Operationalising AI and analytics in real world environments

There is a vast difference between testing analytics in a lab and operationalising it. However, looking at how these technologies can be applied in a smart city environment can provide many ideas on how to best achieve this.

Johannesburg, 10 Jun 2019
Read time 5min 30sec
Kelly Lu, business solutions manager for advanced analytics and AI, SAS.
Kelly Lu, business solutions manager for advanced analytics and AI, SAS.

Analytics and artificial intelligence (AI) offer enormous benefits to organisations in both the public and private sector, but the journey to operationalising these technologies can be difficult. It is, for example, important to strike the right balance between choice and control, to reconcile the need to foster creativity and curiosity on one side with the need to govern the analytics lifecycle on the other side, and it requires adopting a deployment strategy that suits the needs of the business.

The difficulties involved are highlighted by the fact that research firm McKinsey suggests that only 8% of companies effectively scale their analytics. This is partly due to the choices made in their analytics environment, but mostly due to them having no clear deployment pattern. A lack of co-ordination also plays a role, since it is clear that the orchestration needed between IT, business and the data science teams is huge.

Kelly Lu, business solutions manager for advanced analytics and AI at SAS, points out that because these teams essentially don’t speak the same language, it becomes easy to end up with siloed initiatives, where everyone ends up doing their own thing. Add to this the lack of data science skills in the market, and it becomes obvious just how tough it will be to make the transition from a pilot project to one that is in production. This then begs the question: how does one go about taking AI out of the lab and into the real world?

“Here at SAS, we understand just how important the concept of the ‘smart city’ is, so we looked to the urban environment as a way to compose an example of taking AI and analytics into the real world. It is a logical approach, since if we consider history, modern cities are crying out for a smart solution. Just 120 years ago, there were only 12 cities globally with a population of more than one million people. By 2011, there were over 500 of these and over 20 mega-cities of more than 10 million people, so smart cities have never been needed more,” she says.

“In our example, we can look to how analytics are already being successfully implemented within urban environments, and this can serve as a foundation on which to craft a true smart city. Most cities have installed network sensors on the roads that record aspects like traffic and pollution levels. The trouble is that this information usually finds its way into the traffic department’s data warehouse, and is not looked at again. However, if this information can be used more effectively, it can really improve the city in a multitude of ways.”

Lu suggests there are many ways this data can be exploited, such as by building descriptor statistics that look at the data over time, offering information relating to trends, peaks and troughs. But beyond this, there are many other areas this information can impact, such as on pollution control, not to mention that having statistical data that will enable better traffic planning when it comes to major events.

Melissa Jantjies, Associate Systems Engineer at SAS Africa, adds that knowing about traffic peaks can be beneficial to citizens, enabling them to avoid driving during these periods. However, she continues, a truly smart city will do more than just tell people when the traffic is at its worst – it will also offer them alternatives.

“This is where AI comes into the mix, as it can assist with the next level up, which could include solutions like traffic light phasing, congestion charging and car-pooling as different ways to help overcome this,” she says.

“Of course, achieving this requires the city traffic controller and the data scientist to sit together and come up with ideas like this. These concepts can then be developed further and built initially using open source solutions like Python, where users can play around with a multitude of options. Then, once the solution is proven in the test environment, it can be implemented at scale in a production environment using proprietary offerings like SAS.”

Jantjies adds that it must be remembered that applying AI and analytics to just this one area of the city should, in the longer term, have much broader impacts, such as how reducing traffic – and thus pollution – will ultimately impact on the public health budget.

“Over time, the data gathered enables statisticians to understand the various causes of traffic, how different conditions impact it and – using historical data coupled with real-time information – to predict traffic patterns in specific areas. This can then be extrapolated out to other parts of the city, enabling controllers to determine which routes create the largest influx of traffic, so plans can be put in place to better control traffic from the outset.”

Lu adds that the first step in achieving all of the above, however, lies in ensuring the data being utilised is clean, quality information. Then, once the data analysis demonstrates where challenges lie and what is causing these to occur, the city can put plans in place to effect this.

“We have already mentioned how congestion charging and traffic light phasing can be instituted, but other possibilities include vehicle restrictions during periods of high pollution; alert systems so citizens know when and where traffic is at its peak; implementing priority lanes for electric vehicles or public transport; and real-time diversions and signage to make choosing new routes a simpler affair.

“And of course, the learnings from a use case like this can ultimately be applied across multiple industries, since many face similar challenges – like lack of access to data, not enough data scientists, and an inability to combine all the various analytics available into single view. In the end, taking analytics out of the lab and into the field effectively will require a complete analytics lifecycle, which can then be streamlined and ultimately operationalised in a way that provides the users with the ideal balance between choice and control,” she concludes.

Login with