performance should be assessed by the results it generates. On the one hand, the challenge for solution providers will be to create easily consumable AI that can offer customized views based on user personas and their KPIs. On the other hand, enterprises must delineate their business goals for every AI implementation, ensuring alignment with the core growth strategy.

These predictions offer an industry-agnostic overview of what is likely to be a year of refinement and consolidation for enterprise AI. It is important to note, however, that the AI adoption lifecycle is also heavily reliant on the industry vertical. High- tech, finance, and Insurance, for instance, lead the AI adoption wave. Others, like retail, have seen deployments only in specific areas like marketing and sales. That said, there is immense potential for disruption at every stage of the value chain in areas such as store space optimization, merchandising, demand forecasting, and pricing.

A combination of an increasingly challenging market landscape, demanding customers, and the exponential rise of data complexity means that enterprises cannot adopt just any piecemeal AI intervention. To be connected with strategic imperatives, deliver business outcomes, and unlock intelligence at scale, organizations need to rely on AI offered by enterprise-grade platforms.

Enterprise-grade AI platforms deliver exponential benefits over time with continuous learning capabilities. Enterprises that are diligent about their choice of platform and factor scalability into their decision can expect to accelerate time-to- value while setting the foundation for intelligence- driven growth substantially.

One way to reduce the uncertainty of returns on AI investment will be to partner with providers who offer product, domain, and consulting expertise, ensuring a well-rounded business initiative instead of a mere platform acquisition or licensing exercise. Most importantly, enterprises should have clearly defined metrics for AI success complemented by a forward-looking AI roadmap.