Improving the return on your AI and ‘big data’ investment
A machine-learning solutions expert cautions against blindly spending huge sums on large data sets, tech stacks and other AI-driven decisioning resources
Artificial intelligence is rightly hailed as one of the next big economic frontiers.
Recognising this, the Singapore government has identified AI and data science as key “frontier technologies” for growing the digital economy, announcing last month (May 2017) that it will invest up to S$150 million to boost the country’s AI capabilities over the next five years.
It will be some time before we see some of the flashier examples of machine learning, such as humanoid robots or self-driving cars, being widely used.
But AI is already driving a quiet revolution in the data-rich fields of financial services, telecommunications and e-commerce.
In fintech, for instance, robo-advisors are expected to be managing US$8 trillion in assets globally by 2020. And big banks and investment firms have begun using machine-learning solutions for everything from tracking their risk exposure to monitoring phone calls for signs of insider trading.
At Experian, we provide machine-learning solutions and Big Data-driven analytics for companies in these and other sectors, helping them deploy and productise new and smarter algorithms that can leverage vast and often dynamic troves of data.
But even as we see our clients reap significant improvements in areas such as fraud protection and credit applications, we also advise them against blindly spending on large data sets, tech stacks and other AI-related decisioning resources.
Of course, the potential benefits of AI are undeniable.
Improvements in the ability to spot subtle, atypical patterns using increasingly-powerful self-learning algorithms have significantly boosted fraud-protection capabilities in several industries.
Machine learning has also improved the speed and efficiency of credit-scoring and processing.
With its ability to mine large data sets for patterns and insights, AI also facilitates more personalised lending solutions as well as non-bank and online lending, particularly for lendees with “thin files” or limited credit histories. Machine-learning solutions are especially adept at making sense of unstructured data - for example, a business’ website reviews - to perform credit-scoring for thin files.
Companies are understandably eager, therefore, to harness smarter algorithms and decision analytics.
But they would do well to look beyond the hype surrounding this emerging field, and instead adopt a practical, focused approach to their investments - first by establishing a clear purpose for each data accumulation, and second by making sure they can measure the impact of each machine-learning solution.
Many businesses spend huge sums on tech stacks and on acquiring data and capabilities, but without any tangible return, in the hope that AI will somehow give them an edge in the market.
Yet returns from AI hinge on internal understanding and adoption of the new technology across all levels of a company - which typically requires that AI-development staff have access to the business teams, and that the AI solutions never lose sight of the end goal, which is to build or rebuild a product or service.
The first question to ask is how the new machine-learning solution will be used to obtain a specific result, and how this will impact the business.
This may sound obvious, but it is not uncommon to see, say, a major multinational services company investing in AI by creating a central machine-learning team, but without assigning it a specified commercial goal, targeted client or objective.
Over the years, such a team may publish multiple academic papers and develop some algorithmic capabilities, but ultimately not build or contribute to new or existing products.
By contrast, smaller independent teams doing R&D on AI in the same firm, and residing within individual business lines, are sometimes far more effective in building models that add commercial value.
One reason is that such teams typically have direct access to clients, live data and feedback channels to constantly guide them.
In any AI use case, it is also vital to clarify up-front what client segment is being served and how the company wishes to impact it, before looking for the right alternative data and methodologies.
If these are properly clarified, the investment can be smaller. For example, a company may find that fewer data sources can drive a viable solution.
As multiple data sources often imply additional costs and more complex stakeholder structures to manage, being economical on this front makes a huge difference.
Similarly, a clear purpose for each data accumulation should be established. Multiple data sources are often useful, but there is no point setting up data lakes for their own sake.
Attention should also be paid to finding a tech stack and platform advanced enough to support the AI solution end-to-end.
Finally, it is crucial to be able to measure the impact of the solution. A good approach is to test the process for a short time with and without the new solution, and on two separate but equivalent workflows.
Most of Experian’s use cases reference economic KPI such as ROI, but we also operate within corporate social responsibility frameworks that call for different targets.
What is critical is that the target is clear, captures the desired benefits and is aligned with expectations.
The key to getting more businesses to adopt AI and advanced decisioning lies in being able to accurately measure the impact and continually reassess the return.
This will provide the confidence that these new technologies are real, powerful and worth spending investing in.
*Luca Zuccoli is regional head of analytics and data lab for APAC at Experian, an information-services company
Read full article
Collaboration sparks innovation that help foster sustainable economic growth and drive positive outcomes for civil society. Just last week, I was part of Ingenuity19’s first ‘Tech for Good’ Summit in…Learn more