ey-robot-finger-about-to-touch-human-finger

How Irish organisations can realise the benefits of AI at scale

Authors

7 minute read 05 Sep 2022
Related topics
Assurance
Technology
AI
Analytics and big data
Emerging technology
Data and decision intelligence

Machine Learning Operations (MLOps) has emerged as the blueprint for organisations to overcome their challenges and reach their true AI potential.

In brief

  • Embedding and scaling AI across an organisation to deliver value requires a shift from an artisan approach where data scientists and engineers are given creative freedom to an industrialised AI factory with rigour, standards, and processes.
  • Late adopters have an advantage if they can leapfrog the complex and expensive on-premise ML/AI platforms and go straight into the cloud, gaining efficiency benefits of the pre-packaged and mature cloud services.
  • Scaling AI using MLOps principles is not just about technology. People, process and culture will need to change as well.

Deploying Artificial Intelligence (AI) across the business at scale has transformative potential for organisations. Used properly, it can help businesses deliver seamless personalised experiences to customers, optimise production processes, uncover fraud, and accurately predict relevant market events and trends. Globally, the market for AI is expected to break the $500 billion mark by 2024 with a five-year compound annual growth rate (CAGR) of 17.5%¹. Ireland, too, is well positioned as a leader in AI readiness ranking 17th out of 160 countries globally, according to the Oxford Insight AI Readiness report 2021².

How EY can Help

Top-performing organisations with mature AI capabilities on average attribute about 5% of their earnings to AI³. Those organisations also see significant cost savings through AI especially in relation to operations optimisation and contact centre automation use cases.

 

However, a necessary precursor for the successful deployment of AI at scale is for organisations to have solid data foundations. The results of the EY Ireland Tech Horizon 2022 survey suggest that Irish businesses have some way to go in their maturity levels in relation to data, which may be holding them back in realising the full potential of the data they have. The study found that only 14% of Irish organisations say they are data centric, which may suggest that many organisations are grappling with the mammoth task of breaking their data silos and distilling it into a single source of truth.

 

A key finding from our research was that Irish organisations are also underinvesting in data and analytics compared to their global peers, with 38% of Irish respondents saying data and analytics are most likely to account for the largest share of technology investment over the next two years. This compares unfavourably with the 53% of global organisations who said so.

 

While these findings should rightly be the cause for some concern, it is important to cut through some of the misunderstandings that surround AI and its supporting technologies.

 

Demystifying AI

The hype around AI over the last number of years sometimes creates a sense of scepticism. There is, therefore, a need to demystify it.

As organisations define enterprise-wide strategies to harness the power of AI solutions, it is important to understand the types of AI and their business use case.

Almost every breakthrough in the development of narrow AI has been made possible by machine learning, a programming paradigm that finds patterns in data for a specific purpose. Deep learning is a subset of machine learning application that reprogrammes itself as it digests more data to perform the specific task it is designed to perform with increasingly greater accuracy without human intervention. It is specialised for very large and complex data sets. Therefore, it’s reasonable to use the terms AI and machine learning interchangeably in a business context.

A common pitfall for organisations is investing in a “silver bullet” AI project or product without fully understanding the “garbage in, garbage out effect.”

The performance of an AI system is only as good as the data it
consumes and therefore a solid data foundation is essential.

Successful organisations build their data assets piece by piece, by taking an AI use case led approach – prioritising data builds to deliver use cases with the highest value. This approach benefits from a compounding effect. As data requirements for use cases overlap, later use cases benefit from earlier ones. In addition to serving AI use cases, having a central data repository is essential for democratising data across an organisation, raising the overall data literacy of the organisation and helping to deliver open data use cases.

The key benefit of this approach is that it demonstrates business value early and helps get buy-in from key business stakeholders. The buy-in from business is imperative for user adoption and for managing change with innovative solutions being introduced. Agile and iterative methodologies such as design thinking and human-centric design have become very popular in the early phases of AI use case led approaches. Demonstrating business value early and substantial value over time is key to the success of data transformation programmes.

Effective use of AI

The question facing many organisations is where they should apply AI to reap the biggest return on investment. After all, embedding AI across the organisation is a non-trivial task that requires significant investment in talent and technology, as well as sweeping change initiatives to ensure the technology drives meaningful value. As part of an organisation’s data and analytics strategy or specific AI strategy, business use cases that deliver clear returns in terms of additional earnings or cost reductions should be prioritised.

It is no surprise therefore to find that that those areas of thebusiness that have traditionally provided
the most value andthose with the highest running costs tend to be the ones in which AI can have the
biggest impact.

In banking, for example, this would apply to marketing and customer engagement activities and contact centres. While in the utilities sector, it could be the optimisation of generators starts and stops with functioning wind energy.

Another way organisations can determine where to apply AI is simply to look at the functions that are already capitalising on traditional analytics techniques. For instance, a function using a rules-based approach to detect fraud and non-compliance may see a performance uplift of between 20% and 40% by replacing business rules with a supervised machine learning model⁴.

This may be easier said than done, however. All too often, organisations get stuck in what has been termed the “proof-of-concept purgatory,” with 46% of AI projects not making the leap from pilot to production⁵. In many cases, the data scientists were given the creative freedom to fabricate models from scratch and prepare data differently for each solution. Little thought was given to creating reusable data assets and integrating the use of AI into day-to-day business operations.

This highly bespoke and risk-laden approach to AI development is partly a function of decade-old data science practices that focused on delivering excellent algorithms but failed to link them to business value.

The days when organisations could afford to take such an experimental approach to AI pursuing scattered proof-of-concepts without a standard business-wide blueprint to deliver solutions are well and truly gone.

MLOps can accelerate the AI journey

Machine Learning Operations (MLOps) has emerged as that blueprint. It combines the technology, processes, and people within the right operating model to deliver AI reliably at scale. It standardises the full development lifecycle, drives consistency, improves transparency and auditability, and ensures that each team member focuses on what they do best.

Teams within organisations that embrace MLOps now look at AI development less as a set of individual tasks with each one starting from scratch and more as an automated, modular process which delivers solutions to meet the needs of the entire business.

The emergence of MLOps is of particular relevance to those Irish organisations that are lagging behind in terms of AI implementation. There are organisations that have begun investing in their AI journeys but find themselves locked into costly cumbersome on-premise platforms requiring obsolescent skillsets to develop AI models. These organisations are most at risk of failing to deliver on the promise of AI, but they do have a choice now. They can take the bold step to abandon those platforms and join the early and late adopters in embracing MLOps and moving to the cloud.


How EY can Help

Summary

AI maturity is no longer optional, it is directly correlated with the bottom line. Leaders of all organisations, regardless of size or sector, need to demonstrate ownership of and commitment to AI if they are to move forward. Embracing MLOps and moving to the cloud is the way forward to holistic AI adoption. And for AI to be used at scale, there is a pressing need to link it to business value.

About this article

Authors

Related topics
Assurance
Technology
AI
Analytics and big data
Emerging technology
Data and decision intelligence

EY refers to the global organization, and may refer to one or more, of the member firms of Ernst & Young Global Limited, each of which is a separate legal entity. Ernst & Young Global Limited, a UK company limited by guarantee, does not provide services to clients.