Understanding the foundations of successful AI

Despite the disruption to the overall economy and most industries in the wake of COVID-19 pandemic, there are sections of IT that are booming.

As the Wall Street Journal reports, spending on PCs, mobile devices, and storage infrastructure is expected to decline, but artificial intelligence (AI) will surge as it is seen as a way to assist organisations with cutting costs and boosting revenue. As we noted in the previous article, AI can indeed help boost productivity and improve the customer experience, and that makes it an ideal tool to help navigate disrupted conditions.

adobestock-300893166.jpg

For many organisations, this interest in AI is new -- or, at least, they have seen the desire to adopt AI solutions accelerated from a long-term goal to something that is immediately pressing. The struggle that many organisations now face is that AI is an end-goal and requires sound foundations and a well thought out approach to data and analytics first. The first step for many organisations in their efforts to adopt AI will be to bring an operational analytics oriented framework to their data environment first.

Introducing operational analytics

Being able to incorporate analytics into high-volume operational decisions in real time is critical for the success of any AI project that an organisation might undertake. As discovered by SAS in a recent survey, organisations currently deploy less than 50% of their best models, and take more than three months to deploy 90% of them. What that means for those environments is that a proven ROI from AI can only ever be a pipe dream.

It's the organisations that can put these models into production in an automated and rapid manner, but also maintain governance and trustworthy data management, that will be the ones to find success with AI.

The seven key steps for managing operationalised analytics are as follows:

  • Register – First, an organisation should build a centralised model repository with life cycle templates and version control capabilities to ensure transparency, governance, and traceability with every model.
  • Deploy – At this point, an organisation integrates analytical models into a production environment and can start using them to make predictions.
  • Decide – Here is where the models are incorporated into a decisioning engine -- ideally one that allows non-technical stakeholders across the organisation to begin automating and bringing AI into their tasks.
  • Act – This is the point where users can begin publishing the decision flows determined in the previous steps. The decision flows should be able to be executed in real time, with the ability to scale as necessary. Data volume at this point is high so the environment also needs to have been prepared for data throughput.
  • Measure – The system should be structured for capturing feedback on the outcome of actions and the performance of decision flows in real time.
  • Monitor – The system should also have the capacity to allow an organisation's stakeholders to monitor ongoing performance of the environment. This includes the production of performance reports and actionable alerts.
  • Retrain – Finally, the system needs to have the capacity to retrain a model with new data if the model's performance begins to degrade. There should also be the option to revise or replace models entirely if necessary.

Read more information about how you can get the most from your AI investment.

Overcoming the hurdles to operationalising analytics

Operationalising analytics involves taking an all-new approach to analytics, and one that has a similar structure to the one taken by DevOps with application development -- we can term it "ModelOps" in kind. As with anything that drives towards AI, creating a ModelOps framework is a complex process and can easily fall by the wayside unless it is carefully planned and executed.

As noted in InformationWeek, organisations find operationalising analytics difficult in a number of areas: Analytics is considered a technology problem, rather than a business challenge, it's difficult to get buy-in from the executive as a consequence, and there's the potential that analytical results aren't as transparent as they need to be.

Having a partner such as SAS that has a history in structuring and delivering operational analytics is key to addressing this challenge. As Gartner noted when placing SAS as a "leader" in its Data Science Magic Quadrant: "Customers choose SAS for its enterprise-grade platform capabilities and support for the entire analytics life cycle — from exploration to modelling and deployment."  SAS' expertise in this space can help an organisation get the necessary executive buy-in by helping to articulate the benefits of operationalising analytics as part of the journey to AI with a proven roadmap and process behind it.

The shift towards operationalising analytics is a significant one -- it involves taking a process that was previously structured around taking months to deploy to, instead, structuring it as something that is iterative in nature and performed in real time. Furthermore, it takes something that was previously held within the data science team and makes it transparent and accessible to a much broader range of stakeholders. Gaining organisational buy-in on an initiative this significant can be a big challenge, but for organisations that do so, the holy grail of effective AI lies at the other end.

Read more information about how you can get the most from your AI investment.